cli-inference
Run, script, and troubleshoot Chai-1 folding with the chai-lab fold CLI and chai_lab.chai1.run_inference API.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- cli-inference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Run, script, and troubleshoot Chai-1 folding with the chai-lab fold CLI and chai_lab.chai1.run_inference API.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md8.1 KB
- 参考文档references/inference-workflows.md6.0 KB
- 参考文档references/troubleshooting.md5.7 KB
- 脚本scripts/write_inference_template.py8.3 KB
03 SKILL.md 正文
Chai-1 CLI and Inference
Use this sub-skill when the task is to run, template, script, or debug Chai-1 molecular structure prediction through chai-lab fold or chai_lab.chai1.run_inference.
Use This For
- Constructing
chai-lab foldcommands, including device, seed, recycle, sample, and output settings. - Writing Python inference scripts around
chai_lab.chai1.run_inferenceand readingStructureCandidatesoutputs. - Diagnosing inference setup failures: non-empty output directories, CUDA/device problems, model downloads, and long-running jobs.
- Routing advanced custom-context work to
make_all_atom_feature_contextplusrun_folding_on_context.
Route Elsewhere
- FASTA entity syntax, duplicate entity names, ligands, DNA/RNA, and chain-name choices:
../input-data-formats/SKILL.md. - MSA
.aligned.pqt, ColabFold server,a3m-to-pqt, and template hit files:../msa-templates/SKILL.md. - Contact, pocket, covalent, and glycan restraints:
../restraints-glycans/SKILL.md.
Quick Start
Install a pinned public package build or a Git build in a Linux Python 3.10+ environment:
pip install chai_lab==0.6.1
# or: pip install git+https://github.com/chaidiscovery/chai-lab.git
Run a minimal ESM-only fold, writing to a new or empty directory:
chai-lab fold input.fasta outputs/chai-run --seed 42 --device cuda:0
Print citation metadata without running inference:
chai-lab citation
Generate a configurable Python script template without starting a GPU job:
python scripts/write_inference_template.py \
--fasta input.fasta \
--output-dir outputs/chai-run \
--script-out run_chai_inference.py \
--device cuda:0 \
--seed 42
Runtime Expectations
- Chai-1 inference is intended for Linux, Python
>=3.10, and a CUDA GPU with bfloat16 support; do not promise practical CPU inference. - The default device is effectively
cuda:0ifdeviceis not provided. - Model and helper assets download automatically on first use. Set
CHAI_DOWNLOADS_DIR=/path/to/cachewhen the default package-local download location is unsuitable. - Folding can be slow and memory-intensive. Reduce
--num-diffn-samples,--num-diffn-timesteps, or input size for faster/cheaper exploratory runs. output_dirmust not contain existing files; create a fresh run directory or deliberately clean it before invoking inference.
Reference Map
references/inference-workflows.md: command recipes, output handling, and script templates.references/api-reference.md: exact Python APIs, parameters, outputs, and advanced context route.references/troubleshooting.md: common failures and safe diagnosis steps.scripts/write_inference_template.py: bundled helper that writes a configurable Python inference template.
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).