data-formats
Validates OFA TSV rows, selected-column layouts, base64 image payloads, code sequences, and speech manifests before GPU runs.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- data-formats
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Validates OFA TSV rows, selected-column layouts, base64 image payloads, code sequences, and speech manifests before GPU runs.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/data-formats.md3.1 KB
- 参考文档references/troubleshooting.md1.5 KB
- 脚本scripts/encode_image_base64.py1.6 KB
- 脚本scripts/validate_ofa_tsv.py6.8 KB
03 SKILL.md 正文
data-formats
Use this sub-skill when a user needs to prepare, inspect, or validate OFA input files before launching a task.
Trigger phrases
- "What columns does this TSV need?"
- "Why does the data loader crash on this row?"
- "How do I base64-encode OFA images?"
- "Is my VQA / RefCOCO / OCR / MMSpeech manifest correct?"
- "Can you validate the pretraining workspace first?"
What this sub-skill owns
- OFA TSV row layouts and selected-column rules,
- base64 image payload checks,
- image-code sequence validation,
- answer-label and reference-field sanity checks,
- manifest/file-path validation for speech workflows,
- safe preflight checks for pretraining data bundles.
What it excludes
- the training/evaluation command itself -> the task sub-skill,
- model internals and registration ->
model-internals-and-extension, - generic image manipulation unrelated to OFA TSVs -> other vision skills.
Read these files
- references/data-formats.md for row layouts and validation cues.
- references/troubleshooting.md for common row-shape failures.
- scripts/encode_image_base64.py to convert a local image into a TSV-safe payload.
- scripts/validate_ofa_tsv.py for general TSV checks.
Typical workflow
- Identify the task family.
- Confirm the expected row width and selected columns.
- Validate base64, image, code, JSON, or path cells.
- Only then hand the file to the workflow sub-skill for a training or evaluation command.
Notes
- Many OFA workflows use a tab-separated flat file instead of a dataset library format.
- The same base64 image can appear in caption, VQA, RefCOCO, OCR, and image-generation layouts.
- Selected columns matter as much as the row content; a row can look valid but still feed the wrong task.
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).