refactor and support for multi algs fusion

.claude/skills/adapt-new-diffusion-model/SKILL.md

@@ -52,15 +52,12 @@ def is_diffusion_model(model_or_path):
 ```
 
 If your model doesn't have `model_index.json`, either create one in the model
-directory or pass diffusion-specific options through new-architecture
-`ExtraConfig` / `AutoRound` kwargs:
+directory or pass diffusion-specific options through `AutoRound` kwargs:
 
 ```python
-from auto_round.compressors.config import ExtraConfig
-
 ar = AutoRound(
     model,
-    extra_config=ExtraConfig(num_inference_steps=5),
+    num_inference_steps=5,
 )
 ```
 

.claude/skills/add-vlm-model/SKILL.md

@@ -104,7 +104,7 @@ The new architecture routes multimodal calibration through:
 
 If your model works with an existing template/processor, prefer passing
 `template=...`, `processor=...`, or `image_processor=...` through `AutoRound` /
-`ExtraConfig` instead of adding compressor code.
+kwargs instead of adding compressor code.
 
 ## Step 3: Add Calibration Template
 

.gitignore

@@ -12,3 +12,4 @@ tmp_autoround/
 ut_log_dir/
 CLAUDE.local.md
 docs/plan/
+.codegraph/