models-and-inference
Route xTuring model selection, loading, saving, generation, and backend-specific inference choices.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- models-and-inference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Route xTuring model selection, loading, saving, generation, and backend-specific inference choices.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/inference-workflows.md4.2 KB
- 参考文档references/model-catalog.md4.6 KB
- 参考文档references/troubleshooting.md2.9 KB
- 脚本scripts/inspect_xturing_install.py4.7 KB
03 SKILL.md 正文
models-and-inference
Use this sub-skill when the user needs to pick a supported model key, load a saved checkpoint, tune generation settings, save a model, or choose between base, LoRA, INT8, and K-bit variants.
Use this for
BaseModel.create("<key>")for supported registry-backed families.BaseModel.load(path)for saved xTuring checkpoints and bundledx/...hub entries.BaseModel.load(path, model_name="<key>")for a plain Hugging Face checkpoint directory withoutxturing.json.GenericModeland the otherGeneric*wrappers for arbitrary local or hub checkpoints.model.generate(...),model.save(...), andmodel.generation_config().
Route elsewhere for
- dataset schemas, dataset conversion, or self-instruct generation ->
data-prep-and-generation - fine-tuning, DPO, or trainer configuration ->
training-and-alignment - CLI chat, API server, or playground UI ->
cli-api-ui - perplexity or adapter-style evaluation ->
evaluation
Read first
references/model-catalog.mdreferences/inference-workflows.mdreferences/troubleshooting.mdscripts/inspect_xturing_install.py
Known gap
StableDiffusionis registered but placeholder-only; instantiating it raisesNotImplementedError, so do not route users there as a usable inference target.
This sub-skill is self-contained; do not depend on the original source checkout for model-selection or inference guidance.
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).