[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-paidf-anomalygen":3,"mdc-fz5irj-key":34,"related-org-nvidia-paidf-anomalygen":4750,"related-repo-nvidia-paidf-anomalygen":4909},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":32,"mdContent":33},"paidf-anomalygen","generate synthetic anomaly images with PAIDF","Full PAIDF AnomalyGen pipeline — fine-tune on a new anomaly dataset, generate synthetic anomaly images (SDG), evaluate quality (nn_score), and search per-sample (guidance, crop_ratio) parameters. Three modes: full (Phase 0→7: finetune then generate), finetune_only (Phase 0→1: train only), inference_only (Phase 0, 2→7: generate from an existing checkpoint). Use when the user asks to \"fine-tune AnomalyGen\", \"generate anomaly images\", \"run PAIDF SDG\", \"evaluate SDG output quality\", \"run per-sample search\", or run any part of the AnomalyGen pipeline, even if they only mention one phase.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"Machine Learning","machine-learning","tag",{"name":17,"slug":18,"type":15},"Images","images",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Anomaly Detection","anomaly-detection",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-08-05T05:58:12.462677","Apache-2.0",281,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fpaidf-anomalygen","---\nname: paidf-anomalygen\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit and a CUDA GPU. Pulls the `metropolis_sdg.paidf_anomalygen` image declared in `versions.yaml` at the skill bank root.\nmetadata:\n  author: NVIDIA Corporation\n  version: \"0.1.0\"\nallowed-tools: Read Bash\ndescription: >-\n  Full PAIDF AnomalyGen pipeline — fine-tune on a new anomaly dataset, generate\n  synthetic anomaly images (SDG), evaluate quality (nn_score), and search per-sample\n  (guidance, crop_ratio) parameters. Three modes: full (Phase 0→7: finetune then\n  generate), finetune_only (Phase 0→1: train only), inference_only (Phase 0, 2→7:\n  generate from an existing checkpoint). Use when the user asks to \"fine-tune\n  AnomalyGen\", \"generate anomaly images\", \"run PAIDF SDG\", \"evaluate SDG output\n  quality\", \"run per-sample search\", or run any part of the AnomalyGen pipeline,\n  even if they only mention one phase.\ntags:\n- tao\n- data\n---\n\n# PAIDF AnomalyGen\n\nMulti-phase pipeline (0–7); the `mode` flag selects which phases run.\n\n| Phase | What runs | Mode(s) |\n|---|---|---|\n| 0 | Verify \u002F download pretrained checkpoints | all |\n| 1 | Fine-tune on `dataset_dir` | `full`, `finetune_only` |\n| 2 | Prepare inference JSONL (AMP routing) | `full`, `inference_only` |\n| 3 | SDG — generate synthetic anomaly images → `original\u002F` | `full`, `inference_only` |\n| 4 | Eval `original\u002F` — emit `per_sample.csv` + `eval.log`, merge `nn_score` into `SDG_result.csv` | `full`, `inference_only` |\n| 5 | Per-sample `(guidance, crop_ratio)` search rounds → `rounds\u002Fround_NN\u002F` (each round runs SDG + eval) | `full`, `inference_only` |\n| 6 | Assemble best-of-rounds into `searched\u002F` (stitch only), plus `rounds\u002Fsearch_summary.csv` | `full`, `inference_only` |\n| 7 | Filter `searched\u002F` by `nn_threshold` (default `0.4`), regen dropped samples, then canonical bucket eval → `searched\u002F{per_sample.csv, eval.log}` | `full`, `inference_only` |\n\nRun every phase through to completion without mid-run pauses. Collect all\nrequired parameters up front, and run every command from the repo root.\n\n**Shell setup.** All `${ANOMALYGEN_SCRIPTS}` references resolve to the packaged\nhelper-script directory. Inside the container this is preset (`ENV\nANOMALYGEN_SCRIPTS=\u003Cdir>\u002Fscripts\u002Futilities`); on the host, export it once per\nshell:\n\n```bash\nexport ANOMALYGEN_SCRIPTS=\"$(git rev-parse --show-toplevel)\u002Fscripts\u002Futilities\"\n```\n\n`python3 -m scripts.utilities.\u003Cname>` invocations work from any CWD inside the\ncontainer (PYTHONPATH is preset) and from the repo root on the host. When inside\na product container (`ANOMALYGEN_PRODUCT_MODE=1`), invoke `anomalygen-guard`\nbefore any GPU work; if it reports `BLOCKED`, fix the listed issues before\ncontinuing.\n\n## Quick Start\n\nThe pipeline runs inside the `metropolis_sdg.paidf_anomalygen` container\n(declared in `versions.yaml`) or any host with the `cosmos-predict2` conda env\nactive. All phase commands assume that environment, at the repo root, with\n`ANOMALYGEN_SCRIPTS` exported.\n\nMinimal end-to-end run (`mode=full`):\n\n```bash\n# 1. Set the shared variables (see \"Shared variables\" for the full set).\nexport ANOMALYGEN_SCRIPTS=\"$(git rev-parse --show-toplevel)\u002Fscripts\u002Futilities\"\nMODE=full\nNAME=my_exp\nDATASET_DIR=\u002Fdata\u002Fuc1\nDEFECT_DESC=assets\u002Fdefect_spec_template.jsonl\nNUM_SDG=20\nMODEL_SIZE=2b\n\n# 2. Phase 0 — verify \u002F download checkpoints (~140 GB; needs HF_TOKEN).\n${ANOMALYGEN_SCRIPTS}\u002Fcheck.sh || ${ANOMALYGEN_SCRIPTS}\u002Fdownload_checkpoints.sh\n\n# 3. Walk Phases 1→7 in order (see each Phase section).\n```\n\nFor `mode=inference_only` (reuse a checkpoint) also set `CKPT`\u002F`STEP` and skip\nPhase 1. For `mode=finetune_only` run only Phases 0–1.\n\n## Running in Docker — container launch, mounts & permissions\n\nThe `paidf-anomalygen` image runs as a non-root baked-in user (`USER\nanomalygen`, `uid=10000`), independent of your host uid. Docker does not remap\nuids on bind mounts, so a host directory owned by your uid is not writable by\nuid 10000 and the container fails the instant it tries to create a file there.\nRun as your host uid with `--user \"$(id -u):$(id -g)\"` plus the mandatory\n`\u002Fetc\u002Fpasswd`+`\u002Fetc\u002Fgroup` and `HOME`\u002Fcache-redirect companions, and run the\nfail-fast write preflight before Phase 0. See `references\u002Fdocker.md` for the\nfull `docker run` command, the load-bearing-flag table, the preflight snippet,\nand the uid-10000 `chown`\u002F`chmod` fallback.\n\n## Reference files — read before executing phases\n\nRead **`references\u002Ffinetune.md`** before Phase 0\u002F1 and **`references\u002Finference.md`**\nbefore any of Phases 2–7; for `mode=full` read both before starting. The\nremaining references below are on-demand — read when troubleshooting or needing\nfull detail for a specific phase.\n\n| File | Read when |\n|---|---|\n| `references\u002Ffinetune.md` | Before Phase 0\u002F1: env check, checkpoint download, dataset validation, config generation, training commands, best-checkpoint selection |\n| `references\u002Ffinetune-commands.md` | Exact Phase 1 Step 1–4 commands and `CKPT`\u002F`STEP` derivation |\n| `references\u002Finference-commands.md` | Exact Phase 5 `run_round.sh` and Phase 7 `filter_with_regen` commands |\n| `references\u002Finference.md` | Before Phases 2–7: AMP routing, JSONL validation, SDG flags, eval interpretation, search loop, filtering |\n| `references\u002Fsetup.md` | Checkpoint download fails; first-time setup; HF_TOKEN \u002F disk issues |\n| `references\u002Fdatasets.md` | User needs to prepare or obtain a UC1 \u002F UC2 \u002F UC3 dataset; `dataset_dir` doesn't exist yet |\n| `references\u002Fprep-testcase.md` | AMP fails; need full param table, helper script descriptions, allocation invariant |\n| `references\u002Fsdg-inference.md` | NCCL hang; checkpoint validation error; multi-GPU VRAM question; full step list |\n| `references\u002Feval.md` | Unexpected scores; FID column order confusion; eval output format reference |\n| `references\u002Fsdg-refine.md` | draws.json alignment; re-AMP heuristics; search output layout |\n| `references\u002Fguard-and-custom-counts.md` | Full guard preflight command; `--per-defect-counts` example |\n| `references\u002Fdocker.md` | Container launch command, mount-permission flags, write preflight, uid-10000 fallback |\n| `references\u002Foutput-layout.md` | Full `results\u002F\u003Cname>\u002F` directory tree with per-file annotations; post-run Verification checklist |\n| `references\u002Ferror-handling.md` | Pipeline-level failure modes: missing mask dirs, short\u002Fempty AMP, mid-round resume, off-boundary `step` |\n\n---\n\n## Required parameters\n\n`num_SDG` allocation depends on `prep_testcase.sh --mode`: `inference` (default,\nPhase 2) is uniform across defect types, override per-defect via\n`--per-defect-counts`; `validation` (Phase 1's validation JSONL) is proportional\nto training mask counts (largest-remainder rounding) and enforces ≥1 per defect.\nSee `references\u002Fprep-testcase.md` for the full mode table.\n\n| Parameter | Description |\n|---|---|\n| `mode` | `full` (Phase 0→7), `inference_only` (skip Phase 1), or `finetune_only` (Phase 0→1 only). |\n| `name` | Experiment label. |\n| `dataset_dir` | Training\u002Freference dataset root. Drives mask-count allocation, AMP submask templates, and holds `semantic_segmentation_labels.json` for `cad` defects. |\n| `defect_spec` | JSONL tagging each defect `spatial_dependency` as `free`\u002F`text`\u002F`cad`. `text` entries need `roi_prompt_defect_location`. Template: `assets\u002Fdefect_spec_template.jsonl`. |\n| `num_SDG` | Total output samples per bucket. *(Ignored when `mode=finetune_only`.)* |\n\n## Conditionally required\n\n| Parameter | Required when | Description |\n|---|---|---|\n| `checkpoint_dir` \u002F `step` | `mode=inference_only` | Pre-existing fine-tuned model. In `mode=full` these are auto-derived after Phase 1; passing them is an error. In `mode=finetune_only` silently ignored — Phase 1 always trains from scratch (no resume-from-checkpoint support). Both must be present together — supplying only one is an error. |\n\n## Optional parameters\n\n| Parameter | Default | Description |\n|---|---|---|\n| `clean_dir` | `dataset_dir` | Clean images. Set only when they live outside the training dataset. Forwarded as `--clean-dir` to prep-testcase and `--clean-image-path` to finetune. |\n| `validation_jsonl` | auto-generated | Pre-built validation JSONL for Phase 1. When supplied, preflight verifies every `defect_spec` type appears and paths exist. |\n| `num_search_run` | `3` | Per-sample search budget for Phase 5. `0` skips search (only `original\u002F`). *(Ignored when `mode=finetune_only`.)* |\n| `nn_threshold` | `0.4` | `nn_score` cutoff for Phase 7 (DINOv2 correspondence to real defects — key KPI). Samples below are regenerated; final `searched\u002F` always has `num_SDG`. `0` disables filtering. |\n| `max_iter` | `75000` | Phase 1 only. Total fine-tune iterations. |\n| `save_iter` | `5000` | Phase 1 only. Checkpoint save interval. |\n| `validation_iter` | `5000` | Phase 1 only. Validation (`nn_score`) logging interval. |\n| `num_gpus` | `1` | Forwarded to Phase 1 (finetune) and Phase 3 (SDG). Eval and search rounds stay single-GPU. |\n| `model_size` | `2b` | `2b` or `14b`. Used by finetune and SDG. On-disk checkpoint path encodes in upper-case (`2b`→`2B`, `14b`→`14B`). |\n| `lr` | `0.02` | Phase 1 only. Learning rate. |\n| `batch_size` | `2` | Phase 1 only. Per-GPU batch size. |\n| `image_size` | `512` | Phase 1 only. Training resolution (square). |\n| `guidance_range` | `1.5 10.0` | Phase 5 search draw range for guidance. |\n| `crop_ratio_range` | `1.5 10.0` | Phase 5 search draw range for crop_ratio. |\n\n---\n\n## Mode validation (fail fast before any phase)\n\n- `mode` unset → halt: *\"`mode` is required (`full` | `inference_only` | `finetune_only`).\"*\n- `mode=inference_only` missing either `checkpoint_dir` or `step` → halt: *\"inference_only requires both `checkpoint_dir` and `step`.\"*\n- `mode=full` with `checkpoint_dir` or `step` supplied → halt: *\"full mode runs finetune; use `mode=inference_only` to reuse an existing checkpoint.\"*\n\n## Shared variables\n\nSet once before Phase 0:\n\n```bash\nMODE=\u003Cfull|inference_only|finetune_only>\nNAME=\u003Cexp>\nDATASET_DIR=\u003Cdataset_dir>\nCLEAN_DIR=${clean_dir:-${DATASET_DIR}}\nCKPT=\u003Ccheckpoint_dir>      # required iff MODE=inference_only; auto-derived after Phase 1 when MODE=full\nSTEP=\u003Citer>                # required iff MODE=inference_only; auto-derived after Phase 1 when MODE=full\nNUM_SDG=\u003CN>\nDEFECT_DESC=\u003Cdefect_spec.jsonl>\nDEFECTS=(T+A T+B)          # TEXTURE+TYPE names. For mode=inference_only, derive from ${CKPT}\u002Fag_config.yaml → dataloader_train.dataset.anomaly_types (also printed by validate_checkpoint.py in Phase 0). For mode=full, take from DEFECT_DESC entries. See references\u002Finference.md §Phase 0.\nNUM_SEARCH_RUN=${num_search_run:-3}\nNN_THRESHOLD=${nn_threshold:-0.4}\nMODEL_SIZE=\u003C2b|14b>\nNUM_GPUS=${num_gpus:-1}\nMAX_ITER=${max_iter:-75000}\nSAVE_ITER=${save_iter:-5000}\nVALIDATION_ITER=${validation_iter:-5000}\nLR=${lr:-0.02}\nBATCH_SIZE=${batch_size:-2}\nIMAGE_SIZE=${image_size:-512}\nVALIDATION_JSONL=${validation_jsonl:-}  # optional; set by Phase 1 Step 2 if not user-supplied\n\nBASE=results\u002F${NAME}\nJSONL=ag_inference\u002F${NAME}\u002Ftestcase.jsonl\nORIGINAL=${BASE}\u002Foriginal\nSEARCHED=${BASE}\u002Fsearched\nROUNDS=${BASE}\u002Frounds\nREGENS=${BASE}\u002Fregens\n```\n\n## Guard preflight (product mode only)\n\nWhen `ANOMALYGEN_PRODUCT_MODE=1`, run\n`.agents\u002Fskills\u002Fanomalygen-guard\u002Fscripts\u002Fpreflight.py` before any GPU work and\nfix any `BLOCKED` issues. `--validation-jsonl` is forwarded only when the user\nsupplied one; for `MODE=finetune_only` omit `--num-sdg` if not supplied. See\n`references\u002Fguard-and-custom-counts.md` for the full preflight command with all\nforwarded flags and the validation-JSONL \u002F `allocate_samples.py` 0-entry\nchecks.\n\n---\n\n## Phase 0 — checkpoints\n\nRead `references\u002Ffinetune.md §Phase 0` for HF_TOKEN requirements and what gets\ndownloaded (~140 GB). Verify first; download only what is missing.\n\n```bash\n${ANOMALYGEN_SCRIPTS}\u002Fcheck.sh \\\n    || ${ANOMALYGEN_SCRIPTS}\u002Fdownload_checkpoints.sh\n```\n\n---\n\n## Phase 1 — fine-tune (skip when `MODE=inference_only`)\n\nRead `references\u002Ffinetune.md §Phase 1` for dataset structure, config template\ndetails, and best-checkpoint selection. Four steps: (1) validate dataset \u002F\nderive anomaly types, (2) generate the validation JSONL (skip if user supplied\n`VALIDATION_JSONL`), (3) generate the training config — **show it to the user\nand confirm before writing** — (4) launch training in the background. Then\nderive `CKPT` (path encodes upper-case `MODEL_SIZE`) and `STEP` (highest\n`nn_score` step from validation logs). If `MODE=finetune_only`, stop after\ntraining. See `references\u002Ffinetune-commands.md` for the exact Step 1–4 commands\nand the `CKPT`\u002F`STEP` derivation snippet.\n\n---\n\n## Phase 2 — prep-testcase (skip when `MODE=finetune_only`)\n\nRead `references\u002Finference.md §Phase 2` for AMP routing detail and n_seeds\nsizing. Do NOT pass `--seeds` — it is auto-computed and is not a recognized\nflag. `prep_testcase.sh` defaults to `--mode inference` (uniform allocation\nacross defect types, no KPI floor), which Phase 2 always uses.\n\n```bash\n${ANOMALYGEN_SCRIPTS}\u002Fprep_testcase.sh \\\n    --name ${NAME} --num-sdg ${NUM_SDG} \\\n    --dataset-dir ${DATASET_DIR} \\\n    --clean-dir ${CLEAN_DIR} \\\n    --defect-spec ${DEFECT_DESC} \\\n    --amp-output-dir ag_inference\u002F${NAME}\u002Famp \\\n    --output-jsonl ${JSONL}\n```\n\n**Custom per-defect counts:** when the user specifies counts per defect type,\ntranslate to `--num-sdg` plus a `--per-defect-counts` JSON dict (types absent\nfrom the dict get 0; sum should equal `--num-sdg`, else the script warns on\nstderr and uses the override sum). Confirm the allocation when intent is\nambiguous. See `references\u002Fguard-and-custom-counts.md` for the full\n`--per-defect-counts` command example and the ambiguity-handling detail.\n\n---\n\n## Phase 3 — SDG → `original\u002F`\n\nRead `references\u002Finference.md §Phase 3` for JSONL validation against the\ncheckpoint, multi-GPU caveats, and output verification.\n\n```bash\npython3 -m scripts.utilities.validate_checkpoint ${CKPT} --step ${STEP}\npython3 -m scripts.utilities.validate_jsonl ${CKPT} ${JSONL}\n\n${ANOMALYGEN_SCRIPTS}\u002Frun_sdg.sh \\\n    --checkpoint_dir ${CKPT} --step ${STEP} \\\n    --input_jsonl ${JSONL} --output_dir ${ORIGINAL} \\\n    --model_size ${MODEL_SIZE} --num_gpus ${NUM_GPUS}\n\n${ANOMALYGEN_SCRIPTS}\u002Fverify_output.sh ${JSONL} ${ORIGINAL}\n```\n\n---\n\n## Phase 4 — eval `original\u002F`\n\nRead `references\u002Finference.md §Eval` for score interpretation and feature-count\nexplanation. `run_eval.sh` writes `per_sample.csv` and `eval.log` inside\n`original\u002F` and merges `nn_score` into `SDG_result.csv`.\n\n```bash\n${ANOMALYGEN_SCRIPTS}\u002Frun_eval.sh \\\n    --real-path ${DATASET_DIR} --generated-path ${ORIGINAL} \\\n    --anomaly-types ${DEFECTS[@]}\n```\n\n---\n\n## Phase 5 — per-sample search rounds\n\nRead `references\u002Finference.md §Phase 5` for draw strategy, ranges, and re-AMP\nguidance. For `r` in `1..NUM_SEARCH_RUN`:\n\n1. Read prior round's `per_sample.csv` (or `${ORIGINAL}\u002Fper_sample.csv` for `r=1`).\n2. Write `${ROUNDS}\u002Fround_${r}\u002Fdraws.json` with selected `(guidance, crop_ratio)` per sample.\n3. Run round via `${ANOMALYGEN_SCRIPTS}\u002Frun_round.sh` (SDG + eval; the round dir\n   gets its own `sdg\u002F{SDG_result.csv, per_sample.csv, eval.log}`). See\n   `references\u002Finference-commands.md §Phase 5` for the full command and flags.\n\n`NUM_SEARCH_RUN=0` is valid — skip this phase entirely and let Phase 6\nclone `original\u002F` into `searched\u002F`.\n\n---\n\n## Phase 6 — assemble `searched\u002F` (stitch only)\n\nAlways run assemble (works with 0 rounds — `searched\u002F` clones `original\u002F`, so\ndownstream always reads `searched\u002F` regardless of `num_search_run`). Stitch-only:\ncopies winning images per sample-index into `searched\u002F` and carries over\nper-sample `nn_score` \u002F `mnn_score` from each pick's source-round per_sample.csv.\nNo eval — Phase 7 emits the canonical `searched\u002Feval.log`.\n\n```bash\nmkdir -p ${ROUNDS}\npython3 -m scripts.utilities.assemble_searched \\\n    --original-dir ${ORIGINAL} --original-csv ${ORIGINAL}\u002Fper_sample.csv \\\n    --rounds-dir ${ROUNDS} --searched-dir ${SEARCHED}\n```\n\n---\n\n## Phase 7 — filter + regen + eval (default `nn_threshold=0.4`)\n\nPhase 7 **runs by default** (`nn_threshold=0.4`) on every `mode=full` and\n`mode=inference_only` invocation; pass `nn_threshold=0` to skip it. It filters\n`searched\u002F` by `nn_threshold`, regenerates dropped samples via re-AMP (fresh\n`(clean, submask)` pairing in the same defect type) for up to 5 attempts, then\nfalls back to best-scoring non-passing regens and finally to dropped originals,\nso the final bucket always equals `num_SDG`.\n\nRun `python3 -m scripts.utilities.filter_with_regen`. It runs the final\n`run_eval.sh` internally — the only eval against `searched\u002F`. Read\n`references\u002Finference.md §Phase 7` for regen mechanics, source-column tracing,\nand the `regens\u002Fregen_summary.csv` schema; see\n`references\u002Finference-commands.md §Phase 7` for the full command and flags.\n\n---\n\n## Output layout\n\nEvery bucket that gets eval'd carries the same triad of files:\n`SDG_result.csv` (generation params + `nn_score`), `per_sample.csv`\n(per-sample nn + mnn), and `eval.log` (aggregate FID \u002F per-defect avg).\nBuckets live under `results\u002F\u003Cname>\u002F` as `original\u002F` (Phase 3+4), `searched\u002F`\n(Phase 6 stitch + Phase 7 filter+regen+eval), `rounds\u002Fround_NN\u002F` (Phase 5,\nplus `search_summary.csv`), and `regens\u002Fregen_NN\u002F` (Phase 7, plus\n`regen_summary.csv`).\n\nSee `references\u002Foutput-layout.md` for the full directory tree with per-file\nannotations and the post-run **Verification** checklist (image counts per\nbucket, `search_summary.csv` \u002F `regen_summary.csv` row checks, and the per-type\n`nn_score` \u002F `mnn_score` \u002F `fid` fields in each `eval.log`).\n\n## Error handling\n\nCommon pipeline failure modes (missing mask dirs, short\u002Fempty AMP output and\nthe `0 entries written` halt, mid-round SDG failure resume, off-boundary\n`step`) are covered in `references\u002Ferror-handling.md`; see also\n`references\u002Ffinetune.md` and `references\u002Finference.md` for phase-specific\nerror handling.\n",{"data":35,"body":44},{"name":4,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"description":6,"tags":41},"Requires docker + nvidia-container-toolkit and a CUDA GPU. Pulls the `metropolis_sdg.paidf_anomalygen` image declared in `versions.yaml` at the skill bank root.",{"author":38,"version":39},"NVIDIA Corporation","0.1.0","Read Bash",[42,43],"tao","data",{"type":45,"children":46},"root",[47,55,70,422,427,454,521,556,563,600,613,859,896,902,992,998,1032,1347,1351,1357,1406,1609,1615,1686,1692,2189,2192,2198,2321,2327,2332,3060,3066,3132,3135,3141,3153,3204,3207,3220,3301,3304,3316,3352,3547,3592,3595,3606,3618,3908,3911,3922,3974,4063,4066,4072,4100,4179,4202,4205,4218,4279,4414,4417,4430,4498,4548,4551,4557,4640,4698,4704,4744],{"type":48,"tag":49,"props":50,"children":51},"element","h1",{"id":4},[52],{"type":53,"value":54},"text","PAIDF AnomalyGen",{"type":48,"tag":56,"props":57,"children":58},"p",{},[59,61,68],{"type":53,"value":60},"Multi-phase pipeline (0–7); the ",{"type":48,"tag":62,"props":63,"children":65},"code",{"className":64},[],[66],{"type":53,"value":67},"mode",{"type":53,"value":69}," flag selects which phases run.",{"type":48,"tag":71,"props":72,"children":73},"table",{},[74,98],{"type":48,"tag":75,"props":76,"children":77},"thead",{},[78],{"type":48,"tag":79,"props":80,"children":81},"tr",{},[82,88,93],{"type":48,"tag":83,"props":84,"children":85},"th",{},[86],{"type":53,"value":87},"Phase",{"type":48,"tag":83,"props":89,"children":90},{},[91],{"type":53,"value":92},"What runs",{"type":48,"tag":83,"props":94,"children":95},{},[96],{"type":53,"value":97},"Mode(s)",{"type":48,"tag":99,"props":100,"children":101},"tbody",{},[102,121,157,185,218,282,325,366],{"type":48,"tag":79,"props":103,"children":104},{},[105,111,116],{"type":48,"tag":106,"props":107,"children":108},"td",{},[109],{"type":53,"value":110},"0",{"type":48,"tag":106,"props":112,"children":113},{},[114],{"type":53,"value":115},"Verify \u002F download pretrained checkpoints",{"type":48,"tag":106,"props":117,"children":118},{},[119],{"type":53,"value":120},"all",{"type":48,"tag":79,"props":122,"children":123},{},[124,129,140],{"type":48,"tag":106,"props":125,"children":126},{},[127],{"type":53,"value":128},"1",{"type":48,"tag":106,"props":130,"children":131},{},[132,134],{"type":53,"value":133},"Fine-tune on ",{"type":48,"tag":62,"props":135,"children":137},{"className":136},[],[138],{"type":53,"value":139},"dataset_dir",{"type":48,"tag":106,"props":141,"children":142},{},[143,149,151],{"type":48,"tag":62,"props":144,"children":146},{"className":145},[],[147],{"type":53,"value":148},"full",{"type":53,"value":150},", ",{"type":48,"tag":62,"props":152,"children":154},{"className":153},[],[155],{"type":53,"value":156},"finetune_only",{"type":48,"tag":79,"props":158,"children":159},{},[160,165,170],{"type":48,"tag":106,"props":161,"children":162},{},[163],{"type":53,"value":164},"2",{"type":48,"tag":106,"props":166,"children":167},{},[168],{"type":53,"value":169},"Prepare inference JSONL (AMP routing)",{"type":48,"tag":106,"props":171,"children":172},{},[173,178,179],{"type":48,"tag":62,"props":174,"children":176},{"className":175},[],[177],{"type":53,"value":148},{"type":53,"value":150},{"type":48,"tag":62,"props":180,"children":182},{"className":181},[],[183],{"type":53,"value":184},"inference_only",{"type":48,"tag":79,"props":186,"children":187},{},[188,193,204],{"type":48,"tag":106,"props":189,"children":190},{},[191],{"type":53,"value":192},"3",{"type":48,"tag":106,"props":194,"children":195},{},[196,198],{"type":53,"value":197},"SDG — generate synthetic anomaly images → ",{"type":48,"tag":62,"props":199,"children":201},{"className":200},[],[202],{"type":53,"value":203},"original\u002F",{"type":48,"tag":106,"props":205,"children":206},{},[207,212,213],{"type":48,"tag":62,"props":208,"children":210},{"className":209},[],[211],{"type":53,"value":148},{"type":53,"value":150},{"type":48,"tag":62,"props":214,"children":216},{"className":215},[],[217],{"type":53,"value":184},{"type":48,"tag":79,"props":219,"children":220},{},[221,226,268],{"type":48,"tag":106,"props":222,"children":223},{},[224],{"type":53,"value":225},"4",{"type":48,"tag":106,"props":227,"children":228},{},[229,231,236,238,244,246,252,254,260,262],{"type":53,"value":230},"Eval ",{"type":48,"tag":62,"props":232,"children":234},{"className":233},[],[235],{"type":53,"value":203},{"type":53,"value":237}," — emit ",{"type":48,"tag":62,"props":239,"children":241},{"className":240},[],[242],{"type":53,"value":243},"per_sample.csv",{"type":53,"value":245}," + ",{"type":48,"tag":62,"props":247,"children":249},{"className":248},[],[250],{"type":53,"value":251},"eval.log",{"type":53,"value":253},", merge ",{"type":48,"tag":62,"props":255,"children":257},{"className":256},[],[258],{"type":53,"value":259},"nn_score",{"type":53,"value":261}," into ",{"type":48,"tag":62,"props":263,"children":265},{"className":264},[],[266],{"type":53,"value":267},"SDG_result.csv",{"type":48,"tag":106,"props":269,"children":270},{},[271,276,277],{"type":48,"tag":62,"props":272,"children":274},{"className":273},[],[275],{"type":53,"value":148},{"type":53,"value":150},{"type":48,"tag":62,"props":278,"children":280},{"className":279},[],[281],{"type":53,"value":184},{"type":48,"tag":79,"props":283,"children":284},{},[285,290,311],{"type":48,"tag":106,"props":286,"children":287},{},[288],{"type":53,"value":289},"5",{"type":48,"tag":106,"props":291,"children":292},{},[293,295,301,303,309],{"type":53,"value":294},"Per-sample ",{"type":48,"tag":62,"props":296,"children":298},{"className":297},[],[299],{"type":53,"value":300},"(guidance, crop_ratio)",{"type":53,"value":302}," search rounds → ",{"type":48,"tag":62,"props":304,"children":306},{"className":305},[],[307],{"type":53,"value":308},"rounds\u002Fround_NN\u002F",{"type":53,"value":310}," (each round runs SDG + eval)",{"type":48,"tag":106,"props":312,"children":313},{},[314,319,320],{"type":48,"tag":62,"props":315,"children":317},{"className":316},[],[318],{"type":53,"value":148},{"type":53,"value":150},{"type":48,"tag":62,"props":321,"children":323},{"className":322},[],[324],{"type":53,"value":184},{"type":48,"tag":79,"props":326,"children":327},{},[328,333,352],{"type":48,"tag":106,"props":329,"children":330},{},[331],{"type":53,"value":332},"6",{"type":48,"tag":106,"props":334,"children":335},{},[336,338,344,346],{"type":53,"value":337},"Assemble best-of-rounds into ",{"type":48,"tag":62,"props":339,"children":341},{"className":340},[],[342],{"type":53,"value":343},"searched\u002F",{"type":53,"value":345}," (stitch only), plus ",{"type":48,"tag":62,"props":347,"children":349},{"className":348},[],[350],{"type":53,"value":351},"rounds\u002Fsearch_summary.csv",{"type":48,"tag":106,"props":353,"children":354},{},[355,360,361],{"type":48,"tag":62,"props":356,"children":358},{"className":357},[],[359],{"type":53,"value":148},{"type":53,"value":150},{"type":48,"tag":62,"props":362,"children":364},{"className":363},[],[365],{"type":53,"value":184},{"type":48,"tag":79,"props":367,"children":368},{},[369,374,408],{"type":48,"tag":106,"props":370,"children":371},{},[372],{"type":53,"value":373},"7",{"type":48,"tag":106,"props":375,"children":376},{},[377,379,384,386,392,394,400,402],{"type":53,"value":378},"Filter ",{"type":48,"tag":62,"props":380,"children":382},{"className":381},[],[383],{"type":53,"value":343},{"type":53,"value":385}," by ",{"type":48,"tag":62,"props":387,"children":389},{"className":388},[],[390],{"type":53,"value":391},"nn_threshold",{"type":53,"value":393}," (default ",{"type":48,"tag":62,"props":395,"children":397},{"className":396},[],[398],{"type":53,"value":399},"0.4",{"type":53,"value":401},"), regen dropped samples, then canonical bucket eval → ",{"type":48,"tag":62,"props":403,"children":405},{"className":404},[],[406],{"type":53,"value":407},"searched\u002F{per_sample.csv, eval.log}",{"type":48,"tag":106,"props":409,"children":410},{},[411,416,417],{"type":48,"tag":62,"props":412,"children":414},{"className":413},[],[415],{"type":53,"value":148},{"type":53,"value":150},{"type":48,"tag":62,"props":418,"children":420},{"className":419},[],[421],{"type":53,"value":184},{"type":48,"tag":56,"props":423,"children":424},{},[425],{"type":53,"value":426},"Run every phase through to completion without mid-run pauses. Collect all\nrequired parameters up front, and run every command from the repo root.",{"type":48,"tag":56,"props":428,"children":429},{},[430,436,438,444,446,452],{"type":48,"tag":431,"props":432,"children":433},"strong",{},[434],{"type":53,"value":435},"Shell setup.",{"type":53,"value":437}," All ",{"type":48,"tag":62,"props":439,"children":441},{"className":440},[],[442],{"type":53,"value":443},"${ANOMALYGEN_SCRIPTS}",{"type":53,"value":445}," references resolve to the packaged\nhelper-script directory. Inside the container this is preset (",{"type":48,"tag":62,"props":447,"children":449},{"className":448},[],[450],{"type":53,"value":451},"ENV ANOMALYGEN_SCRIPTS=\u003Cdir>\u002Fscripts\u002Futilities",{"type":53,"value":453},"); on the host, export it once per\nshell:",{"type":48,"tag":455,"props":456,"children":461},"pre",{"className":457,"code":458,"language":459,"meta":460,"style":460},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","export ANOMALYGEN_SCRIPTS=\"$(git rev-parse --show-toplevel)\u002Fscripts\u002Futilities\"\n","bash","",[462],{"type":48,"tag":62,"props":463,"children":464},{"__ignoreMap":460},[465],{"type":48,"tag":466,"props":467,"children":470},"span",{"class":468,"line":469},"line",1,[471,477,483,489,494,500,506,511,516],{"type":48,"tag":466,"props":472,"children":474},{"style":473},"--shiki-light:#9C3EDA;--shiki-default:#C792EA;--shiki-dark:#C792EA",[475],{"type":53,"value":476},"export",{"type":48,"tag":466,"props":478,"children":480},{"style":479},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[481],{"type":53,"value":482}," ANOMALYGEN_SCRIPTS",{"type":48,"tag":466,"props":484,"children":486},{"style":485},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[487],{"type":53,"value":488},"=",{"type":48,"tag":466,"props":490,"children":491},{"style":485},[492],{"type":53,"value":493},"\"$(",{"type":48,"tag":466,"props":495,"children":497},{"style":496},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[498],{"type":53,"value":499},"git",{"type":48,"tag":466,"props":501,"children":503},{"style":502},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[504],{"type":53,"value":505}," rev-parse --show-toplevel",{"type":48,"tag":466,"props":507,"children":508},{"style":485},[509],{"type":53,"value":510},")",{"type":48,"tag":466,"props":512,"children":513},{"style":502},[514],{"type":53,"value":515},"\u002Fscripts\u002Futilities",{"type":48,"tag":466,"props":517,"children":518},{"style":485},[519],{"type":53,"value":520},"\"\n",{"type":48,"tag":56,"props":522,"children":523},{},[524,530,532,538,540,546,548,554],{"type":48,"tag":62,"props":525,"children":527},{"className":526},[],[528],{"type":53,"value":529},"python3 -m scripts.utilities.\u003Cname>",{"type":53,"value":531}," invocations work from any CWD inside the\ncontainer (PYTHONPATH is preset) and from the repo root on the host. When inside\na product container (",{"type":48,"tag":62,"props":533,"children":535},{"className":534},[],[536],{"type":53,"value":537},"ANOMALYGEN_PRODUCT_MODE=1",{"type":53,"value":539},"), invoke ",{"type":48,"tag":62,"props":541,"children":543},{"className":542},[],[544],{"type":53,"value":545},"anomalygen-guard",{"type":53,"value":547},"\nbefore any GPU work; if it reports ",{"type":48,"tag":62,"props":549,"children":551},{"className":550},[],[552],{"type":53,"value":553},"BLOCKED",{"type":53,"value":555},", fix the listed issues before\ncontinuing.",{"type":48,"tag":557,"props":558,"children":560},"h2",{"id":559},"quick-start",[561],{"type":53,"value":562},"Quick Start",{"type":48,"tag":56,"props":564,"children":565},{},[566,568,574,576,582,584,590,592,598],{"type":53,"value":567},"The pipeline runs inside the ",{"type":48,"tag":62,"props":569,"children":571},{"className":570},[],[572],{"type":53,"value":573},"metropolis_sdg.paidf_anomalygen",{"type":53,"value":575}," container\n(declared in ",{"type":48,"tag":62,"props":577,"children":579},{"className":578},[],[580],{"type":53,"value":581},"versions.yaml",{"type":53,"value":583},") or any host with the ",{"type":48,"tag":62,"props":585,"children":587},{"className":586},[],[588],{"type":53,"value":589},"cosmos-predict2",{"type":53,"value":591}," conda env\nactive. All phase commands assume that environment, at the repo root, with\n",{"type":48,"tag":62,"props":593,"children":595},{"className":594},[],[596],{"type":53,"value":597},"ANOMALYGEN_SCRIPTS",{"type":53,"value":599}," exported.",{"type":48,"tag":56,"props":601,"children":602},{},[603,605,611],{"type":53,"value":604},"Minimal end-to-end run (",{"type":48,"tag":62,"props":606,"children":608},{"className":607},[],[609],{"type":53,"value":610},"mode=full",{"type":53,"value":612},"):",{"type":48,"tag":455,"props":614,"children":616},{"className":457,"code":615,"language":459,"meta":460,"style":460},"# 1. Set the shared variables (see \"Shared variables\" for the full set).\nexport ANOMALYGEN_SCRIPTS=\"$(git rev-parse --show-toplevel)\u002Fscripts\u002Futilities\"\nMODE=full\nNAME=my_exp\nDATASET_DIR=\u002Fdata\u002Fuc1\nDEFECT_DESC=assets\u002Fdefect_spec_template.jsonl\nNUM_SDG=20\nMODEL_SIZE=2b\n\n# 2. Phase 0 — verify \u002F download checkpoints (~140 GB; needs HF_TOKEN).\n${ANOMALYGEN_SCRIPTS}\u002Fcheck.sh || ${ANOMALYGEN_SCRIPTS}\u002Fdownload_checkpoints.sh\n\n# 3. Walk Phases 1→7 in order (see each Phase section).\n",[617],{"type":48,"tag":62,"props":618,"children":619},{"__ignoreMap":460},[620,629,669,687,705,723,741,759,777,787,796,842,850],{"type":48,"tag":466,"props":621,"children":622},{"class":468,"line":469},[623],{"type":48,"tag":466,"props":624,"children":626},{"style":625},"--shiki-light:#90A4AE;--shiki-light-font-style:italic;--shiki-default:#546E7A;--shiki-default-font-style:italic;--shiki-dark:#676E95;--shiki-dark-font-style:italic",[627],{"type":53,"value":628},"# 1. Set the shared variables (see \"Shared variables\" for the full set).\n",{"type":48,"tag":466,"props":630,"children":632},{"class":468,"line":631},2,[633,637,641,645,649,653,657,661,665],{"type":48,"tag":466,"props":634,"children":635},{"style":473},[636],{"type":53,"value":476},{"type":48,"tag":466,"props":638,"children":639},{"style":479},[640],{"type":53,"value":482},{"type":48,"tag":466,"props":642,"children":643},{"style":485},[644],{"type":53,"value":488},{"type":48,"tag":466,"props":646,"children":647},{"style":485},[648],{"type":53,"value":493},{"type":48,"tag":466,"props":650,"children":651},{"style":496},[652],{"type":53,"value":499},{"type":48,"tag":466,"props":654,"children":655},{"style":502},[656],{"type":53,"value":505},{"type":48,"tag":466,"props":658,"children":659},{"style":485},[660],{"type":53,"value":510},{"type":48,"tag":466,"props":662,"children":663},{"style":502},[664],{"type":53,"value":515},{"type":48,"tag":466,"props":666,"children":667},{"style":485},[668],{"type":53,"value":520},{"type":48,"tag":466,"props":670,"children":672},{"class":468,"line":671},3,[673,678,682],{"type":48,"tag":466,"props":674,"children":675},{"style":479},[676],{"type":53,"value":677},"MODE",{"type":48,"tag":466,"props":679,"children":680},{"style":485},[681],{"type":53,"value":488},{"type":48,"tag":466,"props":683,"children":684},{"style":502},[685],{"type":53,"value":686},"full\n",{"type":48,"tag":466,"props":688,"children":690},{"class":468,"line":689},4,[691,696,700],{"type":48,"tag":466,"props":692,"children":693},{"style":479},[694],{"type":53,"value":695},"NAME",{"type":48,"tag":466,"props":697,"children":698},{"style":485},[699],{"type":53,"value":488},{"type":48,"tag":466,"props":701,"children":702},{"style":502},[703],{"type":53,"value":704},"my_exp\n",{"type":48,"tag":466,"props":706,"children":708},{"class":468,"line":707},5,[709,714,718],{"type":48,"tag":466,"props":710,"children":711},{"style":479},[712],{"type":53,"value":713},"DATASET_DIR",{"type":48,"tag":466,"props":715,"children":716},{"style":485},[717],{"type":53,"value":488},{"type":48,"tag":466,"props":719,"children":720},{"style":502},[721],{"type":53,"value":722},"\u002Fdata\u002Fuc1\n",{"type":48,"tag":466,"props":724,"children":726},{"class":468,"line":725},6,[727,732,736],{"type":48,"tag":466,"props":728,"children":729},{"style":479},[730],{"type":53,"value":731},"DEFECT_DESC",{"type":48,"tag":466,"props":733,"children":734},{"style":485},[735],{"type":53,"value":488},{"type":48,"tag":466,"props":737,"children":738},{"style":502},[739],{"type":53,"value":740},"assets\u002Fdefect_spec_template.jsonl\n",{"type":48,"tag":466,"props":742,"children":744},{"class":468,"line":743},7,[745,750,754],{"type":48,"tag":466,"props":746,"children":747},{"style":479},[748],{"type":53,"value":749},"NUM_SDG",{"type":48,"tag":466,"props":751,"children":752},{"style":485},[753],{"type":53,"value":488},{"type":48,"tag":466,"props":755,"children":756},{"style":502},[757],{"type":53,"value":758},"20\n",{"type":48,"tag":466,"props":760,"children":762},{"class":468,"line":761},8,[763,768,772],{"type":48,"tag":466,"props":764,"children":765},{"style":479},[766],{"type":53,"value":767},"MODEL_SIZE",{"type":48,"tag":466,"props":769,"children":770},{"style":485},[771],{"type":53,"value":488},{"type":48,"tag":466,"props":773,"children":774},{"style":502},[775],{"type":53,"value":776},"2b\n",{"type":48,"tag":466,"props":778,"children":780},{"class":468,"line":779},9,[781],{"type":48,"tag":466,"props":782,"children":784},{"emptyLinePlaceholder":783},true,[785],{"type":53,"value":786},"\n",{"type":48,"tag":466,"props":788,"children":790},{"class":468,"line":789},10,[791],{"type":48,"tag":466,"props":792,"children":793},{"style":625},[794],{"type":53,"value":795},"# 2. Phase 0 — verify \u002F download checkpoints (~140 GB; needs HF_TOKEN).\n",{"type":48,"tag":466,"props":797,"children":799},{"class":468,"line":798},11,[800,805,809,814,819,824,829,833,837],{"type":48,"tag":466,"props":801,"children":802},{"style":485},[803],{"type":53,"value":804},"${",{"type":48,"tag":466,"props":806,"children":807},{"style":479},[808],{"type":53,"value":597},{"type":48,"tag":466,"props":810,"children":811},{"style":485},[812],{"type":53,"value":813},"}",{"type":48,"tag":466,"props":815,"children":816},{"style":479},[817],{"type":53,"value":818},"\u002Fcheck.sh ",{"type":48,"tag":466,"props":820,"children":821},{"style":485},[822],{"type":53,"value":823},"||",{"type":48,"tag":466,"props":825,"children":826},{"style":485},[827],{"type":53,"value":828}," ${",{"type":48,"tag":466,"props":830,"children":831},{"style":479},[832],{"type":53,"value":597},{"type":48,"tag":466,"props":834,"children":835},{"style":485},[836],{"type":53,"value":813},{"type":48,"tag":466,"props":838,"children":839},{"style":496},[840],{"type":53,"value":841},"\u002Fdownload_checkpoints.sh\n",{"type":48,"tag":466,"props":843,"children":845},{"class":468,"line":844},12,[846],{"type":48,"tag":466,"props":847,"children":848},{"emptyLinePlaceholder":783},[849],{"type":53,"value":786},{"type":48,"tag":466,"props":851,"children":853},{"class":468,"line":852},13,[854],{"type":48,"tag":466,"props":855,"children":856},{"style":625},[857],{"type":53,"value":858},"# 3. Walk Phases 1→7 in order (see each Phase section).\n",{"type":48,"tag":56,"props":860,"children":861},{},[862,864,870,872,878,880,886,888,894],{"type":53,"value":863},"For ",{"type":48,"tag":62,"props":865,"children":867},{"className":866},[],[868],{"type":53,"value":869},"mode=inference_only",{"type":53,"value":871}," (reuse a checkpoint) also set ",{"type":48,"tag":62,"props":873,"children":875},{"className":874},[],[876],{"type":53,"value":877},"CKPT",{"type":53,"value":879},"\u002F",{"type":48,"tag":62,"props":881,"children":883},{"className":882},[],[884],{"type":53,"value":885},"STEP",{"type":53,"value":887}," and skip\nPhase 1. For ",{"type":48,"tag":62,"props":889,"children":891},{"className":890},[],[892],{"type":53,"value":893},"mode=finetune_only",{"type":53,"value":895}," run only Phases 0–1.",{"type":48,"tag":557,"props":897,"children":899},{"id":898},"running-in-docker-container-launch-mounts-permissions",[900],{"type":53,"value":901},"Running in Docker — container launch, mounts & permissions",{"type":48,"tag":56,"props":903,"children":904},{},[905,907,912,914,920,921,927,929,935,937,943,945,951,953,959,961,967,969,975,977,983,984,990],{"type":53,"value":906},"The ",{"type":48,"tag":62,"props":908,"children":910},{"className":909},[],[911],{"type":53,"value":4},{"type":53,"value":913}," image runs as a non-root baked-in user (",{"type":48,"tag":62,"props":915,"children":917},{"className":916},[],[918],{"type":53,"value":919},"USER anomalygen",{"type":53,"value":150},{"type":48,"tag":62,"props":922,"children":924},{"className":923},[],[925],{"type":53,"value":926},"uid=10000",{"type":53,"value":928},"), independent of your host uid. Docker does not remap\nuids on bind mounts, so a host directory owned by your uid is not writable by\nuid 10000 and the container fails the instant it tries to create a file there.\nRun as your host uid with ",{"type":48,"tag":62,"props":930,"children":932},{"className":931},[],[933],{"type":53,"value":934},"--user \"$(id -u):$(id -g)\"",{"type":53,"value":936}," plus the mandatory\n",{"type":48,"tag":62,"props":938,"children":940},{"className":939},[],[941],{"type":53,"value":942},"\u002Fetc\u002Fpasswd",{"type":53,"value":944},"+",{"type":48,"tag":62,"props":946,"children":948},{"className":947},[],[949],{"type":53,"value":950},"\u002Fetc\u002Fgroup",{"type":53,"value":952}," and ",{"type":48,"tag":62,"props":954,"children":956},{"className":955},[],[957],{"type":53,"value":958},"HOME",{"type":53,"value":960},"\u002Fcache-redirect companions, and run the\nfail-fast write preflight before Phase 0. See ",{"type":48,"tag":62,"props":962,"children":964},{"className":963},[],[965],{"type":53,"value":966},"references\u002Fdocker.md",{"type":53,"value":968}," for the\nfull ",{"type":48,"tag":62,"props":970,"children":972},{"className":971},[],[973],{"type":53,"value":974},"docker run",{"type":53,"value":976}," command, the load-bearing-flag table, the preflight snippet,\nand the uid-10000 ",{"type":48,"tag":62,"props":978,"children":980},{"className":979},[],[981],{"type":53,"value":982},"chown",{"type":53,"value":879},{"type":48,"tag":62,"props":985,"children":987},{"className":986},[],[988],{"type":53,"value":989},"chmod",{"type":53,"value":991}," fallback.",{"type":48,"tag":557,"props":993,"children":995},{"id":994},"reference-files-read-before-executing-phases",[996],{"type":53,"value":997},"Reference files — read before executing phases",{"type":48,"tag":56,"props":999,"children":1000},{},[1001,1003,1012,1014,1023,1025,1030],{"type":53,"value":1002},"Read ",{"type":48,"tag":431,"props":1004,"children":1005},{},[1006],{"type":48,"tag":62,"props":1007,"children":1009},{"className":1008},[],[1010],{"type":53,"value":1011},"references\u002Ffinetune.md",{"type":53,"value":1013}," before Phase 0\u002F1 and ",{"type":48,"tag":431,"props":1015,"children":1016},{},[1017],{"type":48,"tag":62,"props":1018,"children":1020},{"className":1019},[],[1021],{"type":53,"value":1022},"references\u002Finference.md",{"type":53,"value":1024},"\nbefore any of Phases 2–7; for ",{"type":48,"tag":62,"props":1026,"children":1028},{"className":1027},[],[1029],{"type":53,"value":610},{"type":53,"value":1031}," read both before starting. The\nremaining references below are on-demand — read when troubleshooting or needing\nfull detail for a specific phase.",{"type":48,"tag":71,"props":1033,"children":1034},{},[1035,1051],{"type":48,"tag":75,"props":1036,"children":1037},{},[1038],{"type":48,"tag":79,"props":1039,"children":1040},{},[1041,1046],{"type":48,"tag":83,"props":1042,"children":1043},{},[1044],{"type":53,"value":1045},"File",{"type":48,"tag":83,"props":1047,"children":1048},{},[1049],{"type":53,"value":1050},"Read when",{"type":48,"tag":99,"props":1052,"children":1053},{},[1054,1070,1100,1133,1149,1166,1190,1207,1224,1241,1258,1283,1299,1324],{"type":48,"tag":79,"props":1055,"children":1056},{},[1057,1065],{"type":48,"tag":106,"props":1058,"children":1059},{},[1060],{"type":48,"tag":62,"props":1061,"children":1063},{"className":1062},[],[1064],{"type":53,"value":1011},{"type":48,"tag":106,"props":1066,"children":1067},{},[1068],{"type":53,"value":1069},"Before Phase 0\u002F1: env check, checkpoint download, dataset validation, config generation, training commands, best-checkpoint selection",{"type":48,"tag":79,"props":1071,"children":1072},{},[1073,1082],{"type":48,"tag":106,"props":1074,"children":1075},{},[1076],{"type":48,"tag":62,"props":1077,"children":1079},{"className":1078},[],[1080],{"type":53,"value":1081},"references\u002Ffinetune-commands.md",{"type":48,"tag":106,"props":1083,"children":1084},{},[1085,1087,1092,1093,1098],{"type":53,"value":1086},"Exact Phase 1 Step 1–4 commands and ",{"type":48,"tag":62,"props":1088,"children":1090},{"className":1089},[],[1091],{"type":53,"value":877},{"type":53,"value":879},{"type":48,"tag":62,"props":1094,"children":1096},{"className":1095},[],[1097],{"type":53,"value":885},{"type":53,"value":1099}," derivation",{"type":48,"tag":79,"props":1101,"children":1102},{},[1103,1112],{"type":48,"tag":106,"props":1104,"children":1105},{},[1106],{"type":48,"tag":62,"props":1107,"children":1109},{"className":1108},[],[1110],{"type":53,"value":1111},"references\u002Finference-commands.md",{"type":48,"tag":106,"props":1113,"children":1114},{},[1115,1117,1123,1125,1131],{"type":53,"value":1116},"Exact Phase 5 ",{"type":48,"tag":62,"props":1118,"children":1120},{"className":1119},[],[1121],{"type":53,"value":1122},"run_round.sh",{"type":53,"value":1124}," and Phase 7 ",{"type":48,"tag":62,"props":1126,"children":1128},{"className":1127},[],[1129],{"type":53,"value":1130},"filter_with_regen",{"type":53,"value":1132}," commands",{"type":48,"tag":79,"props":1134,"children":1135},{},[1136,1144],{"type":48,"tag":106,"props":1137,"children":1138},{},[1139],{"type":48,"tag":62,"props":1140,"children":1142},{"className":1141},[],[1143],{"type":53,"value":1022},{"type":48,"tag":106,"props":1145,"children":1146},{},[1147],{"type":53,"value":1148},"Before Phases 2–7: AMP routing, JSONL validation, SDG flags, eval interpretation, search loop, filtering",{"type":48,"tag":79,"props":1150,"children":1151},{},[1152,1161],{"type":48,"tag":106,"props":1153,"children":1154},{},[1155],{"type":48,"tag":62,"props":1156,"children":1158},{"className":1157},[],[1159],{"type":53,"value":1160},"references\u002Fsetup.md",{"type":48,"tag":106,"props":1162,"children":1163},{},[1164],{"type":53,"value":1165},"Checkpoint download fails; first-time setup; HF_TOKEN \u002F disk issues",{"type":48,"tag":79,"props":1167,"children":1168},{},[1169,1178],{"type":48,"tag":106,"props":1170,"children":1171},{},[1172],{"type":48,"tag":62,"props":1173,"children":1175},{"className":1174},[],[1176],{"type":53,"value":1177},"references\u002Fdatasets.md",{"type":48,"tag":106,"props":1179,"children":1180},{},[1181,1183,1188],{"type":53,"value":1182},"User needs to prepare or obtain a UC1 \u002F UC2 \u002F UC3 dataset; ",{"type":48,"tag":62,"props":1184,"children":1186},{"className":1185},[],[1187],{"type":53,"value":139},{"type":53,"value":1189}," doesn't exist yet",{"type":48,"tag":79,"props":1191,"children":1192},{},[1193,1202],{"type":48,"tag":106,"props":1194,"children":1195},{},[1196],{"type":48,"tag":62,"props":1197,"children":1199},{"className":1198},[],[1200],{"type":53,"value":1201},"references\u002Fprep-testcase.md",{"type":48,"tag":106,"props":1203,"children":1204},{},[1205],{"type":53,"value":1206},"AMP fails; need full param table, helper script descriptions, allocation invariant",{"type":48,"tag":79,"props":1208,"children":1209},{},[1210,1219],{"type":48,"tag":106,"props":1211,"children":1212},{},[1213],{"type":48,"tag":62,"props":1214,"children":1216},{"className":1215},[],[1217],{"type":53,"value":1218},"references\u002Fsdg-inference.md",{"type":48,"tag":106,"props":1220,"children":1221},{},[1222],{"type":53,"value":1223},"NCCL hang; checkpoint validation error; multi-GPU VRAM question; full step list",{"type":48,"tag":79,"props":1225,"children":1226},{},[1227,1236],{"type":48,"tag":106,"props":1228,"children":1229},{},[1230],{"type":48,"tag":62,"props":1231,"children":1233},{"className":1232},[],[1234],{"type":53,"value":1235},"references\u002Feval.md",{"type":48,"tag":106,"props":1237,"children":1238},{},[1239],{"type":53,"value":1240},"Unexpected scores; FID column order confusion; eval output format reference",{"type":48,"tag":79,"props":1242,"children":1243},{},[1244,1253],{"type":48,"tag":106,"props":1245,"children":1246},{},[1247],{"type":48,"tag":62,"props":1248,"children":1250},{"className":1249},[],[1251],{"type":53,"value":1252},"references\u002Fsdg-refine.md",{"type":48,"tag":106,"props":1254,"children":1255},{},[1256],{"type":53,"value":1257},"draws.json alignment; re-AMP heuristics; search output layout",{"type":48,"tag":79,"props":1259,"children":1260},{},[1261,1270],{"type":48,"tag":106,"props":1262,"children":1263},{},[1264],{"type":48,"tag":62,"props":1265,"children":1267},{"className":1266},[],[1268],{"type":53,"value":1269},"references\u002Fguard-and-custom-counts.md",{"type":48,"tag":106,"props":1271,"children":1272},{},[1273,1275,1281],{"type":53,"value":1274},"Full guard preflight command; ",{"type":48,"tag":62,"props":1276,"children":1278},{"className":1277},[],[1279],{"type":53,"value":1280},"--per-defect-counts",{"type":53,"value":1282}," example",{"type":48,"tag":79,"props":1284,"children":1285},{},[1286,1294],{"type":48,"tag":106,"props":1287,"children":1288},{},[1289],{"type":48,"tag":62,"props":1290,"children":1292},{"className":1291},[],[1293],{"type":53,"value":966},{"type":48,"tag":106,"props":1295,"children":1296},{},[1297],{"type":53,"value":1298},"Container launch command, mount-permission flags, write preflight, uid-10000 fallback",{"type":48,"tag":79,"props":1300,"children":1301},{},[1302,1311],{"type":48,"tag":106,"props":1303,"children":1304},{},[1305],{"type":48,"tag":62,"props":1306,"children":1308},{"className":1307},[],[1309],{"type":53,"value":1310},"references\u002Foutput-layout.md",{"type":48,"tag":106,"props":1312,"children":1313},{},[1314,1316,1322],{"type":53,"value":1315},"Full ",{"type":48,"tag":62,"props":1317,"children":1319},{"className":1318},[],[1320],{"type":53,"value":1321},"results\u002F\u003Cname>\u002F",{"type":53,"value":1323}," directory tree with per-file annotations; post-run Verification checklist",{"type":48,"tag":79,"props":1325,"children":1326},{},[1327,1336],{"type":48,"tag":106,"props":1328,"children":1329},{},[1330],{"type":48,"tag":62,"props":1331,"children":1333},{"className":1332},[],[1334],{"type":53,"value":1335},"references\u002Ferror-handling.md",{"type":48,"tag":106,"props":1337,"children":1338},{},[1339,1341],{"type":53,"value":1340},"Pipeline-level failure modes: missing mask dirs, short\u002Fempty AMP, mid-round resume, off-boundary ",{"type":48,"tag":62,"props":1342,"children":1344},{"className":1343},[],[1345],{"type":53,"value":1346},"step",{"type":48,"tag":1348,"props":1349,"children":1350},"hr",{},[],{"type":48,"tag":557,"props":1352,"children":1354},{"id":1353},"required-parameters",[1355],{"type":53,"value":1356},"Required parameters",{"type":48,"tag":56,"props":1358,"children":1359},{},[1360,1366,1368,1374,1376,1382,1384,1389,1391,1397,1399,1404],{"type":48,"tag":62,"props":1361,"children":1363},{"className":1362},[],[1364],{"type":53,"value":1365},"num_SDG",{"type":53,"value":1367}," allocation depends on ",{"type":48,"tag":62,"props":1369,"children":1371},{"className":1370},[],[1372],{"type":53,"value":1373},"prep_testcase.sh --mode",{"type":53,"value":1375},": ",{"type":48,"tag":62,"props":1377,"children":1379},{"className":1378},[],[1380],{"type":53,"value":1381},"inference",{"type":53,"value":1383}," (default,\nPhase 2) is uniform across defect types, override per-defect via\n",{"type":48,"tag":62,"props":1385,"children":1387},{"className":1386},[],[1388],{"type":53,"value":1280},{"type":53,"value":1390},"; ",{"type":48,"tag":62,"props":1392,"children":1394},{"className":1393},[],[1395],{"type":53,"value":1396},"validation",{"type":53,"value":1398}," (Phase 1's validation JSONL) is proportional\nto training mask counts (largest-remainder rounding) and enforces ≥1 per defect.\nSee ",{"type":48,"tag":62,"props":1400,"children":1402},{"className":1401},[],[1403],{"type":53,"value":1201},{"type":53,"value":1405}," for the full mode table.",{"type":48,"tag":71,"props":1407,"children":1408},{},[1409,1425],{"type":48,"tag":75,"props":1410,"children":1411},{},[1412],{"type":48,"tag":79,"props":1413,"children":1414},{},[1415,1420],{"type":48,"tag":83,"props":1416,"children":1417},{},[1418],{"type":53,"value":1419},"Parameter",{"type":48,"tag":83,"props":1421,"children":1422},{},[1423],{"type":53,"value":1424},"Description",{"type":48,"tag":99,"props":1426,"children":1427},{},[1428,1463,1480,1512,1580],{"type":48,"tag":79,"props":1429,"children":1430},{},[1431,1439],{"type":48,"tag":106,"props":1432,"children":1433},{},[1434],{"type":48,"tag":62,"props":1435,"children":1437},{"className":1436},[],[1438],{"type":53,"value":67},{"type":48,"tag":106,"props":1440,"children":1441},{},[1442,1447,1449,1454,1456,1461],{"type":48,"tag":62,"props":1443,"children":1445},{"className":1444},[],[1446],{"type":53,"value":148},{"type":53,"value":1448}," (Phase 0→7), ",{"type":48,"tag":62,"props":1450,"children":1452},{"className":1451},[],[1453],{"type":53,"value":184},{"type":53,"value":1455}," (skip Phase 1), or ",{"type":48,"tag":62,"props":1457,"children":1459},{"className":1458},[],[1460],{"type":53,"value":156},{"type":53,"value":1462}," (Phase 0→1 only).",{"type":48,"tag":79,"props":1464,"children":1465},{},[1466,1475],{"type":48,"tag":106,"props":1467,"children":1468},{},[1469],{"type":48,"tag":62,"props":1470,"children":1472},{"className":1471},[],[1473],{"type":53,"value":1474},"name",{"type":48,"tag":106,"props":1476,"children":1477},{},[1478],{"type":53,"value":1479},"Experiment label.",{"type":48,"tag":79,"props":1481,"children":1482},{},[1483,1491],{"type":48,"tag":106,"props":1484,"children":1485},{},[1486],{"type":48,"tag":62,"props":1487,"children":1489},{"className":1488},[],[1490],{"type":53,"value":139},{"type":48,"tag":106,"props":1492,"children":1493},{},[1494,1496,1502,1504,1510],{"type":53,"value":1495},"Training\u002Freference dataset root. Drives mask-count allocation, AMP submask templates, and holds ",{"type":48,"tag":62,"props":1497,"children":1499},{"className":1498},[],[1500],{"type":53,"value":1501},"semantic_segmentation_labels.json",{"type":53,"value":1503}," for ",{"type":48,"tag":62,"props":1505,"children":1507},{"className":1506},[],[1508],{"type":53,"value":1509},"cad",{"type":53,"value":1511}," defects.",{"type":48,"tag":79,"props":1513,"children":1514},{},[1515,1524],{"type":48,"tag":106,"props":1516,"children":1517},{},[1518],{"type":48,"tag":62,"props":1519,"children":1521},{"className":1520},[],[1522],{"type":53,"value":1523},"defect_spec",{"type":48,"tag":106,"props":1525,"children":1526},{},[1527,1529,1535,1537,1543,1544,1549,1550,1555,1557,1562,1564,1570,1572,1578],{"type":53,"value":1528},"JSONL tagging each defect ",{"type":48,"tag":62,"props":1530,"children":1532},{"className":1531},[],[1533],{"type":53,"value":1534},"spatial_dependency",{"type":53,"value":1536}," as ",{"type":48,"tag":62,"props":1538,"children":1540},{"className":1539},[],[1541],{"type":53,"value":1542},"free",{"type":53,"value":879},{"type":48,"tag":62,"props":1545,"children":1547},{"className":1546},[],[1548],{"type":53,"value":53},{"type":53,"value":879},{"type":48,"tag":62,"props":1551,"children":1553},{"className":1552},[],[1554],{"type":53,"value":1509},{"type":53,"value":1556},". ",{"type":48,"tag":62,"props":1558,"children":1560},{"className":1559},[],[1561],{"type":53,"value":53},{"type":53,"value":1563}," entries need ",{"type":48,"tag":62,"props":1565,"children":1567},{"className":1566},[],[1568],{"type":53,"value":1569},"roi_prompt_defect_location",{"type":53,"value":1571},". Template: ",{"type":48,"tag":62,"props":1573,"children":1575},{"className":1574},[],[1576],{"type":53,"value":1577},"assets\u002Fdefect_spec_template.jsonl",{"type":53,"value":1579},".",{"type":48,"tag":79,"props":1581,"children":1582},{},[1583,1591],{"type":48,"tag":106,"props":1584,"children":1585},{},[1586],{"type":48,"tag":62,"props":1587,"children":1589},{"className":1588},[],[1590],{"type":53,"value":1365},{"type":48,"tag":106,"props":1592,"children":1593},{},[1594,1596],{"type":53,"value":1595},"Total output samples per bucket. ",{"type":48,"tag":1597,"props":1598,"children":1599},"em",{},[1600,1602,1607],{"type":53,"value":1601},"(Ignored when ",{"type":48,"tag":62,"props":1603,"children":1605},{"className":1604},[],[1606],{"type":53,"value":893},{"type":53,"value":1608},".)",{"type":48,"tag":557,"props":1610,"children":1612},{"id":1611},"conditionally-required",[1613],{"type":53,"value":1614},"Conditionally required",{"type":48,"tag":71,"props":1616,"children":1617},{},[1618,1637],{"type":48,"tag":75,"props":1619,"children":1620},{},[1621],{"type":48,"tag":79,"props":1622,"children":1623},{},[1624,1628,1633],{"type":48,"tag":83,"props":1625,"children":1626},{},[1627],{"type":53,"value":1419},{"type":48,"tag":83,"props":1629,"children":1630},{},[1631],{"type":53,"value":1632},"Required when",{"type":48,"tag":83,"props":1634,"children":1635},{},[1636],{"type":53,"value":1424},{"type":48,"tag":99,"props":1638,"children":1639},{},[1640],{"type":48,"tag":79,"props":1641,"children":1642},{},[1643,1659,1667],{"type":48,"tag":106,"props":1644,"children":1645},{},[1646,1652,1654],{"type":48,"tag":62,"props":1647,"children":1649},{"className":1648},[],[1650],{"type":53,"value":1651},"checkpoint_dir",{"type":53,"value":1653}," \u002F ",{"type":48,"tag":62,"props":1655,"children":1657},{"className":1656},[],[1658],{"type":53,"value":1346},{"type":48,"tag":106,"props":1660,"children":1661},{},[1662],{"type":48,"tag":62,"props":1663,"children":1665},{"className":1664},[],[1666],{"type":53,"value":869},{"type":48,"tag":106,"props":1668,"children":1669},{},[1670,1672,1677,1679,1684],{"type":53,"value":1671},"Pre-existing fine-tuned model. In ",{"type":48,"tag":62,"props":1673,"children":1675},{"className":1674},[],[1676],{"type":53,"value":610},{"type":53,"value":1678}," these are auto-derived after Phase 1; passing them is an error. In ",{"type":48,"tag":62,"props":1680,"children":1682},{"className":1681},[],[1683],{"type":53,"value":893},{"type":53,"value":1685}," silently ignored — Phase 1 always trains from scratch (no resume-from-checkpoint support). Both must be present together — supplying only one is an error.",{"type":48,"tag":557,"props":1687,"children":1689},{"id":1688},"optional-parameters",[1690],{"type":53,"value":1691},"Optional parameters",{"type":48,"tag":71,"props":1693,"children":1694},{},[1695,1714],{"type":48,"tag":75,"props":1696,"children":1697},{},[1698],{"type":48,"tag":79,"props":1699,"children":1700},{},[1701,1705,1710],{"type":48,"tag":83,"props":1702,"children":1703},{},[1704],{"type":53,"value":1419},{"type":48,"tag":83,"props":1706,"children":1707},{},[1708],{"type":53,"value":1709},"Default",{"type":48,"tag":83,"props":1711,"children":1712},{},[1713],{"type":53,"value":1424},{"type":48,"tag":99,"props":1715,"children":1716},{},[1717,1758,1787,1836,1885,1911,1937,1969,1994,2061,2087,2112,2138,2164],{"type":48,"tag":79,"props":1718,"children":1719},{},[1720,1729,1737],{"type":48,"tag":106,"props":1721,"children":1722},{},[1723],{"type":48,"tag":62,"props":1724,"children":1726},{"className":1725},[],[1727],{"type":53,"value":1728},"clean_dir",{"type":48,"tag":106,"props":1730,"children":1731},{},[1732],{"type":48,"tag":62,"props":1733,"children":1735},{"className":1734},[],[1736],{"type":53,"value":139},{"type":48,"tag":106,"props":1738,"children":1739},{},[1740,1742,1748,1750,1756],{"type":53,"value":1741},"Clean images. Set only when they live outside the training dataset. Forwarded as ",{"type":48,"tag":62,"props":1743,"children":1745},{"className":1744},[],[1746],{"type":53,"value":1747},"--clean-dir",{"type":53,"value":1749}," to prep-testcase and ",{"type":48,"tag":62,"props":1751,"children":1753},{"className":1752},[],[1754],{"type":53,"value":1755},"--clean-image-path",{"type":53,"value":1757}," to finetune.",{"type":48,"tag":79,"props":1759,"children":1760},{},[1761,1770,1775],{"type":48,"tag":106,"props":1762,"children":1763},{},[1764],{"type":48,"tag":62,"props":1765,"children":1767},{"className":1766},[],[1768],{"type":53,"value":1769},"validation_jsonl",{"type":48,"tag":106,"props":1771,"children":1772},{},[1773],{"type":53,"value":1774},"auto-generated",{"type":48,"tag":106,"props":1776,"children":1777},{},[1778,1780,1785],{"type":53,"value":1779},"Pre-built validation JSONL for Phase 1. When supplied, preflight verifies every ",{"type":48,"tag":62,"props":1781,"children":1783},{"className":1782},[],[1784],{"type":53,"value":1523},{"type":53,"value":1786}," type appears and paths exist.",{"type":48,"tag":79,"props":1788,"children":1789},{},[1790,1799,1807],{"type":48,"tag":106,"props":1791,"children":1792},{},[1793],{"type":48,"tag":62,"props":1794,"children":1796},{"className":1795},[],[1797],{"type":53,"value":1798},"num_search_run",{"type":48,"tag":106,"props":1800,"children":1801},{},[1802],{"type":48,"tag":62,"props":1803,"children":1805},{"className":1804},[],[1806],{"type":53,"value":192},{"type":48,"tag":106,"props":1808,"children":1809},{},[1810,1812,1817,1819,1824,1826],{"type":53,"value":1811},"Per-sample search budget for Phase 5. ",{"type":48,"tag":62,"props":1813,"children":1815},{"className":1814},[],[1816],{"type":53,"value":110},{"type":53,"value":1818}," skips search (only ",{"type":48,"tag":62,"props":1820,"children":1822},{"className":1821},[],[1823],{"type":53,"value":203},{"type":53,"value":1825},"). ",{"type":48,"tag":1597,"props":1827,"children":1828},{},[1829,1830,1835],{"type":53,"value":1601},{"type":48,"tag":62,"props":1831,"children":1833},{"className":1832},[],[1834],{"type":53,"value":893},{"type":53,"value":1608},{"type":48,"tag":79,"props":1837,"children":1838},{},[1839,1847,1855],{"type":48,"tag":106,"props":1840,"children":1841},{},[1842],{"type":48,"tag":62,"props":1843,"children":1845},{"className":1844},[],[1846],{"type":53,"value":391},{"type":48,"tag":106,"props":1848,"children":1849},{},[1850],{"type":48,"tag":62,"props":1851,"children":1853},{"className":1852},[],[1854],{"type":53,"value":399},{"type":48,"tag":106,"props":1856,"children":1857},{},[1858,1863,1865,1870,1872,1877,1878,1883],{"type":48,"tag":62,"props":1859,"children":1861},{"className":1860},[],[1862],{"type":53,"value":259},{"type":53,"value":1864}," cutoff for Phase 7 (DINOv2 correspondence to real defects — key KPI). Samples below are regenerated; final ",{"type":48,"tag":62,"props":1866,"children":1868},{"className":1867},[],[1869],{"type":53,"value":343},{"type":53,"value":1871}," always has ",{"type":48,"tag":62,"props":1873,"children":1875},{"className":1874},[],[1876],{"type":53,"value":1365},{"type":53,"value":1556},{"type":48,"tag":62,"props":1879,"children":1881},{"className":1880},[],[1882],{"type":53,"value":110},{"type":53,"value":1884}," disables filtering.",{"type":48,"tag":79,"props":1886,"children":1887},{},[1888,1897,1906],{"type":48,"tag":106,"props":1889,"children":1890},{},[1891],{"type":48,"tag":62,"props":1892,"children":1894},{"className":1893},[],[1895],{"type":53,"value":1896},"max_iter",{"type":48,"tag":106,"props":1898,"children":1899},{},[1900],{"type":48,"tag":62,"props":1901,"children":1903},{"className":1902},[],[1904],{"type":53,"value":1905},"75000",{"type":48,"tag":106,"props":1907,"children":1908},{},[1909],{"type":53,"value":1910},"Phase 1 only. Total fine-tune iterations.",{"type":48,"tag":79,"props":1912,"children":1913},{},[1914,1923,1932],{"type":48,"tag":106,"props":1915,"children":1916},{},[1917],{"type":48,"tag":62,"props":1918,"children":1920},{"className":1919},[],[1921],{"type":53,"value":1922},"save_iter",{"type":48,"tag":106,"props":1924,"children":1925},{},[1926],{"type":48,"tag":62,"props":1927,"children":1929},{"className":1928},[],[1930],{"type":53,"value":1931},"5000",{"type":48,"tag":106,"props":1933,"children":1934},{},[1935],{"type":53,"value":1936},"Phase 1 only. Checkpoint save interval.",{"type":48,"tag":79,"props":1938,"children":1939},{},[1940,1949,1957],{"type":48,"tag":106,"props":1941,"children":1942},{},[1943],{"type":48,"tag":62,"props":1944,"children":1946},{"className":1945},[],[1947],{"type":53,"value":1948},"validation_iter",{"type":48,"tag":106,"props":1950,"children":1951},{},[1952],{"type":48,"tag":62,"props":1953,"children":1955},{"className":1954},[],[1956],{"type":53,"value":1931},{"type":48,"tag":106,"props":1958,"children":1959},{},[1960,1962,1967],{"type":53,"value":1961},"Phase 1 only. Validation (",{"type":48,"tag":62,"props":1963,"children":1965},{"className":1964},[],[1966],{"type":53,"value":259},{"type":53,"value":1968},") logging interval.",{"type":48,"tag":79,"props":1970,"children":1971},{},[1972,1981,1989],{"type":48,"tag":106,"props":1973,"children":1974},{},[1975],{"type":48,"tag":62,"props":1976,"children":1978},{"className":1977},[],[1979],{"type":53,"value":1980},"num_gpus",{"type":48,"tag":106,"props":1982,"children":1983},{},[1984],{"type":48,"tag":62,"props":1985,"children":1987},{"className":1986},[],[1988],{"type":53,"value":128},{"type":48,"tag":106,"props":1990,"children":1991},{},[1992],{"type":53,"value":1993},"Forwarded to Phase 1 (finetune) and Phase 3 (SDG). Eval and search rounds stay single-GPU.",{"type":48,"tag":79,"props":1995,"children":1996},{},[1997,2006,2015],{"type":48,"tag":106,"props":1998,"children":1999},{},[2000],{"type":48,"tag":62,"props":2001,"children":2003},{"className":2002},[],[2004],{"type":53,"value":2005},"model_size",{"type":48,"tag":106,"props":2007,"children":2008},{},[2009],{"type":48,"tag":62,"props":2010,"children":2012},{"className":2011},[],[2013],{"type":53,"value":2014},"2b",{"type":48,"tag":106,"props":2016,"children":2017},{},[2018,2023,2025,2031,2033,2038,2040,2046,2047,2052,2053,2059],{"type":48,"tag":62,"props":2019,"children":2021},{"className":2020},[],[2022],{"type":53,"value":2014},{"type":53,"value":2024}," or ",{"type":48,"tag":62,"props":2026,"children":2028},{"className":2027},[],[2029],{"type":53,"value":2030},"14b",{"type":53,"value":2032},". Used by finetune and SDG. On-disk checkpoint path encodes in upper-case (",{"type":48,"tag":62,"props":2034,"children":2036},{"className":2035},[],[2037],{"type":53,"value":2014},{"type":53,"value":2039},"→",{"type":48,"tag":62,"props":2041,"children":2043},{"className":2042},[],[2044],{"type":53,"value":2045},"2B",{"type":53,"value":150},{"type":48,"tag":62,"props":2048,"children":2050},{"className":2049},[],[2051],{"type":53,"value":2030},{"type":53,"value":2039},{"type":48,"tag":62,"props":2054,"children":2056},{"className":2055},[],[2057],{"type":53,"value":2058},"14B",{"type":53,"value":2060},").",{"type":48,"tag":79,"props":2062,"children":2063},{},[2064,2073,2082],{"type":48,"tag":106,"props":2065,"children":2066},{},[2067],{"type":48,"tag":62,"props":2068,"children":2070},{"className":2069},[],[2071],{"type":53,"value":2072},"lr",{"type":48,"tag":106,"props":2074,"children":2075},{},[2076],{"type":48,"tag":62,"props":2077,"children":2079},{"className":2078},[],[2080],{"type":53,"value":2081},"0.02",{"type":48,"tag":106,"props":2083,"children":2084},{},[2085],{"type":53,"value":2086},"Phase 1 only. Learning rate.",{"type":48,"tag":79,"props":2088,"children":2089},{},[2090,2099,2107],{"type":48,"tag":106,"props":2091,"children":2092},{},[2093],{"type":48,"tag":62,"props":2094,"children":2096},{"className":2095},[],[2097],{"type":53,"value":2098},"batch_size",{"type":48,"tag":106,"props":2100,"children":2101},{},[2102],{"type":48,"tag":62,"props":2103,"children":2105},{"className":2104},[],[2106],{"type":53,"value":164},{"type":48,"tag":106,"props":2108,"children":2109},{},[2110],{"type":53,"value":2111},"Phase 1 only. Per-GPU batch size.",{"type":48,"tag":79,"props":2113,"children":2114},{},[2115,2124,2133],{"type":48,"tag":106,"props":2116,"children":2117},{},[2118],{"type":48,"tag":62,"props":2119,"children":2121},{"className":2120},[],[2122],{"type":53,"value":2123},"image_size",{"type":48,"tag":106,"props":2125,"children":2126},{},[2127],{"type":48,"tag":62,"props":2128,"children":2130},{"className":2129},[],[2131],{"type":53,"value":2132},"512",{"type":48,"tag":106,"props":2134,"children":2135},{},[2136],{"type":53,"value":2137},"Phase 1 only. Training resolution (square).",{"type":48,"tag":79,"props":2139,"children":2140},{},[2141,2150,2159],{"type":48,"tag":106,"props":2142,"children":2143},{},[2144],{"type":48,"tag":62,"props":2145,"children":2147},{"className":2146},[],[2148],{"type":53,"value":2149},"guidance_range",{"type":48,"tag":106,"props":2151,"children":2152},{},[2153],{"type":48,"tag":62,"props":2154,"children":2156},{"className":2155},[],[2157],{"type":53,"value":2158},"1.5 10.0",{"type":48,"tag":106,"props":2160,"children":2161},{},[2162],{"type":53,"value":2163},"Phase 5 search draw range for guidance.",{"type":48,"tag":79,"props":2165,"children":2166},{},[2167,2176,2184],{"type":48,"tag":106,"props":2168,"children":2169},{},[2170],{"type":48,"tag":62,"props":2171,"children":2173},{"className":2172},[],[2174],{"type":53,"value":2175},"crop_ratio_range",{"type":48,"tag":106,"props":2177,"children":2178},{},[2179],{"type":48,"tag":62,"props":2180,"children":2182},{"className":2181},[],[2183],{"type":53,"value":2158},{"type":48,"tag":106,"props":2185,"children":2186},{},[2187],{"type":53,"value":2188},"Phase 5 search draw range for crop_ratio.",{"type":48,"tag":1348,"props":2190,"children":2191},{},[],{"type":48,"tag":557,"props":2193,"children":2195},{"id":2194},"mode-validation-fail-fast-before-any-phase",[2196],{"type":53,"value":2197},"Mode validation (fail fast before any phase)",{"type":48,"tag":2199,"props":2200,"children":2201},"ul",{},[2202,2245,2286],{"type":48,"tag":2203,"props":2204,"children":2205},"li",{},[2206,2211,2213],{"type":48,"tag":62,"props":2207,"children":2209},{"className":2208},[],[2210],{"type":53,"value":67},{"type":53,"value":2212}," unset → halt: ",{"type":48,"tag":1597,"props":2214,"children":2215},{},[2216,2218,2223,2225,2230,2232,2237,2238,2243],{"type":53,"value":2217},"\"",{"type":48,"tag":62,"props":2219,"children":2221},{"className":2220},[],[2222],{"type":53,"value":67},{"type":53,"value":2224}," is required (",{"type":48,"tag":62,"props":2226,"children":2228},{"className":2227},[],[2229],{"type":53,"value":148},{"type":53,"value":2231}," | ",{"type":48,"tag":62,"props":2233,"children":2235},{"className":2234},[],[2236],{"type":53,"value":184},{"type":53,"value":2231},{"type":48,"tag":62,"props":2239,"children":2241},{"className":2240},[],[2242],{"type":53,"value":156},{"type":53,"value":2244},").\"",{"type":48,"tag":2203,"props":2246,"children":2247},{},[2248,2253,2255,2260,2261,2266,2268],{"type":48,"tag":62,"props":2249,"children":2251},{"className":2250},[],[2252],{"type":53,"value":869},{"type":53,"value":2254}," missing either ",{"type":48,"tag":62,"props":2256,"children":2258},{"className":2257},[],[2259],{"type":53,"value":1651},{"type":53,"value":2024},{"type":48,"tag":62,"props":2262,"children":2264},{"className":2263},[],[2265],{"type":53,"value":1346},{"type":53,"value":2267}," → halt: ",{"type":48,"tag":1597,"props":2269,"children":2270},{},[2271,2273,2278,2279,2284],{"type":53,"value":2272},"\"inference_only requires both ",{"type":48,"tag":62,"props":2274,"children":2276},{"className":2275},[],[2277],{"type":53,"value":1651},{"type":53,"value":952},{"type":48,"tag":62,"props":2280,"children":2282},{"className":2281},[],[2283],{"type":53,"value":1346},{"type":53,"value":2285},".\"",{"type":48,"tag":2203,"props":2287,"children":2288},{},[2289,2294,2296,2301,2302,2307,2309],{"type":48,"tag":62,"props":2290,"children":2292},{"className":2291},[],[2293],{"type":53,"value":610},{"type":53,"value":2295}," with ",{"type":48,"tag":62,"props":2297,"children":2299},{"className":2298},[],[2300],{"type":53,"value":1651},{"type":53,"value":2024},{"type":48,"tag":62,"props":2303,"children":2305},{"className":2304},[],[2306],{"type":53,"value":1346},{"type":53,"value":2308}," supplied → halt: ",{"type":48,"tag":1597,"props":2310,"children":2311},{},[2312,2314,2319],{"type":53,"value":2313},"\"full mode runs finetune; use ",{"type":48,"tag":62,"props":2315,"children":2317},{"className":2316},[],[2318],{"type":53,"value":869},{"type":53,"value":2320}," to reuse an existing checkpoint.\"",{"type":48,"tag":557,"props":2322,"children":2324},{"id":2323},"shared-variables",[2325],{"type":53,"value":2326},"Shared variables",{"type":48,"tag":56,"props":2328,"children":2329},{},[2330],{"type":53,"value":2331},"Set once before Phase 0:",{"type":48,"tag":455,"props":2333,"children":2335},{"className":457,"code":2334,"language":459,"meta":460,"style":460},"MODE=\u003Cfull|inference_only|finetune_only>\nNAME=\u003Cexp>\nDATASET_DIR=\u003Cdataset_dir>\nCLEAN_DIR=${clean_dir:-${DATASET_DIR}}\nCKPT=\u003Ccheckpoint_dir>      # required iff MODE=inference_only; auto-derived after Phase 1 when MODE=full\nSTEP=\u003Citer>                # required iff MODE=inference_only; auto-derived after Phase 1 when MODE=full\nNUM_SDG=\u003CN>\nDEFECT_DESC=\u003Cdefect_spec.jsonl>\nDEFECTS=(T+A T+B)          # TEXTURE+TYPE names. For mode=inference_only, derive from ${CKPT}\u002Fag_config.yaml → dataloader_train.dataset.anomaly_types (also printed by validate_checkpoint.py in Phase 0). For mode=full, take from DEFECT_DESC entries. See references\u002Finference.md §Phase 0.\nNUM_SEARCH_RUN=${num_search_run:-3}\nNN_THRESHOLD=${nn_threshold:-0.4}\nMODEL_SIZE=\u003C2b|14b>\nNUM_GPUS=${num_gpus:-1}\nMAX_ITER=${max_iter:-75000}\nSAVE_ITER=${save_iter:-5000}\nVALIDATION_ITER=${validation_iter:-5000}\nLR=${lr:-0.02}\nBATCH_SIZE=${batch_size:-2}\nIMAGE_SIZE=${image_size:-512}\nVALIDATION_JSONL=${validation_jsonl:-}  # optional; set by Phase 1 Step 2 if not user-supplied\n\nBASE=results\u002F${NAME}\nJSONL=ag_inference\u002F${NAME}\u002Ftestcase.jsonl\nORIGINAL=${BASE}\u002Foriginal\nSEARCHED=${BASE}\u002Fsearched\nROUNDS=${BASE}\u002Frounds\nREGENS=${BASE}\u002Fregens\n",[2336],{"type":48,"tag":62,"props":2337,"children":2338},{"__ignoreMap":460},[2339,2377,2397,2416,2447,2472,2497,2517,2537,2569,2599,2627,2654,2682,2711,2740,2769,2798,2827,2856,2883,2891,2921,2956,2982,3008,3034],{"type":48,"tag":466,"props":2340,"children":2341},{"class":468,"line":469},[2342,2346,2351,2355,2360,2364,2368,2372],{"type":48,"tag":466,"props":2343,"children":2344},{"style":479},[2345],{"type":53,"value":677},{"type":48,"tag":466,"props":2347,"children":2348},{"style":485},[2349],{"type":53,"value":2350},"=\u003C",{"type":48,"tag":466,"props":2352,"children":2353},{"style":502},[2354],{"type":53,"value":148},{"type":48,"tag":466,"props":2356,"children":2357},{"style":485},[2358],{"type":53,"value":2359},"|",{"type":48,"tag":466,"props":2361,"children":2362},{"style":496},[2363],{"type":53,"value":184},{"type":48,"tag":466,"props":2365,"children":2366},{"style":485},[2367],{"type":53,"value":2359},{"type":48,"tag":466,"props":2369,"children":2370},{"style":496},[2371],{"type":53,"value":156},{"type":48,"tag":466,"props":2373,"children":2374},{"style":479},[2375],{"type":53,"value":2376},">\n",{"type":48,"tag":466,"props":2378,"children":2379},{"class":468,"line":631},[2380,2384,2388,2393],{"type":48,"tag":466,"props":2381,"children":2382},{"style":479},[2383],{"type":53,"value":695},{"type":48,"tag":466,"props":2385,"children":2386},{"style":485},[2387],{"type":53,"value":2350},{"type":48,"tag":466,"props":2389,"children":2390},{"style":502},[2391],{"type":53,"value":2392},"exp",{"type":48,"tag":466,"props":2394,"children":2395},{"style":485},[2396],{"type":53,"value":2376},{"type":48,"tag":466,"props":2398,"children":2399},{"class":468,"line":671},[2400,2404,2408,2412],{"type":48,"tag":466,"props":2401,"children":2402},{"style":479},[2403],{"type":53,"value":713},{"type":48,"tag":466,"props":2405,"children":2406},{"style":485},[2407],{"type":53,"value":2350},{"type":48,"tag":466,"props":2409,"children":2410},{"style":502},[2411],{"type":53,"value":139},{"type":48,"tag":466,"props":2413,"children":2414},{"style":485},[2415],{"type":53,"value":2376},{"type":48,"tag":466,"props":2417,"children":2418},{"class":468,"line":689},[2419,2424,2429,2433,2438,2442],{"type":48,"tag":466,"props":2420,"children":2421},{"style":479},[2422],{"type":53,"value":2423},"CLEAN_DIR",{"type":48,"tag":466,"props":2425,"children":2426},{"style":485},[2427],{"type":53,"value":2428},"=${",{"type":48,"tag":466,"props":2430,"children":2431},{"style":479},[2432],{"type":53,"value":1728},{"type":48,"tag":466,"props":2434,"children":2435},{"style":485},[2436],{"type":53,"value":2437},":-${",{"type":48,"tag":466,"props":2439,"children":2440},{"style":479},[2441],{"type":53,"value":713},{"type":48,"tag":466,"props":2443,"children":2444},{"style":485},[2445],{"type":53,"value":2446},"}}\n",{"type":48,"tag":466,"props":2448,"children":2449},{"class":468,"line":707},[2450,2454,2458,2462,2467],{"type":48,"tag":466,"props":2451,"children":2452},{"style":479},[2453],{"type":53,"value":877},{"type":48,"tag":466,"props":2455,"children":2456},{"style":485},[2457],{"type":53,"value":2350},{"type":48,"tag":466,"props":2459,"children":2460},{"style":502},[2461],{"type":53,"value":1651},{"type":48,"tag":466,"props":2463,"children":2464},{"style":485},[2465],{"type":53,"value":2466},">",{"type":48,"tag":466,"props":2468,"children":2469},{"style":625},[2470],{"type":53,"value":2471},"      # required iff MODE=inference_only; auto-derived after Phase 1 when MODE=full\n",{"type":48,"tag":466,"props":2473,"children":2474},{"class":468,"line":725},[2475,2479,2483,2488,2492],{"type":48,"tag":466,"props":2476,"children":2477},{"style":479},[2478],{"type":53,"value":885},{"type":48,"tag":466,"props":2480,"children":2481},{"style":485},[2482],{"type":53,"value":2350},{"type":48,"tag":466,"props":2484,"children":2485},{"style":502},[2486],{"type":53,"value":2487},"iter",{"type":48,"tag":466,"props":2489,"children":2490},{"style":485},[2491],{"type":53,"value":2466},{"type":48,"tag":466,"props":2493,"children":2494},{"style":625},[2495],{"type":53,"value":2496},"                # required iff MODE=inference_only; auto-derived after Phase 1 when MODE=full\n",{"type":48,"tag":466,"props":2498,"children":2499},{"class":468,"line":743},[2500,2504,2508,2513],{"type":48,"tag":466,"props":2501,"children":2502},{"style":479},[2503],{"type":53,"value":749},{"type":48,"tag":466,"props":2505,"children":2506},{"style":485},[2507],{"type":53,"value":2350},{"type":48,"tag":466,"props":2509,"children":2510},{"style":502},[2511],{"type":53,"value":2512},"N",{"type":48,"tag":466,"props":2514,"children":2515},{"style":485},[2516],{"type":53,"value":2376},{"type":48,"tag":466,"props":2518,"children":2519},{"class":468,"line":761},[2520,2524,2528,2533],{"type":48,"tag":466,"props":2521,"children":2522},{"style":479},[2523],{"type":53,"value":731},{"type":48,"tag":466,"props":2525,"children":2526},{"style":485},[2527],{"type":53,"value":2350},{"type":48,"tag":466,"props":2529,"children":2530},{"style":502},[2531],{"type":53,"value":2532},"defect_spec.jsonl",{"type":48,"tag":466,"props":2534,"children":2535},{"style":485},[2536],{"type":53,"value":2376},{"type":48,"tag":466,"props":2538,"children":2539},{"class":468,"line":779},[2540,2545,2550,2555,2560,2564],{"type":48,"tag":466,"props":2541,"children":2542},{"style":479},[2543],{"type":53,"value":2544},"DEFECTS",{"type":48,"tag":466,"props":2546,"children":2547},{"style":485},[2548],{"type":53,"value":2549},"=(",{"type":48,"tag":466,"props":2551,"children":2552},{"style":502},[2553],{"type":53,"value":2554},"T+A",{"type":48,"tag":466,"props":2556,"children":2557},{"style":502},[2558],{"type":53,"value":2559}," T+B",{"type":48,"tag":466,"props":2561,"children":2562},{"style":485},[2563],{"type":53,"value":510},{"type":48,"tag":466,"props":2565,"children":2566},{"style":625},[2567],{"type":53,"value":2568},"          # TEXTURE+TYPE names. For mode=inference_only, derive from ${CKPT}\u002Fag_config.yaml → dataloader_train.dataset.anomaly_types (also printed by validate_checkpoint.py in Phase 0). For mode=full, take from DEFECT_DESC entries. See references\u002Finference.md §Phase 0.\n",{"type":48,"tag":466,"props":2570,"children":2571},{"class":468,"line":789},[2572,2577,2581,2585,2590,2594],{"type":48,"tag":466,"props":2573,"children":2574},{"style":479},[2575],{"type":53,"value":2576},"NUM_SEARCH_RUN",{"type":48,"tag":466,"props":2578,"children":2579},{"style":485},[2580],{"type":53,"value":2428},{"type":48,"tag":466,"props":2582,"children":2583},{"style":479},[2584],{"type":53,"value":1798},{"type":48,"tag":466,"props":2586,"children":2587},{"style":485},[2588],{"type":53,"value":2589},":-",{"type":48,"tag":466,"props":2591,"children":2592},{"style":479},[2593],{"type":53,"value":192},{"type":48,"tag":466,"props":2595,"children":2596},{"style":485},[2597],{"type":53,"value":2598},"}\n",{"type":48,"tag":466,"props":2600,"children":2601},{"class":468,"line":798},[2602,2607,2611,2615,2619,2623],{"type":48,"tag":466,"props":2603,"children":2604},{"style":479},[2605],{"type":53,"value":2606},"NN_THRESHOLD",{"type":48,"tag":466,"props":2608,"children":2609},{"style":485},[2610],{"type":53,"value":2428},{"type":48,"tag":466,"props":2612,"children":2613},{"style":479},[2614],{"type":53,"value":391},{"type":48,"tag":466,"props":2616,"children":2617},{"style":485},[2618],{"type":53,"value":2589},{"type":48,"tag":466,"props":2620,"children":2621},{"style":479},[2622],{"type":53,"value":399},{"type":48,"tag":466,"props":2624,"children":2625},{"style":485},[2626],{"type":53,"value":2598},{"type":48,"tag":466,"props":2628,"children":2629},{"class":468,"line":844},[2630,2634,2638,2642,2646,2650],{"type":48,"tag":466,"props":2631,"children":2632},{"style":479},[2633],{"type":53,"value":767},{"type":48,"tag":466,"props":2635,"children":2636},{"style":485},[2637],{"type":53,"value":2350},{"type":48,"tag":466,"props":2639,"children":2640},{"style":502},[2641],{"type":53,"value":2014},{"type":48,"tag":466,"props":2643,"children":2644},{"style":485},[2645],{"type":53,"value":2359},{"type":48,"tag":466,"props":2647,"children":2648},{"style":496},[2649],{"type":53,"value":2030},{"type":48,"tag":466,"props":2651,"children":2652},{"style":479},[2653],{"type":53,"value":2376},{"type":48,"tag":466,"props":2655,"children":2656},{"class":468,"line":852},[2657,2662,2666,2670,2674,2678],{"type":48,"tag":466,"props":2658,"children":2659},{"style":479},[2660],{"type":53,"value":2661},"NUM_GPUS",{"type":48,"tag":466,"props":2663,"children":2664},{"style":485},[2665],{"type":53,"value":2428},{"type":48,"tag":466,"props":2667,"children":2668},{"style":479},[2669],{"type":53,"value":1980},{"type":48,"tag":466,"props":2671,"children":2672},{"style":485},[2673],{"type":53,"value":2589},{"type":48,"tag":466,"props":2675,"children":2676},{"style":479},[2677],{"type":53,"value":128},{"type":48,"tag":466,"props":2679,"children":2680},{"style":485},[2681],{"type":53,"value":2598},{"type":48,"tag":466,"props":2683,"children":2685},{"class":468,"line":2684},14,[2686,2691,2695,2699,2703,2707],{"type":48,"tag":466,"props":2687,"children":2688},{"style":479},[2689],{"type":53,"value":2690},"MAX_ITER",{"type":48,"tag":466,"props":2692,"children":2693},{"style":485},[2694],{"type":53,"value":2428},{"type":48,"tag":466,"props":2696,"children":2697},{"style":479},[2698],{"type":53,"value":1896},{"type":48,"tag":466,"props":2700,"children":2701},{"style":485},[2702],{"type":53,"value":2589},{"type":48,"tag":466,"props":2704,"children":2705},{"style":479},[2706],{"type":53,"value":1905},{"type":48,"tag":466,"props":2708,"children":2709},{"style":485},[2710],{"type":53,"value":2598},{"type":48,"tag":466,"props":2712,"children":2714},{"class":468,"line":2713},15,[2715,2720,2724,2728,2732,2736],{"type":48,"tag":466,"props":2716,"children":2717},{"style":479},[2718],{"type":53,"value":2719},"SAVE_ITER",{"type":48,"tag":466,"props":2721,"children":2722},{"style":485},[2723],{"type":53,"value":2428},{"type":48,"tag":466,"props":2725,"children":2726},{"style":479},[2727],{"type":53,"value":1922},{"type":48,"tag":466,"props":2729,"children":2730},{"style":485},[2731],{"type":53,"value":2589},{"type":48,"tag":466,"props":2733,"children":2734},{"style":479},[2735],{"type":53,"value":1931},{"type":48,"tag":466,"props":2737,"children":2738},{"style":485},[2739],{"type":53,"value":2598},{"type":48,"tag":466,"props":2741,"children":2743},{"class":468,"line":2742},16,[2744,2749,2753,2757,2761,2765],{"type":48,"tag":466,"props":2745,"children":2746},{"style":479},[2747],{"type":53,"value":2748},"VALIDATION_ITER",{"type":48,"tag":466,"props":2750,"children":2751},{"style":485},[2752],{"type":53,"value":2428},{"type":48,"tag":466,"props":2754,"children":2755},{"style":479},[2756],{"type":53,"value":1948},{"type":48,"tag":466,"props":2758,"children":2759},{"style":485},[2760],{"type":53,"value":2589},{"type":48,"tag":466,"props":2762,"children":2763},{"style":479},[2764],{"type":53,"value":1931},{"type":48,"tag":466,"props":2766,"children":2767},{"style":485},[2768],{"type":53,"value":2598},{"type":48,"tag":466,"props":2770,"children":2772},{"class":468,"line":2771},17,[2773,2778,2782,2786,2790,2794],{"type":48,"tag":466,"props":2774,"children":2775},{"style":479},[2776],{"type":53,"value":2777},"LR",{"type":48,"tag":466,"props":2779,"children":2780},{"style":485},[2781],{"type":53,"value":2428},{"type":48,"tag":466,"props":2783,"children":2784},{"style":479},[2785],{"type":53,"value":2072},{"type":48,"tag":466,"props":2787,"children":2788},{"style":485},[2789],{"type":53,"value":2589},{"type":48,"tag":466,"props":2791,"children":2792},{"style":479},[2793],{"type":53,"value":2081},{"type":48,"tag":466,"props":2795,"children":2796},{"style":485},[2797],{"type":53,"value":2598},{"type":48,"tag":466,"props":2799,"children":2801},{"class":468,"line":2800},18,[2802,2807,2811,2815,2819,2823],{"type":48,"tag":466,"props":2803,"children":2804},{"style":479},[2805],{"type":53,"value":2806},"BATCH_SIZE",{"type":48,"tag":466,"props":2808,"children":2809},{"style":485},[2810],{"type":53,"value":2428},{"type":48,"tag":466,"props":2812,"children":2813},{"style":479},[2814],{"type":53,"value":2098},{"type":48,"tag":466,"props":2816,"children":2817},{"style":485},[2818],{"type":53,"value":2589},{"type":48,"tag":466,"props":2820,"children":2821},{"style":479},[2822],{"type":53,"value":164},{"type":48,"tag":466,"props":2824,"children":2825},{"style":485},[2826],{"type":53,"value":2598},{"type":48,"tag":466,"props":2828,"children":2830},{"class":468,"line":2829},19,[2831,2836,2840,2844,2848,2852],{"type":48,"tag":466,"props":2832,"children":2833},{"style":479},[2834],{"type":53,"value":2835},"IMAGE_SIZE",{"type":48,"tag":466,"props":2837,"children":2838},{"style":485},[2839],{"type":53,"value":2428},{"type":48,"tag":466,"props":2841,"children":2842},{"style":479},[2843],{"type":53,"value":2123},{"type":48,"tag":466,"props":2845,"children":2846},{"style":485},[2847],{"type":53,"value":2589},{"type":48,"tag":466,"props":2849,"children":2850},{"style":479},[2851],{"type":53,"value":2132},{"type":48,"tag":466,"props":2853,"children":2854},{"style":485},[2855],{"type":53,"value":2598},{"type":48,"tag":466,"props":2857,"children":2859},{"class":468,"line":2858},20,[2860,2865,2869,2873,2878],{"type":48,"tag":466,"props":2861,"children":2862},{"style":479},[2863],{"type":53,"value":2864},"VALIDATION_JSONL",{"type":48,"tag":466,"props":2866,"children":2867},{"style":485},[2868],{"type":53,"value":2428},{"type":48,"tag":466,"props":2870,"children":2871},{"style":479},[2872],{"type":53,"value":1769},{"type":48,"tag":466,"props":2874,"children":2875},{"style":485},[2876],{"type":53,"value":2877},":-}",{"type":48,"tag":466,"props":2879,"children":2880},{"style":625},[2881],{"type":53,"value":2882},"  # optional; set by Phase 1 Step 2 if not user-supplied\n",{"type":48,"tag":466,"props":2884,"children":2886},{"class":468,"line":2885},21,[2887],{"type":48,"tag":466,"props":2888,"children":2889},{"emptyLinePlaceholder":783},[2890],{"type":53,"value":786},{"type":48,"tag":466,"props":2892,"children":2894},{"class":468,"line":2893},22,[2895,2900,2904,2909,2913,2917],{"type":48,"tag":466,"props":2896,"children":2897},{"style":479},[2898],{"type":53,"value":2899},"BASE",{"type":48,"tag":466,"props":2901,"children":2902},{"style":485},[2903],{"type":53,"value":488},{"type":48,"tag":466,"props":2905,"children":2906},{"style":502},[2907],{"type":53,"value":2908},"results\u002F",{"type":48,"tag":466,"props":2910,"children":2911},{"style":485},[2912],{"type":53,"value":804},{"type":48,"tag":466,"props":2914,"children":2915},{"style":479},[2916],{"type":53,"value":695},{"type":48,"tag":466,"props":2918,"children":2919},{"style":485},[2920],{"type":53,"value":2598},{"type":48,"tag":466,"props":2922,"children":2924},{"class":468,"line":2923},23,[2925,2930,2934,2939,2943,2947,2951],{"type":48,"tag":466,"props":2926,"children":2927},{"style":479},[2928],{"type":53,"value":2929},"JSONL",{"type":48,"tag":466,"props":2931,"children":2932},{"style":485},[2933],{"type":53,"value":488},{"type":48,"tag":466,"props":2935,"children":2936},{"style":502},[2937],{"type":53,"value":2938},"ag_inference\u002F",{"type":48,"tag":466,"props":2940,"children":2941},{"style":485},[2942],{"type":53,"value":804},{"type":48,"tag":466,"props":2944,"children":2945},{"style":479},[2946],{"type":53,"value":695},{"type":48,"tag":466,"props":2948,"children":2949},{"style":485},[2950],{"type":53,"value":813},{"type":48,"tag":466,"props":2952,"children":2953},{"style":502},[2954],{"type":53,"value":2955},"\u002Ftestcase.jsonl\n",{"type":48,"tag":466,"props":2957,"children":2959},{"class":468,"line":2958},24,[2960,2965,2969,2973,2977],{"type":48,"tag":466,"props":2961,"children":2962},{"style":479},[2963],{"type":53,"value":2964},"ORIGINAL",{"type":48,"tag":466,"props":2966,"children":2967},{"style":485},[2968],{"type":53,"value":2428},{"type":48,"tag":466,"props":2970,"children":2971},{"style":479},[2972],{"type":53,"value":2899},{"type":48,"tag":466,"props":2974,"children":2975},{"style":485},[2976],{"type":53,"value":813},{"type":48,"tag":466,"props":2978,"children":2979},{"style":502},[2980],{"type":53,"value":2981},"\u002Foriginal\n",{"type":48,"tag":466,"props":2983,"children":2985},{"class":468,"line":2984},25,[2986,2991,2995,2999,3003],{"type":48,"tag":466,"props":2987,"children":2988},{"style":479},[2989],{"type":53,"value":2990},"SEARCHED",{"type":48,"tag":466,"props":2992,"children":2993},{"style":485},[2994],{"type":53,"value":2428},{"type":48,"tag":466,"props":2996,"children":2997},{"style":479},[2998],{"type":53,"value":2899},{"type":48,"tag":466,"props":3000,"children":3001},{"style":485},[3002],{"type":53,"value":813},{"type":48,"tag":466,"props":3004,"children":3005},{"style":502},[3006],{"type":53,"value":3007},"\u002Fsearched\n",{"type":48,"tag":466,"props":3009,"children":3011},{"class":468,"line":3010},26,[3012,3017,3021,3025,3029],{"type":48,"tag":466,"props":3013,"children":3014},{"style":479},[3015],{"type":53,"value":3016},"ROUNDS",{"type":48,"tag":466,"props":3018,"children":3019},{"style":485},[3020],{"type":53,"value":2428},{"type":48,"tag":466,"props":3022,"children":3023},{"style":479},[3024],{"type":53,"value":2899},{"type":48,"tag":466,"props":3026,"children":3027},{"style":485},[3028],{"type":53,"value":813},{"type":48,"tag":466,"props":3030,"children":3031},{"style":502},[3032],{"type":53,"value":3033},"\u002Frounds\n",{"type":48,"tag":466,"props":3035,"children":3037},{"class":468,"line":3036},27,[3038,3043,3047,3051,3055],{"type":48,"tag":466,"props":3039,"children":3040},{"style":479},[3041],{"type":53,"value":3042},"REGENS",{"type":48,"tag":466,"props":3044,"children":3045},{"style":485},[3046],{"type":53,"value":2428},{"type":48,"tag":466,"props":3048,"children":3049},{"style":479},[3050],{"type":53,"value":2899},{"type":48,"tag":466,"props":3052,"children":3053},{"style":485},[3054],{"type":53,"value":813},{"type":48,"tag":466,"props":3056,"children":3057},{"style":502},[3058],{"type":53,"value":3059},"\u002Fregens\n",{"type":48,"tag":557,"props":3061,"children":3063},{"id":3062},"guard-preflight-product-mode-only",[3064],{"type":53,"value":3065},"Guard preflight (product mode only)",{"type":48,"tag":56,"props":3067,"children":3068},{},[3069,3071,3076,3078,3084,3086,3091,3093,3099,3101,3107,3109,3115,3117,3122,3124,3130],{"type":53,"value":3070},"When ",{"type":48,"tag":62,"props":3072,"children":3074},{"className":3073},[],[3075],{"type":53,"value":537},{"type":53,"value":3077},", run\n",{"type":48,"tag":62,"props":3079,"children":3081},{"className":3080},[],[3082],{"type":53,"value":3083},".agents\u002Fskills\u002Fanomalygen-guard\u002Fscripts\u002Fpreflight.py",{"type":53,"value":3085}," before any GPU work and\nfix any ",{"type":48,"tag":62,"props":3087,"children":3089},{"className":3088},[],[3090],{"type":53,"value":553},{"type":53,"value":3092}," issues. ",{"type":48,"tag":62,"props":3094,"children":3096},{"className":3095},[],[3097],{"type":53,"value":3098},"--validation-jsonl",{"type":53,"value":3100}," is forwarded only when the user\nsupplied one; for ",{"type":48,"tag":62,"props":3102,"children":3104},{"className":3103},[],[3105],{"type":53,"value":3106},"MODE=finetune_only",{"type":53,"value":3108}," omit ",{"type":48,"tag":62,"props":3110,"children":3112},{"className":3111},[],[3113],{"type":53,"value":3114},"--num-sdg",{"type":53,"value":3116}," if not supplied. See\n",{"type":48,"tag":62,"props":3118,"children":3120},{"className":3119},[],[3121],{"type":53,"value":1269},{"type":53,"value":3123}," for the full preflight command with all\nforwarded flags and the validation-JSONL \u002F ",{"type":48,"tag":62,"props":3125,"children":3127},{"className":3126},[],[3128],{"type":53,"value":3129},"allocate_samples.py",{"type":53,"value":3131}," 0-entry\nchecks.",{"type":48,"tag":1348,"props":3133,"children":3134},{},[],{"type":48,"tag":557,"props":3136,"children":3138},{"id":3137},"phase-0-checkpoints",[3139],{"type":53,"value":3140},"Phase 0 — checkpoints",{"type":48,"tag":56,"props":3142,"children":3143},{},[3144,3145,3151],{"type":53,"value":1002},{"type":48,"tag":62,"props":3146,"children":3148},{"className":3147},[],[3149],{"type":53,"value":3150},"references\u002Ffinetune.md §Phase 0",{"type":53,"value":3152}," for HF_TOKEN requirements and what gets\ndownloaded (~140 GB). Verify first; download only what is missing.",{"type":48,"tag":455,"props":3154,"children":3156},{"className":457,"code":3155,"language":459,"meta":460,"style":460},"${ANOMALYGEN_SCRIPTS}\u002Fcheck.sh \\\n    || ${ANOMALYGEN_SCRIPTS}\u002Fdownload_checkpoints.sh\n",[3157],{"type":48,"tag":62,"props":3158,"children":3159},{"__ignoreMap":460},[3160,3180],{"type":48,"tag":466,"props":3161,"children":3162},{"class":468,"line":469},[3163,3167,3171,3175],{"type":48,"tag":466,"props":3164,"children":3165},{"style":485},[3166],{"type":53,"value":804},{"type":48,"tag":466,"props":3168,"children":3169},{"style":479},[3170],{"type":53,"value":597},{"type":48,"tag":466,"props":3172,"children":3173},{"style":485},[3174],{"type":53,"value":813},{"type":48,"tag":466,"props":3176,"children":3177},{"style":479},[3178],{"type":53,"value":3179},"\u002Fcheck.sh \\\n",{"type":48,"tag":466,"props":3181,"children":3182},{"class":468,"line":631},[3183,3188,3192,3196,3200],{"type":48,"tag":466,"props":3184,"children":3185},{"style":485},[3186],{"type":53,"value":3187},"    ||",{"type":48,"tag":466,"props":3189,"children":3190},{"style":485},[3191],{"type":53,"value":828},{"type":48,"tag":466,"props":3193,"children":3194},{"style":479},[3195],{"type":53,"value":597},{"type":48,"tag":466,"props":3197,"children":3198},{"style":485},[3199],{"type":53,"value":813},{"type":48,"tag":466,"props":3201,"children":3202},{"style":496},[3203],{"type":53,"value":841},{"type":48,"tag":1348,"props":3205,"children":3206},{},[],{"type":48,"tag":557,"props":3208,"children":3210},{"id":3209},"phase-1-fine-tune-skip-when-modeinference_only",[3211,3213,3219],{"type":53,"value":3212},"Phase 1 — fine-tune (skip when ",{"type":48,"tag":62,"props":3214,"children":3216},{"className":3215},[],[3217],{"type":53,"value":3218},"MODE=inference_only",{"type":53,"value":510},{"type":48,"tag":56,"props":3221,"children":3222},{},[3223,3224,3230,3232,3237,3239,3244,3246,3251,3253,3258,3260,3265,3267,3272,3274,3279,3281,3286,3288,3293,3294,3299],{"type":53,"value":1002},{"type":48,"tag":62,"props":3225,"children":3227},{"className":3226},[],[3228],{"type":53,"value":3229},"references\u002Ffinetune.md §Phase 1",{"type":53,"value":3231}," for dataset structure, config template\ndetails, and best-checkpoint selection. Four steps: (1) validate dataset \u002F\nderive anomaly types, (2) generate the validation JSONL (skip if user supplied\n",{"type":48,"tag":62,"props":3233,"children":3235},{"className":3234},[],[3236],{"type":53,"value":2864},{"type":53,"value":3238},"), (3) generate the training config — ",{"type":48,"tag":431,"props":3240,"children":3241},{},[3242],{"type":53,"value":3243},"show it to the user\nand confirm before writing",{"type":53,"value":3245}," — (4) launch training in the background. Then\nderive ",{"type":48,"tag":62,"props":3247,"children":3249},{"className":3248},[],[3250],{"type":53,"value":877},{"type":53,"value":3252}," (path encodes upper-case ",{"type":48,"tag":62,"props":3254,"children":3256},{"className":3255},[],[3257],{"type":53,"value":767},{"type":53,"value":3259},") and ",{"type":48,"tag":62,"props":3261,"children":3263},{"className":3262},[],[3264],{"type":53,"value":885},{"type":53,"value":3266}," (highest\n",{"type":48,"tag":62,"props":3268,"children":3270},{"className":3269},[],[3271],{"type":53,"value":259},{"type":53,"value":3273}," step from validation logs). If ",{"type":48,"tag":62,"props":3275,"children":3277},{"className":3276},[],[3278],{"type":53,"value":3106},{"type":53,"value":3280},", stop after\ntraining. See ",{"type":48,"tag":62,"props":3282,"children":3284},{"className":3283},[],[3285],{"type":53,"value":1081},{"type":53,"value":3287}," for the exact Step 1–4 commands\nand the ",{"type":48,"tag":62,"props":3289,"children":3291},{"className":3290},[],[3292],{"type":53,"value":877},{"type":53,"value":879},{"type":48,"tag":62,"props":3295,"children":3297},{"className":3296},[],[3298],{"type":53,"value":885},{"type":53,"value":3300}," derivation snippet.",{"type":48,"tag":1348,"props":3302,"children":3303},{},[],{"type":48,"tag":557,"props":3305,"children":3307},{"id":3306},"phase-2-prep-testcase-skip-when-modefinetune_only",[3308,3310,3315],{"type":53,"value":3309},"Phase 2 — prep-testcase (skip when ",{"type":48,"tag":62,"props":3311,"children":3313},{"className":3312},[],[3314],{"type":53,"value":3106},{"type":53,"value":510},{"type":48,"tag":56,"props":3317,"children":3318},{},[3319,3320,3326,3328,3334,3336,3342,3344,3350],{"type":53,"value":1002},{"type":48,"tag":62,"props":3321,"children":3323},{"className":3322},[],[3324],{"type":53,"value":3325},"references\u002Finference.md §Phase 2",{"type":53,"value":3327}," for AMP routing detail and n_seeds\nsizing. Do NOT pass ",{"type":48,"tag":62,"props":3329,"children":3331},{"className":3330},[],[3332],{"type":53,"value":3333},"--seeds",{"type":53,"value":3335}," — it is auto-computed and is not a recognized\nflag. ",{"type":48,"tag":62,"props":3337,"children":3339},{"className":3338},[],[3340],{"type":53,"value":3341},"prep_testcase.sh",{"type":53,"value":3343}," defaults to ",{"type":48,"tag":62,"props":3345,"children":3347},{"className":3346},[],[3348],{"type":53,"value":3349},"--mode inference",{"type":53,"value":3351}," (uniform allocation\nacross defect types, no KPI floor), which Phase 2 always uses.",{"type":48,"tag":455,"props":3353,"children":3355},{"className":457,"code":3354,"language":459,"meta":460,"style":460},"${ANOMALYGEN_SCRIPTS}\u002Fprep_testcase.sh \\\n    --name ${NAME} --num-sdg ${NUM_SDG} \\\n    --dataset-dir ${DATASET_DIR} \\\n    --clean-dir ${CLEAN_DIR} \\\n    --defect-spec ${DEFECT_DESC} \\\n    --amp-output-dir ag_inference\u002F${NAME}\u002Famp \\\n    --output-jsonl ${JSONL}\n",[3356],{"type":48,"tag":62,"props":3357,"children":3358},{"__ignoreMap":460},[3359,3379,3421,3445,3469,3493,3527],{"type":48,"tag":466,"props":3360,"children":3361},{"class":468,"line":469},[3362,3366,3370,3374],{"type":48,"tag":466,"props":3363,"children":3364},{"style":485},[3365],{"type":53,"value":804},{"type":48,"tag":466,"props":3367,"children":3368},{"style":479},[3369],{"type":53,"value":597},{"type":48,"tag":466,"props":3371,"children":3372},{"style":485},[3373],{"type":53,"value":813},{"type":48,"tag":466,"props":3375,"children":3376},{"style":479},[3377],{"type":53,"value":3378},"\u002Fprep_testcase.sh \\\n",{"type":48,"tag":466,"props":3380,"children":3381},{"class":468,"line":631},[3382,3387,3391,3395,3399,3404,3408,3412,3416],{"type":48,"tag":466,"props":3383,"children":3384},{"style":496},[3385],{"type":53,"value":3386},"    --name",{"type":48,"tag":466,"props":3388,"children":3389},{"style":485},[3390],{"type":53,"value":828},{"type":48,"tag":466,"props":3392,"children":3393},{"style":479},[3394],{"type":53,"value":695},{"type":48,"tag":466,"props":3396,"children":3397},{"style":485},[3398],{"type":53,"value":813},{"type":48,"tag":466,"props":3400,"children":3401},{"style":502},[3402],{"type":53,"value":3403}," --num-sdg",{"type":48,"tag":466,"props":3405,"children":3406},{"style":485},[3407],{"type":53,"value":828},{"type":48,"tag":466,"props":3409,"children":3410},{"style":479},[3411],{"type":53,"value":749},{"type":48,"tag":466,"props":3413,"children":3414},{"style":485},[3415],{"type":53,"value":813},{"type":48,"tag":466,"props":3417,"children":3418},{"style":479},[3419],{"type":53,"value":3420}," \\\n",{"type":48,"tag":466,"props":3422,"children":3423},{"class":468,"line":671},[3424,3429,3433,3437,3441],{"type":48,"tag":466,"props":3425,"children":3426},{"style":502},[3427],{"type":53,"value":3428},"    --dataset-dir",{"type":48,"tag":466,"props":3430,"children":3431},{"style":485},[3432],{"type":53,"value":828},{"type":48,"tag":466,"props":3434,"children":3435},{"style":479},[3436],{"type":53,"value":713},{"type":48,"tag":466,"props":3438,"children":3439},{"style":485},[3440],{"type":53,"value":813},{"type":48,"tag":466,"props":3442,"children":3443},{"style":479},[3444],{"type":53,"value":3420},{"type":48,"tag":466,"props":3446,"children":3447},{"class":468,"line":689},[3448,3453,3457,3461,3465],{"type":48,"tag":466,"props":3449,"children":3450},{"style":502},[3451],{"type":53,"value":3452},"    --clean-dir",{"type":48,"tag":466,"props":3454,"children":3455},{"style":485},[3456],{"type":53,"value":828},{"type":48,"tag":466,"props":3458,"children":3459},{"style":479},[3460],{"type":53,"value":2423},{"type":48,"tag":466,"props":3462,"children":3463},{"style":485},[3464],{"type":53,"value":813},{"type":48,"tag":466,"props":3466,"children":3467},{"style":479},[3468],{"type":53,"value":3420},{"type":48,"tag":466,"props":3470,"children":3471},{"class":468,"line":707},[3472,3477,3481,3485,3489],{"type":48,"tag":466,"props":3473,"children":3474},{"style":502},[3475],{"type":53,"value":3476},"    --defect-spec",{"type":48,"tag":466,"props":3478,"children":3479},{"style":485},[3480],{"type":53,"value":828},{"type":48,"tag":466,"props":3482,"children":3483},{"style":479},[3484],{"type":53,"value":731},{"type":48,"tag":466,"props":3486,"children":3487},{"style":485},[3488],{"type":53,"value":813},{"type":48,"tag":466,"props":3490,"children":3491},{"style":479},[3492],{"type":53,"value":3420},{"type":48,"tag":466,"props":3494,"children":3495},{"class":468,"line":725},[3496,3501,3506,3510,3514,3518,3523],{"type":48,"tag":466,"props":3497,"children":3498},{"style":502},[3499],{"type":53,"value":3500},"    --amp-output-dir",{"type":48,"tag":466,"props":3502,"children":3503},{"style":502},[3504],{"type":53,"value":3505}," ag_inference\u002F",{"type":48,"tag":466,"props":3507,"children":3508},{"style":485},[3509],{"type":53,"value":804},{"type":48,"tag":466,"props":3511,"children":3512},{"style":479},[3513],{"type":53,"value":695},{"type":48,"tag":466,"props":3515,"children":3516},{"style":485},[3517],{"type":53,"value":813},{"type":48,"tag":466,"props":3519,"children":3520},{"style":502},[3521],{"type":53,"value":3522},"\u002Famp",{"type":48,"tag":466,"props":3524,"children":3525},{"style":479},[3526],{"type":53,"value":3420},{"type":48,"tag":466,"props":3528,"children":3529},{"class":468,"line":743},[3530,3535,3539,3543],{"type":48,"tag":466,"props":3531,"children":3532},{"style":502},[3533],{"type":53,"value":3534},"    --output-jsonl",{"type":48,"tag":466,"props":3536,"children":3537},{"style":485},[3538],{"type":53,"value":828},{"type":48,"tag":466,"props":3540,"children":3541},{"style":479},[3542],{"type":53,"value":2929},{"type":48,"tag":466,"props":3544,"children":3545},{"style":485},[3546],{"type":53,"value":2598},{"type":48,"tag":56,"props":3548,"children":3549},{},[3550,3555,3557,3562,3564,3569,3571,3576,3578,3583,3585,3590],{"type":48,"tag":431,"props":3551,"children":3552},{},[3553],{"type":53,"value":3554},"Custom per-defect counts:",{"type":53,"value":3556}," when the user specifies counts per defect type,\ntranslate to ",{"type":48,"tag":62,"props":3558,"children":3560},{"className":3559},[],[3561],{"type":53,"value":3114},{"type":53,"value":3563}," plus a ",{"type":48,"tag":62,"props":3565,"children":3567},{"className":3566},[],[3568],{"type":53,"value":1280},{"type":53,"value":3570}," JSON dict (types absent\nfrom the dict get 0; sum should equal ",{"type":48,"tag":62,"props":3572,"children":3574},{"className":3573},[],[3575],{"type":53,"value":3114},{"type":53,"value":3577},", else the script warns on\nstderr and uses the override sum). Confirm the allocation when intent is\nambiguous. See ",{"type":48,"tag":62,"props":3579,"children":3581},{"className":3580},[],[3582],{"type":53,"value":1269},{"type":53,"value":3584}," for the full\n",{"type":48,"tag":62,"props":3586,"children":3588},{"className":3587},[],[3589],{"type":53,"value":1280},{"type":53,"value":3591}," command example and the ambiguity-handling detail.",{"type":48,"tag":1348,"props":3593,"children":3594},{},[],{"type":48,"tag":557,"props":3596,"children":3598},{"id":3597},"phase-3-sdg-original",[3599,3601],{"type":53,"value":3600},"Phase 3 — SDG → ",{"type":48,"tag":62,"props":3602,"children":3604},{"className":3603},[],[3605],{"type":53,"value":203},{"type":48,"tag":56,"props":3607,"children":3608},{},[3609,3610,3616],{"type":53,"value":1002},{"type":48,"tag":62,"props":3611,"children":3613},{"className":3612},[],[3614],{"type":53,"value":3615},"references\u002Finference.md §Phase 3",{"type":53,"value":3617}," for JSONL validation against the\ncheckpoint, multi-GPU caveats, and output verification.",{"type":48,"tag":455,"props":3619,"children":3621},{"className":457,"code":3620,"language":459,"meta":460,"style":460},"python3 -m scripts.utilities.validate_checkpoint ${CKPT} --step ${STEP}\npython3 -m scripts.utilities.validate_jsonl ${CKPT} ${JSONL}\n\n${ANOMALYGEN_SCRIPTS}\u002Frun_sdg.sh \\\n    --checkpoint_dir ${CKPT} --step ${STEP} \\\n    --input_jsonl ${JSONL} --output_dir ${ORIGINAL} \\\n    --model_size ${MODEL_SIZE} --num_gpus ${NUM_GPUS}\n\n${ANOMALYGEN_SCRIPTS}\u002Fverify_output.sh ${JSONL} ${ORIGINAL}\n",[3622],{"type":48,"tag":62,"props":3623,"children":3624},{"__ignoreMap":460},[3625,3672,3712,3719,3739,3779,3820,3857,3864],{"type":48,"tag":466,"props":3626,"children":3627},{"class":468,"line":469},[3628,3633,3638,3643,3647,3651,3655,3660,3664,3668],{"type":48,"tag":466,"props":3629,"children":3630},{"style":496},[3631],{"type":53,"value":3632},"python3",{"type":48,"tag":466,"props":3634,"children":3635},{"style":502},[3636],{"type":53,"value":3637}," -m",{"type":48,"tag":466,"props":3639,"children":3640},{"style":502},[3641],{"type":53,"value":3642}," scripts.utilities.validate_checkpoint",{"type":48,"tag":466,"props":3644,"children":3645},{"style":485},[3646],{"type":53,"value":828},{"type":48,"tag":466,"props":3648,"children":3649},{"style":479},[3650],{"type":53,"value":877},{"type":48,"tag":466,"props":3652,"children":3653},{"style":485},[3654],{"type":53,"value":813},{"type":48,"tag":466,"props":3656,"children":3657},{"style":502},[3658],{"type":53,"value":3659}," --step",{"type":48,"tag":466,"props":3661,"children":3662},{"style":485},[3663],{"type":53,"value":828},{"type":48,"tag":466,"props":3665,"children":3666},{"style":479},[3667],{"type":53,"value":885},{"type":48,"tag":466,"props":3669,"children":3670},{"style":485},[3671],{"type":53,"value":2598},{"type":48,"tag":466,"props":3673,"children":3674},{"class":468,"line":631},[3675,3679,3683,3688,3692,3696,3700,3704,3708],{"type":48,"tag":466,"props":3676,"children":3677},{"style":496},[3678],{"type":53,"value":3632},{"type":48,"tag":466,"props":3680,"children":3681},{"style":502},[3682],{"type":53,"value":3637},{"type":48,"tag":466,"props":3684,"children":3685},{"style":502},[3686],{"type":53,"value":3687}," scripts.utilities.validate_jsonl",{"type":48,"tag":466,"props":3689,"children":3690},{"style":485},[3691],{"type":53,"value":828},{"type":48,"tag":466,"props":3693,"children":3694},{"style":479},[3695],{"type":53,"value":877},{"type":48,"tag":466,"props":3697,"children":3698},{"style":485},[3699],{"type":53,"value":813},{"type":48,"tag":466,"props":3701,"children":3702},{"style":485},[3703],{"type":53,"value":828},{"type":48,"tag":466,"props":3705,"children":3706},{"style":479},[3707],{"type":53,"value":2929},{"type":48,"tag":466,"props":3709,"children":3710},{"style":485},[3711],{"type":53,"value":2598},{"type":48,"tag":466,"props":3713,"children":3714},{"class":468,"line":671},[3715],{"type":48,"tag":466,"props":3716,"children":3717},{"emptyLinePlaceholder":783},[3718],{"type":53,"value":786},{"type":48,"tag":466,"props":3720,"children":3721},{"class":468,"line":689},[3722,3726,3730,3734],{"type":48,"tag":466,"props":3723,"children":3724},{"style":485},[3725],{"type":53,"value":804},{"type":48,"tag":466,"props":3727,"children":3728},{"style":479},[3729],{"type":53,"value":597},{"type":48,"tag":466,"props":3731,"children":3732},{"style":485},[3733],{"type":53,"value":813},{"type":48,"tag":466,"props":3735,"children":3736},{"style":479},[3737],{"type":53,"value":3738},"\u002Frun_sdg.sh \\\n",{"type":48,"tag":466,"props":3740,"children":3741},{"class":468,"line":707},[3742,3747,3751,3755,3759,3763,3767,3771,3775],{"type":48,"tag":466,"props":3743,"children":3744},{"style":496},[3745],{"type":53,"value":3746},"    --checkpoint_dir",{"type":48,"tag":466,"props":3748,"children":3749},{"style":485},[3750],{"type":53,"value":828},{"type":48,"tag":466,"props":3752,"children":3753},{"style":479},[3754],{"type":53,"value":877},{"type":48,"tag":466,"props":3756,"children":3757},{"style":485},[3758],{"type":53,"value":813},{"type":48,"tag":466,"props":3760,"children":3761},{"style":502},[3762],{"type":53,"value":3659},{"type":48,"tag":466,"props":3764,"children":3765},{"style":485},[3766],{"type":53,"value":828},{"type":48,"tag":466,"props":3768,"children":3769},{"style":479},[3770],{"type":53,"value":885},{"type":48,"tag":466,"props":3772,"children":3773},{"style":485},[3774],{"type":53,"value":813},{"type":48,"tag":466,"props":3776,"children":3777},{"style":479},[3778],{"type":53,"value":3420},{"type":48,"tag":466,"props":3780,"children":3781},{"class":468,"line":725},[3782,3787,3791,3795,3799,3804,3808,3812,3816],{"type":48,"tag":466,"props":3783,"children":3784},{"style":502},[3785],{"type":53,"value":3786},"    --input_jsonl",{"type":48,"tag":466,"props":3788,"children":3789},{"style":485},[3790],{"type":53,"value":828},{"type":48,"tag":466,"props":3792,"children":3793},{"style":479},[3794],{"type":53,"value":2929},{"type":48,"tag":466,"props":3796,"children":3797},{"style":485},[3798],{"type":53,"value":813},{"type":48,"tag":466,"props":3800,"children":3801},{"style":502},[3802],{"type":53,"value":3803}," --output_dir",{"type":48,"tag":466,"props":3805,"children":3806},{"style":485},[3807],{"type":53,"value":828},{"type":48,"tag":466,"props":3809,"children":3810},{"style":479},[3811],{"type":53,"value":2964},{"type":48,"tag":466,"props":3813,"children":3814},{"style":485},[3815],{"type":53,"value":813},{"type":48,"tag":466,"props":3817,"children":3818},{"style":479},[3819],{"type":53,"value":3420},{"type":48,"tag":466,"props":3821,"children":3822},{"class":468,"line":743},[3823,3828,3832,3836,3840,3845,3849,3853],{"type":48,"tag":466,"props":3824,"children":3825},{"style":502},[3826],{"type":53,"value":3827},"    --model_size",{"type":48,"tag":466,"props":3829,"children":3830},{"style":485},[3831],{"type":53,"value":828},{"type":48,"tag":466,"props":3833,"children":3834},{"style":479},[3835],{"type":53,"value":767},{"type":48,"tag":466,"props":3837,"children":3838},{"style":485},[3839],{"type":53,"value":813},{"type":48,"tag":466,"props":3841,"children":3842},{"style":502},[3843],{"type":53,"value":3844}," --num_gpus",{"type":48,"tag":466,"props":3846,"children":3847},{"style":485},[3848],{"type":53,"value":828},{"type":48,"tag":466,"props":3850,"children":3851},{"style":479},[3852],{"type":53,"value":2661},{"type":48,"tag":466,"props":3854,"children":3855},{"style":485},[3856],{"type":53,"value":2598},{"type":48,"tag":466,"props":3858,"children":3859},{"class":468,"line":761},[3860],{"type":48,"tag":466,"props":3861,"children":3862},{"emptyLinePlaceholder":783},[3863],{"type":53,"value":786},{"type":48,"tag":466,"props":3865,"children":3866},{"class":468,"line":779},[3867,3871,3875,3879,3884,3888,3892,3896,3900,3904],{"type":48,"tag":466,"props":3868,"children":3869},{"style":485},[3870],{"type":53,"value":804},{"type":48,"tag":466,"props":3872,"children":3873},{"style":479},[3874],{"type":53,"value":597},{"type":48,"tag":466,"props":3876,"children":3877},{"style":485},[3878],{"type":53,"value":813},{"type":48,"tag":466,"props":3880,"children":3881},{"style":479},[3882],{"type":53,"value":3883},"\u002Fverify_output.sh ",{"type":48,"tag":466,"props":3885,"children":3886},{"style":485},[3887],{"type":53,"value":804},{"type":48,"tag":466,"props":3889,"children":3890},{"style":479},[3891],{"type":53,"value":2929},{"type":48,"tag":466,"props":3893,"children":3894},{"style":485},[3895],{"type":53,"value":813},{"type":48,"tag":466,"props":3897,"children":3898},{"style":485},[3899],{"type":53,"value":828},{"type":48,"tag":466,"props":3901,"children":3902},{"style":479},[3903],{"type":53,"value":2964},{"type":48,"tag":466,"props":3905,"children":3906},{"style":485},[3907],{"type":53,"value":2598},{"type":48,"tag":1348,"props":3909,"children":3910},{},[],{"type":48,"tag":557,"props":3912,"children":3914},{"id":3913},"phase-4-eval-original",[3915,3917],{"type":53,"value":3916},"Phase 4 — eval ",{"type":48,"tag":62,"props":3918,"children":3920},{"className":3919},[],[3921],{"type":53,"value":203},{"type":48,"tag":56,"props":3923,"children":3924},{},[3925,3926,3932,3934,3940,3942,3947,3948,3953,3955,3960,3962,3967,3968,3973],{"type":53,"value":1002},{"type":48,"tag":62,"props":3927,"children":3929},{"className":3928},[],[3930],{"type":53,"value":3931},"references\u002Finference.md §Eval",{"type":53,"value":3933}," for score interpretation and feature-count\nexplanation. ",{"type":48,"tag":62,"props":3935,"children":3937},{"className":3936},[],[3938],{"type":53,"value":3939},"run_eval.sh",{"type":53,"value":3941}," writes ",{"type":48,"tag":62,"props":3943,"children":3945},{"className":3944},[],[3946],{"type":53,"value":243},{"type":53,"value":952},{"type":48,"tag":62,"props":3949,"children":3951},{"className":3950},[],[3952],{"type":53,"value":251},{"type":53,"value":3954}," inside\n",{"type":48,"tag":62,"props":3956,"children":3958},{"className":3957},[],[3959],{"type":53,"value":203},{"type":53,"value":3961}," and merges ",{"type":48,"tag":62,"props":3963,"children":3965},{"className":3964},[],[3966],{"type":53,"value":259},{"type":53,"value":261},{"type":48,"tag":62,"props":3969,"children":3971},{"className":3970},[],[3972],{"type":53,"value":267},{"type":53,"value":1579},{"type":48,"tag":455,"props":3975,"children":3977},{"className":457,"code":3976,"language":459,"meta":460,"style":460},"${ANOMALYGEN_SCRIPTS}\u002Frun_eval.sh \\\n    --real-path ${DATASET_DIR} --generated-path ${ORIGINAL} \\\n    --anomaly-types ${DEFECTS[@]}\n",[3978],{"type":48,"tag":62,"props":3979,"children":3980},{"__ignoreMap":460},[3981,4001,4042],{"type":48,"tag":466,"props":3982,"children":3983},{"class":468,"line":469},[3984,3988,3992,3996],{"type":48,"tag":466,"props":3985,"children":3986},{"style":485},[3987],{"type":53,"value":804},{"type":48,"tag":466,"props":3989,"children":3990},{"style":479},[3991],{"type":53,"value":597},{"type":48,"tag":466,"props":3993,"children":3994},{"style":485},[3995],{"type":53,"value":813},{"type":48,"tag":466,"props":3997,"children":3998},{"style":479},[3999],{"type":53,"value":4000},"\u002Frun_eval.sh \\\n",{"type":48,"tag":466,"props":4002,"children":4003},{"class":468,"line":631},[4004,4009,4013,4017,4021,4026,4030,4034,4038],{"type":48,"tag":466,"props":4005,"children":4006},{"style":496},[4007],{"type":53,"value":4008},"    --real-path",{"type":48,"tag":466,"props":4010,"children":4011},{"style":485},[4012],{"type":53,"value":828},{"type":48,"tag":466,"props":4014,"children":4015},{"style":479},[4016],{"type":53,"value":713},{"type":48,"tag":466,"props":4018,"children":4019},{"style":485},[4020],{"type":53,"value":813},{"type":48,"tag":466,"props":4022,"children":4023},{"style":502},[4024],{"type":53,"value":4025}," --generated-path",{"type":48,"tag":466,"props":4027,"children":4028},{"style":485},[4029],{"type":53,"value":828},{"type":48,"tag":466,"props":4031,"children":4032},{"style":479},[4033],{"type":53,"value":2964},{"type":48,"tag":466,"props":4035,"children":4036},{"style":485},[4037],{"type":53,"value":813},{"type":48,"tag":466,"props":4039,"children":4040},{"style":479},[4041],{"type":53,"value":3420},{"type":48,"tag":466,"props":4043,"children":4044},{"class":468,"line":671},[4045,4050,4054,4058],{"type":48,"tag":466,"props":4046,"children":4047},{"style":502},[4048],{"type":53,"value":4049},"    --anomaly-types",{"type":48,"tag":466,"props":4051,"children":4052},{"style":485},[4053],{"type":53,"value":828},{"type":48,"tag":466,"props":4055,"children":4056},{"style":479},[4057],{"type":53,"value":2544},{"type":48,"tag":466,"props":4059,"children":4060},{"style":485},[4061],{"type":53,"value":4062},"[@]}\n",{"type":48,"tag":1348,"props":4064,"children":4065},{},[],{"type":48,"tag":557,"props":4067,"children":4069},{"id":4068},"phase-5-per-sample-search-rounds",[4070],{"type":53,"value":4071},"Phase 5 — per-sample search rounds",{"type":48,"tag":56,"props":4073,"children":4074},{},[4075,4076,4082,4084,4090,4092,4098],{"type":53,"value":1002},{"type":48,"tag":62,"props":4077,"children":4079},{"className":4078},[],[4080],{"type":53,"value":4081},"references\u002Finference.md §Phase 5",{"type":53,"value":4083}," for draw strategy, ranges, and re-AMP\nguidance. For ",{"type":48,"tag":62,"props":4085,"children":4087},{"className":4086},[],[4088],{"type":53,"value":4089},"r",{"type":53,"value":4091}," in ",{"type":48,"tag":62,"props":4093,"children":4095},{"className":4094},[],[4096],{"type":53,"value":4097},"1..NUM_SEARCH_RUN",{"type":53,"value":4099},":",{"type":48,"tag":4101,"props":4102,"children":4103},"ol",{},[4104,4130,4150],{"type":48,"tag":2203,"props":4105,"children":4106},{},[4107,4109,4114,4116,4122,4123,4129],{"type":53,"value":4108},"Read prior round's ",{"type":48,"tag":62,"props":4110,"children":4112},{"className":4111},[],[4113],{"type":53,"value":243},{"type":53,"value":4115}," (or ",{"type":48,"tag":62,"props":4117,"children":4119},{"className":4118},[],[4120],{"type":53,"value":4121},"${ORIGINAL}\u002Fper_sample.csv",{"type":53,"value":1503},{"type":48,"tag":62,"props":4124,"children":4126},{"className":4125},[],[4127],{"type":53,"value":4128},"r=1",{"type":53,"value":2060},{"type":48,"tag":2203,"props":4131,"children":4132},{},[4133,4135,4141,4143,4148],{"type":53,"value":4134},"Write ",{"type":48,"tag":62,"props":4136,"children":4138},{"className":4137},[],[4139],{"type":53,"value":4140},"${ROUNDS}\u002Fround_${r}\u002Fdraws.json",{"type":53,"value":4142}," with selected ",{"type":48,"tag":62,"props":4144,"children":4146},{"className":4145},[],[4147],{"type":53,"value":300},{"type":53,"value":4149}," per sample.",{"type":48,"tag":2203,"props":4151,"children":4152},{},[4153,4155,4161,4163,4169,4171,4177],{"type":53,"value":4154},"Run round via ",{"type":48,"tag":62,"props":4156,"children":4158},{"className":4157},[],[4159],{"type":53,"value":4160},"${ANOMALYGEN_SCRIPTS}\u002Frun_round.sh",{"type":53,"value":4162}," (SDG + eval; the round dir\ngets its own ",{"type":48,"tag":62,"props":4164,"children":4166},{"className":4165},[],[4167],{"type":53,"value":4168},"sdg\u002F{SDG_result.csv, per_sample.csv, eval.log}",{"type":53,"value":4170},"). See\n",{"type":48,"tag":62,"props":4172,"children":4174},{"className":4173},[],[4175],{"type":53,"value":4176},"references\u002Finference-commands.md §Phase 5",{"type":53,"value":4178}," for the full command and flags.",{"type":48,"tag":56,"props":4180,"children":4181},{},[4182,4188,4190,4195,4196,4201],{"type":48,"tag":62,"props":4183,"children":4185},{"className":4184},[],[4186],{"type":53,"value":4187},"NUM_SEARCH_RUN=0",{"type":53,"value":4189}," is valid — skip this phase entirely and let Phase 6\nclone ",{"type":48,"tag":62,"props":4191,"children":4193},{"className":4192},[],[4194],{"type":53,"value":203},{"type":53,"value":261},{"type":48,"tag":62,"props":4197,"children":4199},{"className":4198},[],[4200],{"type":53,"value":343},{"type":53,"value":1579},{"type":48,"tag":1348,"props":4203,"children":4204},{},[],{"type":48,"tag":557,"props":4206,"children":4208},{"id":4207},"phase-6-assemble-searched-stitch-only",[4209,4211,4216],{"type":53,"value":4210},"Phase 6 — assemble ",{"type":48,"tag":62,"props":4212,"children":4214},{"className":4213},[],[4215],{"type":53,"value":343},{"type":53,"value":4217}," (stitch only)",{"type":48,"tag":56,"props":4219,"children":4220},{},[4221,4223,4228,4230,4235,4237,4242,4244,4249,4251,4256,4258,4263,4264,4270,4272,4278],{"type":53,"value":4222},"Always run assemble (works with 0 rounds — ",{"type":48,"tag":62,"props":4224,"children":4226},{"className":4225},[],[4227],{"type":53,"value":343},{"type":53,"value":4229}," clones ",{"type":48,"tag":62,"props":4231,"children":4233},{"className":4232},[],[4234],{"type":53,"value":203},{"type":53,"value":4236},", so\ndownstream always reads ",{"type":48,"tag":62,"props":4238,"children":4240},{"className":4239},[],[4241],{"type":53,"value":343},{"type":53,"value":4243}," regardless of ",{"type":48,"tag":62,"props":4245,"children":4247},{"className":4246},[],[4248],{"type":53,"value":1798},{"type":53,"value":4250},"). Stitch-only:\ncopies winning images per sample-index into ",{"type":48,"tag":62,"props":4252,"children":4254},{"className":4253},[],[4255],{"type":53,"value":343},{"type":53,"value":4257}," and carries over\nper-sample ",{"type":48,"tag":62,"props":4259,"children":4261},{"className":4260},[],[4262],{"type":53,"value":259},{"type":53,"value":1653},{"type":48,"tag":62,"props":4265,"children":4267},{"className":4266},[],[4268],{"type":53,"value":4269},"mnn_score",{"type":53,"value":4271}," from each pick's source-round per_sample.csv.\nNo eval — Phase 7 emits the canonical ",{"type":48,"tag":62,"props":4273,"children":4275},{"className":4274},[],[4276],{"type":53,"value":4277},"searched\u002Feval.log",{"type":53,"value":1579},{"type":48,"tag":455,"props":4280,"children":4282},{"className":457,"code":4281,"language":459,"meta":460,"style":460},"mkdir -p ${ROUNDS}\npython3 -m scripts.utilities.assemble_searched \\\n    --original-dir ${ORIGINAL} --original-csv ${ORIGINAL}\u002Fper_sample.csv \\\n    --rounds-dir ${ROUNDS} --searched-dir ${SEARCHED}\n",[4283],{"type":48,"tag":62,"props":4284,"children":4285},{"__ignoreMap":460},[4286,4311,4331,4377],{"type":48,"tag":466,"props":4287,"children":4288},{"class":468,"line":469},[4289,4294,4299,4303,4307],{"type":48,"tag":466,"props":4290,"children":4291},{"style":496},[4292],{"type":53,"value":4293},"mkdir",{"type":48,"tag":466,"props":4295,"children":4296},{"style":502},[4297],{"type":53,"value":4298}," -p",{"type":48,"tag":466,"props":4300,"children":4301},{"style":485},[4302],{"type":53,"value":828},{"type":48,"tag":466,"props":4304,"children":4305},{"style":479},[4306],{"type":53,"value":3016},{"type":48,"tag":466,"props":4308,"children":4309},{"style":485},[4310],{"type":53,"value":2598},{"type":48,"tag":466,"props":4312,"children":4313},{"class":468,"line":631},[4314,4318,4322,4327],{"type":48,"tag":466,"props":4315,"children":4316},{"style":496},[4317],{"type":53,"value":3632},{"type":48,"tag":466,"props":4319,"children":4320},{"style":502},[4321],{"type":53,"value":3637},{"type":48,"tag":466,"props":4323,"children":4324},{"style":502},[4325],{"type":53,"value":4326}," scripts.utilities.assemble_searched",{"type":48,"tag":466,"props":4328,"children":4329},{"style":479},[4330],{"type":53,"value":3420},{"type":48,"tag":466,"props":4332,"children":4333},{"class":468,"line":671},[4334,4339,4343,4347,4351,4356,4360,4364,4368,4373],{"type":48,"tag":466,"props":4335,"children":4336},{"style":502},[4337],{"type":53,"value":4338},"    --original-dir",{"type":48,"tag":466,"props":4340,"children":4341},{"style":485},[4342],{"type":53,"value":828},{"type":48,"tag":466,"props":4344,"children":4345},{"style":479},[4346],{"type":53,"value":2964},{"type":48,"tag":466,"props":4348,"children":4349},{"style":485},[4350],{"type":53,"value":813},{"type":48,"tag":466,"props":4352,"children":4353},{"style":502},[4354],{"type":53,"value":4355}," --original-csv",{"type":48,"tag":466,"props":4357,"children":4358},{"style":485},[4359],{"type":53,"value":828},{"type":48,"tag":466,"props":4361,"children":4362},{"style":479},[4363],{"type":53,"value":2964},{"type":48,"tag":466,"props":4365,"children":4366},{"style":485},[4367],{"type":53,"value":813},{"type":48,"tag":466,"props":4369,"children":4370},{"style":502},[4371],{"type":53,"value":4372},"\u002Fper_sample.csv",{"type":48,"tag":466,"props":4374,"children":4375},{"style":479},[4376],{"type":53,"value":3420},{"type":48,"tag":466,"props":4378,"children":4379},{"class":468,"line":689},[4380,4385,4389,4393,4397,4402,4406,4410],{"type":48,"tag":466,"props":4381,"children":4382},{"style":502},[4383],{"type":53,"value":4384},"    --rounds-dir",{"type":48,"tag":466,"props":4386,"children":4387},{"style":485},[4388],{"type":53,"value":828},{"type":48,"tag":466,"props":4390,"children":4391},{"style":479},[4392],{"type":53,"value":3016},{"type":48,"tag":466,"props":4394,"children":4395},{"style":485},[4396],{"type":53,"value":813},{"type":48,"tag":466,"props":4398,"children":4399},{"style":502},[4400],{"type":53,"value":4401}," --searched-dir",{"type":48,"tag":466,"props":4403,"children":4404},{"style":485},[4405],{"type":53,"value":828},{"type":48,"tag":466,"props":4407,"children":4408},{"style":479},[4409],{"type":53,"value":2990},{"type":48,"tag":466,"props":4411,"children":4412},{"style":485},[4413],{"type":53,"value":2598},{"type":48,"tag":1348,"props":4415,"children":4416},{},[],{"type":48,"tag":557,"props":4418,"children":4420},{"id":4419},"phase-7-filter-regen-eval-default-nn_threshold04",[4421,4423,4429],{"type":53,"value":4422},"Phase 7 — filter + regen + eval (default ",{"type":48,"tag":62,"props":4424,"children":4426},{"className":4425},[],[4427],{"type":53,"value":4428},"nn_threshold=0.4",{"type":53,"value":510},{"type":48,"tag":56,"props":4431,"children":4432},{},[4433,4435,4440,4442,4447,4449,4454,4456,4461,4463,4469,4471,4476,4477,4482,4484,4490,4492,4497],{"type":53,"value":4434},"Phase 7 ",{"type":48,"tag":431,"props":4436,"children":4437},{},[4438],{"type":53,"value":4439},"runs by default",{"type":53,"value":4441}," (",{"type":48,"tag":62,"props":4443,"children":4445},{"className":4444},[],[4446],{"type":53,"value":4428},{"type":53,"value":4448},") on every ",{"type":48,"tag":62,"props":4450,"children":4452},{"className":4451},[],[4453],{"type":53,"value":610},{"type":53,"value":4455}," and\n",{"type":48,"tag":62,"props":4457,"children":4459},{"className":4458},[],[4460],{"type":53,"value":869},{"type":53,"value":4462}," invocation; pass ",{"type":48,"tag":62,"props":4464,"children":4466},{"className":4465},[],[4467],{"type":53,"value":4468},"nn_threshold=0",{"type":53,"value":4470}," to skip it. It filters\n",{"type":48,"tag":62,"props":4472,"children":4474},{"className":4473},[],[4475],{"type":53,"value":343},{"type":53,"value":385},{"type":48,"tag":62,"props":4478,"children":4480},{"className":4479},[],[4481],{"type":53,"value":391},{"type":53,"value":4483},", regenerates dropped samples via re-AMP (fresh\n",{"type":48,"tag":62,"props":4485,"children":4487},{"className":4486},[],[4488],{"type":53,"value":4489},"(clean, submask)",{"type":53,"value":4491}," pairing in the same defect type) for up to 5 attempts, then\nfalls back to best-scoring non-passing regens and finally to dropped originals,\nso the final bucket always equals ",{"type":48,"tag":62,"props":4493,"children":4495},{"className":4494},[],[4496],{"type":53,"value":1365},{"type":53,"value":1579},{"type":48,"tag":56,"props":4499,"children":4500},{},[4501,4503,4509,4511,4516,4518,4523,4525,4531,4533,4539,4541,4547],{"type":53,"value":4502},"Run ",{"type":48,"tag":62,"props":4504,"children":4506},{"className":4505},[],[4507],{"type":53,"value":4508},"python3 -m scripts.utilities.filter_with_regen",{"type":53,"value":4510},". It runs the final\n",{"type":48,"tag":62,"props":4512,"children":4514},{"className":4513},[],[4515],{"type":53,"value":3939},{"type":53,"value":4517}," internally — the only eval against ",{"type":48,"tag":62,"props":4519,"children":4521},{"className":4520},[],[4522],{"type":53,"value":343},{"type":53,"value":4524},". Read\n",{"type":48,"tag":62,"props":4526,"children":4528},{"className":4527},[],[4529],{"type":53,"value":4530},"references\u002Finference.md §Phase 7",{"type":53,"value":4532}," for regen mechanics, source-column tracing,\nand the ",{"type":48,"tag":62,"props":4534,"children":4536},{"className":4535},[],[4537],{"type":53,"value":4538},"regens\u002Fregen_summary.csv",{"type":53,"value":4540}," schema; see\n",{"type":48,"tag":62,"props":4542,"children":4544},{"className":4543},[],[4545],{"type":53,"value":4546},"references\u002Finference-commands.md §Phase 7",{"type":53,"value":4178},{"type":48,"tag":1348,"props":4549,"children":4550},{},[],{"type":48,"tag":557,"props":4552,"children":4554},{"id":4553},"output-layout",[4555],{"type":53,"value":4556},"Output layout",{"type":48,"tag":56,"props":4558,"children":4559},{},[4560,4562,4567,4569,4574,4576,4581,4583,4588,4590,4595,4596,4601,4603,4608,4610,4615,4617,4623,4625,4631,4633,4639],{"type":53,"value":4561},"Every bucket that gets eval'd carries the same triad of files:\n",{"type":48,"tag":62,"props":4563,"children":4565},{"className":4564},[],[4566],{"type":53,"value":267},{"type":53,"value":4568}," (generation params + ",{"type":48,"tag":62,"props":4570,"children":4572},{"className":4571},[],[4573],{"type":53,"value":259},{"type":53,"value":4575},"), ",{"type":48,"tag":62,"props":4577,"children":4579},{"className":4578},[],[4580],{"type":53,"value":243},{"type":53,"value":4582},"\n(per-sample nn + mnn), and ",{"type":48,"tag":62,"props":4584,"children":4586},{"className":4585},[],[4587],{"type":53,"value":251},{"type":53,"value":4589}," (aggregate FID \u002F per-defect avg).\nBuckets live under ",{"type":48,"tag":62,"props":4591,"children":4593},{"className":4592},[],[4594],{"type":53,"value":1321},{"type":53,"value":1536},{"type":48,"tag":62,"props":4597,"children":4599},{"className":4598},[],[4600],{"type":53,"value":203},{"type":53,"value":4602}," (Phase 3+4), ",{"type":48,"tag":62,"props":4604,"children":4606},{"className":4605},[],[4607],{"type":53,"value":343},{"type":53,"value":4609},"\n(Phase 6 stitch + Phase 7 filter+regen+eval), ",{"type":48,"tag":62,"props":4611,"children":4613},{"className":4612},[],[4614],{"type":53,"value":308},{"type":53,"value":4616}," (Phase 5,\nplus ",{"type":48,"tag":62,"props":4618,"children":4620},{"className":4619},[],[4621],{"type":53,"value":4622},"search_summary.csv",{"type":53,"value":4624},"), and ",{"type":48,"tag":62,"props":4626,"children":4628},{"className":4627},[],[4629],{"type":53,"value":4630},"regens\u002Fregen_NN\u002F",{"type":53,"value":4632}," (Phase 7, plus\n",{"type":48,"tag":62,"props":4634,"children":4636},{"className":4635},[],[4637],{"type":53,"value":4638},"regen_summary.csv",{"type":53,"value":2060},{"type":48,"tag":56,"props":4641,"children":4642},{},[4643,4645,4650,4652,4657,4659,4664,4665,4670,4672,4677,4678,4683,4684,4690,4692,4697],{"type":53,"value":4644},"See ",{"type":48,"tag":62,"props":4646,"children":4648},{"className":4647},[],[4649],{"type":53,"value":1310},{"type":53,"value":4651}," for the full directory tree with per-file\nannotations and the post-run ",{"type":48,"tag":431,"props":4653,"children":4654},{},[4655],{"type":53,"value":4656},"Verification",{"type":53,"value":4658}," checklist (image counts per\nbucket, ",{"type":48,"tag":62,"props":4660,"children":4662},{"className":4661},[],[4663],{"type":53,"value":4622},{"type":53,"value":1653},{"type":48,"tag":62,"props":4666,"children":4668},{"className":4667},[],[4669],{"type":53,"value":4638},{"type":53,"value":4671}," row checks, and the per-type\n",{"type":48,"tag":62,"props":4673,"children":4675},{"className":4674},[],[4676],{"type":53,"value":259},{"type":53,"value":1653},{"type":48,"tag":62,"props":4679,"children":4681},{"className":4680},[],[4682],{"type":53,"value":4269},{"type":53,"value":1653},{"type":48,"tag":62,"props":4685,"children":4687},{"className":4686},[],[4688],{"type":53,"value":4689},"fid",{"type":53,"value":4691}," fields in each ",{"type":48,"tag":62,"props":4693,"children":4695},{"className":4694},[],[4696],{"type":53,"value":251},{"type":53,"value":2060},{"type":48,"tag":557,"props":4699,"children":4701},{"id":4700},"error-handling",[4702],{"type":53,"value":4703},"Error handling",{"type":48,"tag":56,"props":4705,"children":4706},{},[4707,4709,4715,4717,4722,4724,4729,4731,4736,4737,4742],{"type":53,"value":4708},"Common pipeline failure modes (missing mask dirs, short\u002Fempty AMP output and\nthe ",{"type":48,"tag":62,"props":4710,"children":4712},{"className":4711},[],[4713],{"type":53,"value":4714},"0 entries written",{"type":53,"value":4716}," halt, mid-round SDG failure resume, off-boundary\n",{"type":48,"tag":62,"props":4718,"children":4720},{"className":4719},[],[4721],{"type":53,"value":1346},{"type":53,"value":4723},") are covered in ",{"type":48,"tag":62,"props":4725,"children":4727},{"className":4726},[],[4728],{"type":53,"value":1335},{"type":53,"value":4730},"; see also\n",{"type":48,"tag":62,"props":4732,"children":4734},{"className":4733},[],[4735],{"type":53,"value":1011},{"type":53,"value":952},{"type":48,"tag":62,"props":4738,"children":4740},{"className":4739},[],[4741],{"type":53,"value":1022},{"type":53,"value":4743}," for phase-specific\nerror handling.",{"type":48,"tag":4745,"props":4746,"children":4747},"style",{},[4748],{"type":53,"value":4749},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":4751,"total":4908},[4752,4770,4788,4799,4811,4825,4838,4850,4863,4874,4888,4897],{"slug":4753,"name":4753,"fn":4754,"description":4755,"org":4756,"tags":4757,"stars":4767,"repoUrl":4768,"updatedAt":4769},"nemoclaw-user-guide","retrieve NemoClaw documentation and configuration","Guides human users' AI agents to the NemoClaw docs MCP server and canonical Fern documentation in Markdown form. Use when users ask how to install, configure, operate, troubleshoot, secure, or learn NemoClaw with an AI coding assistant. Trigger keywords - nemoclaw docs, use nemoclaw with ai agent, nemoclaw mcp docs, nemoclaw install help, nemoclaw quickstart, nemoclaw markdown docs, llms.txt, agent skills.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4758,4761,4764],{"name":4759,"slug":4760,"type":15},"Documentation","documentation",{"name":4762,"slug":4763,"type":15},"MCP","mcp",{"name":4765,"slug":4766,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":4771,"name":4771,"fn":4772,"description":4773,"org":4774,"tags":4775,"stars":4785,"repoUrl":4786,"updatedAt":4787},"mcore-build-and-dependency","manage Megatron-LM development environments","Container-based dev environment setup and dependency management for Megatron-LM. Covers acquiring and launching the CI container, uv package management, and updating uv.lock.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4776,4779,4782],{"name":4777,"slug":4778,"type":15},"Containers","containers",{"name":4780,"slug":4781,"type":15},"Deployment","deployment",{"name":4783,"slug":4784,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":4789,"name":4789,"fn":4790,"description":4791,"org":4792,"tags":4793,"stars":4785,"repoUrl":4786,"updatedAt":4798},"mcore-bump-base-image","update NVIDIA PyTorch base images","Bump the NVIDIA PyTorch base image (`nvcr.io\u002Fnvidia\u002Fpytorch:YY.MM-py3`) used by Megatron-LM CI. Covers the two pin sites (GitHub CI in `docker\u002F.ngc_version.dev` and GitLab CI in `.gitlab\u002Fstages\u002F01.build.yml`), the post-bump CI loop (re-run functional tests, refresh golden values, mark broken tests), and the gotchas that bit PRs",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4794,4797],{"name":4795,"slug":4796,"type":15},"CI\u002FCD","ci-cd",{"name":4780,"slug":4781,"type":15},"2026-07-14T05:25:59.97109",{"slug":4800,"name":4800,"fn":4801,"description":4802,"org":4803,"tags":4804,"stars":4785,"repoUrl":4786,"updatedAt":4810},"mcore-cicd","manage CI\u002FCD pipelines for Megatron-LM","CI\u002FCD reference for Megatron-LM. Covers CI pipeline structure, PR scope labels, triggering internal GitLab CI (which force-pushes the current branch to a pull-request\u002FBRANCH ref — always dry-run and verify the destination first; never run against shared or protected branches), and CI failure investigation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4805,4806,4807],{"name":4795,"slug":4796,"type":15},{"name":4780,"slug":4781,"type":15},{"name":4808,"slug":4809,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":4812,"name":4812,"fn":4813,"description":4814,"org":4815,"tags":4816,"stars":4785,"repoUrl":4786,"updatedAt":4824},"mcore-create-issue","investigate CI failures and create issues","Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4817,4820,4821],{"name":4818,"slug":4819,"type":15},"Debugging","debugging",{"name":4808,"slug":4809,"type":15},{"name":4822,"slug":4823,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":4826,"name":4826,"fn":4827,"description":4828,"org":4829,"tags":4830,"stars":4785,"repoUrl":4786,"updatedAt":4837},"mcore-linting-and-formatting","lint and format Megatron-LM code","Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4831,4834],{"name":4832,"slug":4833,"type":15},"Best Practices","best-practices",{"name":4835,"slug":4836,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":4839,"name":4839,"fn":4840,"description":4841,"org":4842,"tags":4843,"stars":4785,"repoUrl":4786,"updatedAt":4849},"mcore-migrate-gpt-to-hybrid","migrate Megatron-LM models to HybridModel","Migration guide for moving Megatron Core GPTModel checkpoints, model providers, training commands, and layer mappings to HybridModel.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4844,4845,4848],{"name":13,"slug":14,"type":15},{"name":4846,"slug":4847,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":4851,"name":4851,"fn":4852,"description":4853,"org":4854,"tags":4855,"stars":4785,"repoUrl":4786,"updatedAt":4862},"mcore-onboard-gb200-1node-tests","onboard functional tests for GB200","Onboard 1-node GitHub MR functional tests for GB200 from existing mr-scoped 2-node tests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4856,4859],{"name":4857,"slug":4858,"type":15},"QA","qa",{"name":4860,"slug":4861,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":4864,"name":4864,"fn":4865,"description":4866,"org":4867,"tags":4868,"stars":4785,"repoUrl":4786,"updatedAt":4873},"mcore-run-on-slurm","launch distributed training jobs on SLURM","How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4869,4870],{"name":4780,"slug":4781,"type":15},{"name":4871,"slug":4872,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":4875,"name":4875,"fn":4876,"description":4877,"org":4878,"tags":4879,"stars":4785,"repoUrl":4786,"updatedAt":4887},"mcore-split-pr","split pull requests to reduce review load","Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4880,4883,4884],{"name":4881,"slug":4882,"type":15},"Code Review","code-review",{"name":4808,"slug":4809,"type":15},{"name":4885,"slug":4886,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":4889,"name":4889,"fn":4890,"description":4891,"org":4892,"tags":4893,"stars":4785,"repoUrl":4786,"updatedAt":4896},"mcore-testing","run and manage Megatron-LM tests","Test system for Megatron-LM. Covers test layout, recipe YAML structure, adding and running unit and functional tests, golden values, marker filters, and CI parity.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4894,4895],{"name":4857,"slug":4858,"type":15},{"name":4860,"slug":4861,"type":15},"2026-07-14T05:25:54.928983",{"slug":4898,"name":4898,"fn":4899,"description":4900,"org":4901,"tags":4902,"stars":4785,"repoUrl":4786,"updatedAt":4907},"nightly-sync","manage nightly main-to-dev sync workflows","Domain knowledge for the nightly main-to-dev sync workflow. Covers merge strategy, CI architecture, failure investigation, and known issues.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4903,4906],{"name":4904,"slug":4905,"type":15},"Automation","automation",{"name":4795,"slug":4796,"type":15},"2026-07-30T05:29:03.275638",525,{"items":4910,"total":5006},[4911,4928,4938,4952,4962,4977,4992],{"slug":4912,"name":4912,"fn":4913,"description":4914,"org":4915,"tags":4916,"stars":23,"repoUrl":24,"updatedAt":4927},"accelerated-computing-cudf","accelerate data processing with cuDF","Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV\u002FParquet I\u002FO, nullable semantics, and multi-GPU DataFrame workloads.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4917,4920,4923,4924],{"name":4918,"slug":4919,"type":15},"Data Analysis","data-analysis",{"name":4921,"slug":4922,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":4925,"slug":4926,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":4929,"name":4929,"fn":4930,"description":4931,"org":4932,"tags":4933,"stars":23,"repoUrl":24,"updatedAt":4937},"aiq-deploy","deploy and manage NVIDIA AI-Q infrastructure","Use when asked to install, deploy, run, validate, troubleshoot, or stop NVIDIA AI-Q Blueprint infrastructure.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4934,4935,4936],{"name":4780,"slug":4781,"type":15},{"name":4871,"slug":4872,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":4939,"name":4939,"fn":4940,"description":4941,"org":4942,"tags":4943,"stars":23,"repoUrl":24,"updatedAt":4951},"aiq-research","conduct deep research with AI-Q","Use when asked to run deep research or AI-Q research through a reachable NVIDIA AI-Q Blueprint backend.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4944,4947,4948],{"name":4945,"slug":4946,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":4949,"slug":4950,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":4953,"name":4953,"fn":4954,"description":4955,"org":4956,"tags":4957,"stars":23,"repoUrl":24,"updatedAt":4961},"amc-run-sample-calibration","run AMC sample dataset calibration","Run end-to-end calibration on the shipped sample dataset (sdg_08_2_sample_data_010926.zip) against a running AMC microservice. Use when user says 'test sample dataset', 'run sample calibration', 'verify AMC install', or 'launch and test'.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4958,4959,4960],{"name":4918,"slug":4919,"type":15},{"name":9,"slug":8,"type":15},{"name":4860,"slug":4861,"type":15},"2026-07-17T05:29:03.913266",{"slug":4963,"name":4963,"fn":4964,"description":4965,"org":4966,"tags":4967,"stars":23,"repoUrl":24,"updatedAt":4976},"amc-run-video-calibration","calibrate video datasets with AutoMagicCalib","Calibrate a new dataset from pre-recorded video files via the AutoMagicCalib REST API. Use when user has local MP4s and says 'calibrate my videos', 'run AMC on these videos', or similar. For RTSP\u002Flive streams, use amc-run-rtsp-calibration instead.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4968,4969,4972,4973],{"name":4904,"slug":4905,"type":15},{"name":4970,"slug":4971,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":4974,"slug":4975,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":4978,"name":4978,"fn":4979,"description":4980,"org":4981,"tags":4982,"stars":23,"repoUrl":24,"updatedAt":4991},"amc-setup-calibration-stack","deploy AutoMagicCalib microservice with Docker","Launch AutoMagicCalib microservice and web UI from NGC release images via Docker Compose. Use when user says 'deploy auto calibration', 'launch auto calibration', 'launch AMC', 'start MS+UI', or 'set up auto-magic-calib'. Requires NGC API key.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4983,4984,4987,4988],{"name":4780,"slug":4781,"type":15},{"name":4985,"slug":4986,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":4989,"slug":4990,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":4993,"name":4993,"fn":4994,"description":4995,"org":4996,"tags":4997,"stars":23,"repoUrl":24,"updatedAt":5005},"cudaq-guide","develop quantum applications with CUDA-Q","CUDA-Q onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[4998,4999,5002],{"name":9,"slug":8,"type":15},{"name":5000,"slug":5001,"type":15},"Quantum Computing","quantum-computing",{"name":5003,"slug":5004,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",310]