training
Launch, resume, and troubleshoot OpenFlamingo training runs.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- training
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Launch, resume, and troubleshoot OpenFlamingo training runs.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/cli-reference.md6.5 KB
- 参考文档references/data-formats.md3.2 KB
- 参考文档references/training-workflows.md4.3 KB
- 参考文档references/troubleshooting.md4.0 KB
- 脚本scripts/build_train_command.py11.5 KB
- 脚本scripts/run_training_entrypoint.py1.5 KB
03 SKILL.md 正文
training
Use this sub-skill for OpenFlamingo training and fine-tuning tasks: configuring model and data flags, generating launch commands, selecting DDP or FSDP, resuming checkpoints, and diagnosing runtime failures.
Read first
What this sub-skill covers
- Vision encoder, language model, tokenizer, and cross-attention setup.
- LAION and MMC4 WebDataset inputs.
- DDP, Slurm-style launches, Horovod, and FSDP caveats.
- Precision, gradient checkpointing, loss weighting, and checkpointing.
- W&B logging, offline mode, and restart behavior.
Typical flow
- Confirm the LAION and MMC4 shard layouts and sample counts.
- Pick DDP or FSDP and decide whether to use
--dataset_resampled. - Generate a launch command with
scripts/build_train_command.py. - Run the printed command from the repository root or another shell environment that can resolve the training entrypoint path.
- Monitor
run_name/checkpoint_<epoch>.ptand resume with--resume_from_checkpointwhen needed.
Guardrails
- Do not launch full training from this skill without explicit user intent and the required data and model weights.
- If the language model uses tied input/output embeddings, prefer DDP or freeze LM embeddings when using FSDP.
- Keep LAION and MMC4 batch counts aligned; the training loop asserts equal batches per epoch.
- If the shard metadata is missing, supply explicit sample counts for both datasets.
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).