r1v4-api-testing
Draft, inspect, parse, and summarize Skywork R1V4 API batch tests without leaking API keys.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- r1v4-api-testing
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Draft, inspect, parse, and summarize Skywork R1V4 API batch tests without leaking API keys.
- disable-model-invocation
- true
02 资源文件 8
- 参考文档references/api-batch-workflows.md3.1 KB
- 参考文档references/data-formats.md3.2 KB
- 参考文档references/result-analysis.md3.0 KB
- 参考文档references/troubleshooting.md3.9 KB
- 脚本scripts/build_api_payload.py5.6 KB
- 脚本scripts/parse_r1v4_response.py12.0 KB
- 脚本scripts/summarize_results.py5.0 KB
- 脚本scripts/validate_cases.py5.9 KB
03 SKILL.md 正文
R1V4 API Testing
Use this sub-skill when you need to prepare Skywork R1V4 API batch tests, dry-run request payloads, parse tagged tool-use responses, summarize result JSONL, or troubleshoot the batch viewer safely.
Start here
- Read
references/api-batch-workflows.mdfor the four batch modes and Chinese terminology. - Check
references/data-formats.mdfortest_cases.jsonl, request payload, and result JSONL schemas. - Validate input cases with
scripts/validate_cases.pybefore any API run. - Build a no-network payload preview with
scripts/build_api_payload.py. - Parse a tagged response or a result file with
scripts/parse_r1v4_response.py. - Summarize a result JSONL with
scripts/summarize_results.py.
Bundled files
references/api-batch-workflows.md— workflow map forbatch_nonstream.py,batch_stream.py,batch_planner_nonstream.py, andbatch_planner_stream.py.references/data-formats.md— input/output schemas and tagged response grammar.references/result-analysis.md— safe headless analysis path and visualizer guidance.references/troubleshooting.md— API, image, stream, parse, and viewer failure modes.scripts/validate_cases.py— schema and optional image-path validation.scripts/build_api_payload.py— OpenAI-compatible payload builder with data URL support.scripts/parse_r1v4_response.py— tagged response parser and JSONL parser CLI.scripts/summarize_results.py— batch result counters and response-length summary.
Routing notes
- Use
skywork/r1v4-litefor the regular batch modes andskywork/r1v4-vl-planner-litefor planner batch modes. - Keep the message order as image first, then text.
- Bundled helpers do not make live API calls by default.
- Leave API keys to env or caller config; do not hard-code them in payloads or scripts.
- Use the bundled summary helper instead of copying the full Flask viewer into a workflow.
- Route local Transformers/vLLM inference, CUDA setup, and model-weight issues to the sibling
local-inferencesub-skill. - Route VLMEvalKit, EMMA, MMK12, and benchmark reproduction tasks to the sibling
evaluation-reproductionsub-skill.
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).