AGENT SKILL

generation-and-api

Use DALLE2-pytorch public generation and model APIs: DALLE2, diffusion prior, decoder, CLIP adapters, inpainting, latent diffusion/VQGAN, checkpoints, tokenizer, and dream CLI.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
generation-and-api
license
MIT
metadata
{ "disco-role": "operating" }
description
Use DALLE2-pytorch public generation and model APIs: DALLE2, diffusion prior, decoder, CLIP adapters, inpainting, latent diffusion/VQGAN, checkpoints, tokenizer, and dream CLI.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

generation-and-api

Use this sub-skill when the task is to construct or call DALLE2-pytorch public Python generation APIs, diagnose model/checkpoint generation failures, or use the installed dream console command for a trained DALL-E 2 checkpoint.

Load This When

  • Building DALLE2, DiffusionPriorNetwork, DiffusionPrior, Unet, or Decoder objects from Python.
  • Choosing between OpenAIClipAdapter, OpenClipAdapter, package CLIP, precomputed embeddings, and trained checkpoints.
  • Sampling with a prior, decoder, or chained DALLE2 model, including cond_scale and prior_cond_scale.
  • Debugging decoder inpainting, cascading decoder resolution order, latent diffusion with VQGanVAE, or tokenizer length issues.
  • Checking that the installed dalle2-pytorch package can import or run a tiny CPU-safe forward-loss smoke test.

Route Elsewhere

  • JSON training launchers, config classes, trainer loops, Accelerate/DeepSpeed commands, checkpoint save/resume during training: ../training-and-configs/SKILL.md.
  • WebDataset, EmbeddingReader, shard layouts, experiment trackers, W&B/HuggingFace/S3 save/load configuration: ../data-and-tracking/SKILL.md.

Runtime Contract

  • Public package: pip install dalle2-pytorch.
  • Public import namespace: import dalle2_pytorch; top-level exports include DALLE2, DiffusionPriorNetwork, DiffusionPrior, Unet, Decoder, OpenAIClipAdapter, OpenClipAdapter, DecoderTrainer, DiffusionPriorTrainer, VQGanVAE, and CLIP.
  • Version covered by this skill: dalle2-pytorch 1.15.6.
  • Full text-to-image generation needs trained prior and decoder weights; realistic CLIP/decoder/prior sampling usually needs GPU memory. CPU is suitable for imports, CLI help, config checks, and tiny synthetic forward-loss smoke tests only.

References And Bundled Script

First Checks

python -m pip show dalle2-pytorch
python scripts/check_dalle2_runtime.py --mode imports
python scripts/check_dalle2_runtime.py --mode cli-help

For a CPU-only synthetic smoke test that does not download model weights:

python scripts/check_dalle2_runtime.py --mode tiny-forward

If any check fails, use references/troubleshooting.md before escalating to training/config or data/tracking guidance.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351