multimodal-i-o
Use GenAI Processors audio, video, PDF, web, file, and speech connectors.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- multimodal-i-o
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use GenAI Processors audio, video, PDF, web, file, and speech connectors.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md3.4 KB
- 参考文档references/troubleshooting.md2.8 KB
- 参考文档references/workflows.md2.8 KB
- 脚本scripts/smoke_io.py1.3 KB
03 SKILL.md 正文
Multimodal I/O
Use this sub-skill when the task is about getting content into or out of a processor pipeline: microphones, speakers, cameras, screens, PDFs, URLs, GitHub, Google Drive, files, speech recognition, speech synthesis, VAD, and realtime media flow.
Read when
- The task names
PyAudioIn,PyAudioOut,VideoIn,AudioToWav,Vad,SpeechToText,TextToSpeech,PDFExtract,UrlFetch,GithubProcessor,Drive,RateLimitAudio,EventDetection,add_timestamps,to_timestamp, orWindow. - The task is about MIME types, audio sample rates, video frames, PDFs, URLs, files, or Google Workspace documents.
- The task needs a realtime audio/video pipeline before or after a model wrapper.
- The task needs optional dependency or device troubleshooting.
Boundaries
This sub-skill owns connectors and media/document processing. It does not own
model selection or function-calling strategy; use ../model-backends/ for
that. It does not own complete app/demo layout; use ../examples-and-apps/ for
full CLIs and AI Studio applets.
Workflow checklist
- Identify source modality: terminal text, audio, camera/screen frames, file glob, PDF bytes, URL, GitHub URL, Drive doc/sheet/slide, or tool output.
- Confirm MIME type and substream expectations before composing with model processors.
- Verify optional imports with
scripts/smoke_io.pybefore using devices or credentials. - For audio/video, check device permissions and prefer headphones unless using browser echo cancellation.
- For Google Speech/TTS or Drive, validate credentials and project/service setup before running full pipelines.
- For URLs/GitHub/PDFs, separate safe parser/import smoke from network or file reads.
- When combining with live/realtime models, cross-check trigger and substream
behavior in
../model-backends/.
References and scripts
references/api-reference.mdlists connector classes and processor roles.references/workflows.mdprovides audio, video, PDF, URL, Drive/GitHub, and realtime media patterns.references/troubleshooting.mdcovers PortAudio, device permissions, Google Cloud auth, PDF rendering, OpenCV/AV, and web fetch failures.scripts/smoke_io.pyimports optional I/O modules without opening devices, making network requests, or requiring credentials.
Native evidence anchors
Evidence comes from source modules under genai_processors/core/, docs under
documentation/docs/development/ and documentation/docs/concepts/realtime.md,
and tests including audio_test.py, rate_limit_audio_test.py,
speech_to_text_test.py, text_to_speech_test.py, vad_test.py,
video_test.py, pdf_test.py, web_test.py, github_test.py,
drive_test.py, filesystem_test.py, event_detection_test.py,
timestamp_test.py, and window_test.py.
Usability checkpoints
A good answer using this sub-skill should:
- Name exact optional packages and environment variables before runnable code.
- Avoid opening microphones, speakers, cameras, browsers, or remote services in a smoke check.
- Preserve non-text parts until a downstream processor intentionally converts them.
- Include concrete MIME types and substream names when they affect routing.
- Route model-specific or applet-specific follow-up work to the proper sibling sub-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).