evaluation-and-metrics
Guides MMGeneration evaluation commands, metric selection, and inception-stat preparation workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- evaluation-and-metrics
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides MMGeneration evaluation commands, metric selection, and inception-stat preparation workflows.
- disable-model-invocation
- true
02 资源文件 2
- 参考文档references/troubleshooting.md2.8 KB
- 参考文档references/workflows.md2.6 KB
03 SKILL.md 正文
Evaluation and Metrics
Use this sub-skill when the user wants to score generated images, precompute real-image statistics, or understand the repo's evaluation scripts and metric classes.
Typical triggers
- "How do I evaluate this checkpoint?"
- "How do I compute FID or IS?"
- "How do I precompute inception stats?"
- "How do I run online vs offline evaluation?"
- "Why does distributed evaluation reject this metric?"
Include here
tools/evaluation.pytools/dist_eval.sh,tools/slurm_eval.sh,tools/slurm_eval_multi_gpu.shtools/utils/inception_stat.pytools/utils/translation_eval.pymmgen.core.evaluationbuilders, hooks, and metrics- FID, IS, PPL, PR, SWD, MS-SSIM, and GaussianKLD usage
- online/offline evaluation mode differences
Exclude here
- Training launchers and resume behavior ->
training-and-distribution - Image sampling demos ->
inference-and-sampling - Config registry editing ->
configuration-and-extension - Latent editing and packaging ->
applications-and-deployment
Read these files first
references/workflows.mdreferences/troubleshooting.md../../references/api-reference.md../../references/cli-reference.md../../references/model-overview.md../../references/data-formats.md
What good guidance looks like
A future agent should be able to:
- Pick the right metric and command family for a GAN or translation model.
- Decide when online evaluation is faster than disk-backed evaluation.
- Explain what cached inception statistics are for and when they are needed.
- Recognize which metrics can run in the distributed path.
- Understand when a metric helper returns a tensor, a dict, or a scalar.
Common failure modes
- A distributed run asks for a metric the distributed path does not support.
- Inception-stat extraction needs a different runtime or cached model file.
- Translation evaluation is pointed at the wrong target domain.
- The helper is told to sample images only, but the caller expects a metric score.
- A metric config uses a missing cached Inception/VGG asset.
For concrete recovery steps, read references/troubleshooting.md.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).