client-inference
Use Petals client APIs for distributed model loading, generation, inference sessions, routing, retries, and swarm selection.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- client-inference
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use Petals client APIs for distributed model loading, generation, inference sessions, routing, retries, and swarm selection.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md2.5 KB
- 参考文档references/troubleshooting.md1.3 KB
- 参考文档references/workflows.md2.2 KB
- 脚本scripts/client_smoke_check.py2.2 KB
03 SKILL.md 正文
Petals Client Inference
Use this sub-skill to write, debug, or adapt Petals client code for distributed text generation, model forward passes, sequence classification, speculative Llama generation, inference-session cache reuse, or route/retry controls.
Route elsewhere
- Server hosting, DHT bootstrap, public/private server commands:
server-swarms. - Trainable prompts, deep prompt tuning, optimizer loops:
prompt-tuning. - Block conversion, tensor parallel, quantization internals:
distributed-blocks. - Benchmark command construction and maintainer test selection:
benchmarks-maintenance.
Workflow
- Read references/api-reference.md for entry points, config fields, and constraints.
- Use references/workflows.md for generation, private-swarm selection, inference-session, classification, and speculative-generation recipes.
- Run the safe no-network checker:
python scripts/client_smoke_check.py --pretty. - For route, retry, tokenizer, model access, or session errors, use references/troubleshooting.md.
Principles
Treat Petals classes as Transformers-compatible wrappers whose embeddings/heads run locally and transformer blocks run on remote servers. Make swarm choice explicit when reproducibility matters. When generating without an active session, pass exactly one of max_new_tokens or max_length; for interactive continuation, create one inference_session(max_length=...) large enough for the entire exchange.
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).