text-conditional
Router for LlamaGen caption-conditioned training, sampling, prompt batches, and t2i evaluation.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- text-conditional
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Router for LlamaGen caption-conditioned training, sampling, prompt batches, and t2i evaluation.
- disable-model-invocation
- true
02 资源文件 10
- 参考文档references/cli-reference.md2.7 KB
- 参考文档references/evaluation.md1.0 KB
- 参考文档references/t5.md1.2 KB
- 参考文档references/troubleshooting.md2.2 KB
- 参考文档references/workflows.md2.6 KB
- 脚本scripts/evaluate_t2i.sh213 B
- 脚本scripts/sample_t2i_coco.sh420 B
- 脚本scripts/sample_t2i_parti.sh420 B
- 脚本scripts/train_t2i_stage1.sh523 B
- 脚本scripts/train_t2i_stage2.sh601 B
03 SKILL.md 正文
Text-Conditional
Use this sub-skill for the text-conditioned LlamaGen path: stage-1 and stage-2 training, prompt sampling, and evaluation of caption-conditioned generations.
Owns
- Stage-1 and stage-2 t2i training.
- COCO and Parti prompt sampling.
- T5 embedder behavior, padding, and cache layout for the text-conditioning path.
- t2i evaluation inputs, outputs, and packaging expectations.
- Checkpoint / precision /
--from-fsdpguidance for the text-conditioned path.
Routes out
- Tokenizer training, finetuning, reconstruction, and code/image round trips ->
tokenizers - Precomputed code or T5 feature extraction, OpenImages manifests, and cache prep ->
data-preparation - ImageNet class-conditional generation, serving, and c2i evaluation ->
class-conditional - Remote publishing or upload helpers -> excluded
Best entry points
- Training:
scripts/train_t2i_stage1.sh,scripts/train_t2i_stage2.sh - Sampling:
scripts/sample_t2i_coco.sh,scripts/sample_t2i_parti.sh - Evaluation:
scripts/evaluate_t2i.sh - T5 behavior:
references/t5.md - Prompt / batch layout:
references/evaluation.md - Workflow selection:
references/workflows.md
Read before answering
references/workflows.mdreferences/cli-reference.mdreferences/t5.mdreferences/evaluation.mdreferences/troubleshooting.md
Fast routing rules
- If the user asks to create or refresh text features, route to
data-preparation. - If the user asks for VQ, VQGAN, VAE, or Consistency Decoder reconstruction, route to
tokenizers. - If the user asks for ImageNet c2i training, sampling, serving, or c2i evaluation, route to
class-conditional. - If the user asks about t2i only, stay here and use the bundled wrappers.
- Keep
--no-left-padding, prompt CSV/TSV formatting, and evaluation folder layout in mind before running a job.
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).