app-deployment
Install, configure, launch, and sanity-check InternGPT/InternChat services.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- app-deployment
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Install, configure, launch, and sanity-check InternGPT/InternChat services.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/cli-and-config.md5.6 KB
- 参考文档references/deployment-recipes.md5.6 KB
- 参考文档references/model-zoo.md5.6 KB
- 参考文档references/troubleshooting.md6.8 KB
- 脚本scripts/validate_load_plan.py12.8 KB
03 SKILL.md 正文
app-deployment
Use this sub-skill when the task is to install, configure, launch, containerize, or sanity-check an InternGPT/InternChat Gradio service. It owns app-level CLI planning, model-zoo placement, HTTPS/OpenAI setup, Docker GPU deployment, and static validation of --load/--tab choices.
Route detailed workflow questions elsewhere:
- Read ../visual-dialogue-tools/SKILL.md for detailed image, mask, OCR, SAM, Husky VQA, inpainting, and image-tool behavior after the service is running.
- Read ../cross-modal-generation/SKILL.md for detailed ImageBind audio/thermal/image-conditioned generation and DragGAN point-editing usage.
- Read ../video-understanding/SKILL.md for detailed video caption/action/dense-caption/TikTok workflows.
Deployment routing
- For CLI flags, supported
--loadclasses, tab names, and launch-plan sanity checks, read references/cli-and-config.md. - For
model_zoo/, checkpoints, Husky/LLaMA conversion, SAM, LaMa, Tag2Text, GRiT, StyleGAN, ImageBind, and StableUnCLIP prerequisites, read references/model-zoo.md. - For basic, full, DragGAN-only, HTTPS, OpenAI, and Docker GPU launch recipes, read references/deployment-recipes.md.
- For common failures such as missing
model_zoo, bad LLaMA/Husky partial conversion, missing OpenAI key,OPENAI_API_BASE, certificate errors, CUDA/VRAM mismatch, Docker placeholder volumes, detectron2/OpenCV pitfalls, and-eexpectations, read references/troubleshooting.md. - Run scripts/validate_load_plan.py before recommending a launch command or compose command when the user asks which
--loador--tabstring to use.
Quick decision guide
- Choose the smallest tab set that matches the user's task:
DragGANfor StyleGAN point editing,Imagefor image upload/click/OCR/segmentation/dialogue,Audiofor ImageBind audio input, andVideofor video upload/analysis. - Choose the smallest direct
--loadlist that provides those tabs' tools; template tools are auto-created only after their prerequisites load. - Prefer
-e/--e-modefor constrained VRAM, but do not describe it as CPU-only: the app still expects CUDA for normal launches, and the speech model is initialized oncuda:0. - Check
model_zoo/,certificate/, and OpenAI configuration before treating a launch failure as a Python bug. - For Docker, replace every placeholder volume with real host directories and keep GPU reservation/container-toolkit requirements explicit.
Safe default recipes
- Basic image-dialogue service:
HuskyVQA_cuda:0,SegmentAnything_cuda:0,ImageOCRRecognition_cuda:0with-eon a CUDA machine and a populatedmodel_zoo/. - Low-memory DragGAN-only HTTPS service:
StyleGAN_cuda:0,--tab DragGAN,--https, and-e, pluscertificate/cert.pem,certificate/key.pem, and StyleGAN checkpoint readiness. - Full multimodal demo: use the long full-feature load string in
references/deployment-recipes.md; warn that it is heavy and requires multiple model families, CUDA, and usually far more setup than a task-specific launch.
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).