kali20gakki/msAgent / skills/quantizer/quant-tuning-infer-dit
quant-tuning-infer-dit
Verified在 W8A8 动态量化后,用 MindIE-SD 冒烟测试 DiT 量化权重是否可加载(可选 skill)。与 `quant-tuning-evaluate`(DiT 扩展节)的区别:本 skill 仅做"能跑通 + 权重可用"的轻量验证(4~16 条 prompt,几分钟出结果),不算分;后者产出视频集交 `quant-tuning-score-dit`(AISBench-VBench 16 维度)评分,是调优回路一环。不引入 vLLM-Ascend、不做精度指标评估(FID/CLIP-score 等)。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 7.1 KB |
| assets/inference.example.yaml | config | 2.6 KB |
assets/inference.example.yaml
View on GitHub# DiT 量化后推理配置示例
#
# 本示例以 HunyuanVideo(视频生成)为参考。
# 适用:HunyuanVideo、Wan2.2-T2V/I2V/TI2V、FLUX.1-dev(已迁移)、SD3、Sana、
# HunyuanDiT、CogViewX、SD1.5/SDXL 等所有走 MultimodalPipelineInterface 的 DiT。
#
# 原则:
# - 推理后端统一走 MindIE-SD;
# - 不引入 vLLM-Ascend / Tritonserver;
# - 不做精度指标评估(FID/CLIP-score 等);
# - 仅验证"能跑通 + 权重可用"。
#
# 注意:本 YAML 是 quant-tuning-infer-dit 子 skill 的执行配置,
# 不被 msmodelslim quant CLI 加载,因此可以含顶层路由字段。
# 与量化 YAML(w8a8_dynamic.example.yaml)不同,后者必须严格符合
# MultimodalSDModelslimV1QuantConfig schema。
#
# 推理参数命名差异:字段名 / 默认值 / 入口脚本详见
# ../../msmodelslim-model-adapt/references/dit/inference_config_field_map.md
# 定位方法见 inference_repo 的 README.md 推理示例命令。
apiversion: multimodal_sd_modelslim_v1
metadata:
config_id: hunyuan_video_post_quant_infer
label:
model_family: dit
backend: mindie-sd
purpose: post_quant_verification
spec:
pipeline_factory: mindie-sd
quantized_path: <workdir>/round_1/quantized
inference_repo: /path/to/hunyuan_video_inference_repo # HunyuanVideo 推理仓路径,由用户提供
output_dir: <workdir>/infer_outputs
# HunyuanVideo 参考默认值(与 hyvideo.config parse_args 对齐):
infer_steps: 50 # HunyuanVideo 默认 50;Wan2.2-T2V-A14B 默认 27/sample_steps;FLUX 默认 20/num_inference_steps
cfg_scale: 1.0 # HunyuanVideo 默认 1.0;Wan 系默认 7.5/sample_guide_scale;FLUX 默认 3.5/guidance_scale
embedded_cfg_scale: 6.0 # HunyuanVideo 默认 6.0
flow_shift: 7.0 # HunyuanVideo 默认 7.0;Wan2 系 sample_shift 默认 5.0
model_resolution: "720p" # HunyuanVideo 特有
video_size: [720, 1280] # HunyuanVideo 特有
prompt_list:
- "A stylish woman walks down a Tokyo street filled with warm glowing neon and animated city signage."
- "Two anthropomorphic cats in comfy boxing gear fight intensely on a spotlighted stage."
- "A serene mountain landscape at sunrise, cinematic style."
- "A futuristic city skyline with neon lights at night."
backend:
type: mindie-sd
device: npu:0
metrics: [] # 不做精度指标评估
# 模型族标签(orchestrator 路由用,仅 infer skill 读取)
model_family: ditRelated skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |