[Docs] Add supported model tables to pretrain_sft advanced tutorial

.agents/skills

@@ -0,0 +1 @@
+../.claude/skills

.claude/skills/xtuner-sync-supported-models/SKILL.md

@@ -0,0 +1,123 @@
+---
+name: xtuner-sync-supported-models
+description: Synchronize xtuner's supported model documentation (docs/en/pretrain_sft/advanced_tutorial/model.md and docs/zh_cn/pretrain_sft/advanced_tutorial/model.md) with the actual Config classes defined under xtuner/v1/model/. Use when (1) new TransformerConfig, MoEConfig, or BaseComposeConfig subclasses are added, removed, or renamed in xtuner/v1/model/, (2) existing model configs change their inheritance hierarchy, scale, or HuggingFace counterpart, or (3) a code review or user request points out that model.md is out of sync with the codebase.
+---
+
+# Update XTuner Supported Model Docs
+
+Keep the English and Chinese `model.md` files synchronized with the actual Config classes in `xtuner/v1/model/`.
+
+## Scan the Codebase
+
+Run the bundled scan script from the xtuner project root to discover all Config classes and their inheritance:
+
+```bash
+python3 .agents/skills/xtuner-sync-supported-models/scripts/scan_model_configs.py
+```
+
+The script outputs JSON with two keys:
+- `configs`: list of every `*Config` class under `xtuner/v1/model/` with its parent classes and file path
+- `children`: parent-to-children mapping for the hierarchy tree
+
+## What to Update
+
+Compare the script output against the two files:
+- `docs/en/pretrain_sft/advanced_tutorial/model.md`
+- `docs/zh_cn/pretrain_sft/advanced_tutorial/model.md`
+
+Both files share the same structure and must stay in sync:
+
+1. **Base Config Classes** — configs that directly inherit from `TransformerConfig` (or `MoEConfig`) and provide a `from_hf` classmethod for loading HuggingFace weights
+2. **Concrete Model Configs** — fixed-scale subclasses of the base configs above
+3. **Compose Models** — multimodal configs that inherit from `BaseComposeConfig`
+4. **Inheritance Hierarchy** — a text tree showing the full `XTunerBaseModelConfig` hierarchy
+
+### Rules for the Base Config table
+
+Include these direct descendants of `TransformerConfig`/`MoEConfig`:
+- `Qwen2DenseConfig`
+- `Qwen3DenseConfig`
+- `DeepSeekV3Config`
+- `GptOssConfig`
+- `Qwen3MoEConfig`
+
+Exclude from the base table:
+- `MoEConfig` — it is an intermediate base class, not a usable model family
+- `Qwen3_5_VLTextMoEConfig` — it is an intermediate base with only one concrete child; its child `Qwen3_5_VLTextMoE35BA3BConfig` belongs under the MoE concrete table
+
+### Rules for the Concrete Model table
+
+Include every concrete subclass that has fixed parameter defaults. For each row note:
+- `Config Class`
+- `Base Class / Family`
+- `Architecture Type`: `Dense`, `MoE`, `Dense (VL backbone)`, `MoE (VL backbone)`
+- `Scale / Notes`: parameter count or total/activated size; for VL backbones note "for multimodal"
+
+`DeepSeekV3Config` appears here even though it has no separate base entry (it is both base and concrete).
+
+### Rules for the Compose Models section
+
+Include three sub-tables:
+1. **Compose Base Config Classes** — `Qwen3VLBaseConfig`, `InternVLBaseConfig`, `InternS1BaseConfig`
+   - `Qwen3VLBaseConfig`: VL model based on Qwen3 text backbone
+   - `InternVLBaseConfig`: VL model based on InternViT + Qwen3
+   - `InternS1BaseConfig`: Science multimodal model based on InternViT + Qwen3
+2. **Concrete Compose Model Configs** — every subclass of the above bases; for each row note the wrapped `Text Config` and scale
+
+### Rules for the Inheritance Hierarchy tree
+
+Rebuild the tree from `XTunerBaseModelConfig` with two top-level branches:
+
+```text
+XTunerBaseModelConfig
+├── TransformerConfig
+│   ├── Dense Models
+│   │   ├── Qwen2DenseConfig
+│   │   │   └── Qwen2Dense7BConfig
+│   │   └── Qwen3DenseConfig
+│   │       ├── Qwen3Dense8BConfig
+│   │       ├── Qwen3Dense4BConfig
+│   │       ├── Qwen3Dense0P6BConfig
+│   │       ├── Qwen3VLTextDense4BConfig
+│   │       └── Qwen3VLTextDense8BConfig
+│   └── MoE Models (via MoEConfig)
+│       ├── DeepSeekV3Config
+│       ├── GptOssConfig
+│       │   ├── GptOss21BA3P6Config
+│       │   └── GptOss117BA5P8Config
+│       ├── Qwen3MoEConfig
+│       │   ├── Qwen3MoE30BA3Config
+│       │   ├── Qwen3MoE235BA22Config
+│       │   ├── Qwen3MoEFoPEConfig
+│       │   ├── Qwen3VLTextMoE30BA3Config
+│       │   └── Qwen3VLTextMoE235BA22Config
+│       └── Qwen3_5_VLTextMoEConfig
+│           └── Qwen3_5_VLTextMoE35BA3BConfig
+└── BaseComposeConfig
+    ├── Qwen3VLBaseConfig
+    │   ├── Qwen3VLMoE30BA3Config
+    │   ├── Qwen3VLMoE235BA22Config
+    │   ├── Qwen3VLDense4BConfig
+    │   ├── Qwen3VLDense8BConfig
+    │   └── Qwen3_5_BaseConfig
+    │       └── Qwen3_5_VLMoE35BA3Config
+    ├── InternVLBaseConfig
+    │   ├── InternVL3P5Dense8BConfig
+    │   ├── InternVL3P5MoE30BA3Config
+    │   └── InternVL3P5Dense1BConfig
+    └── InternS1BaseConfig
+        ├── InternS1Config
+        └── InternS1MiniConfig
+```
+
+When new configs are added, insert them into the appropriate branch following the same indentation style.
+
+## Translation Notes
+
+Keep the Chinese `model.md` (`docs/zh_cn/...`) structurally identical to the English one. Translate:
+- Section headings
+- Table header cells
+- Description cells (e.g., "Image / Video + Text" → "图像/视频 + 文本")
+- Scale descriptions (e.g., "~7B parameters" → "约 7B 参数", "FoPE variant" → "FoPE 变体")
+
+Do **not** translate Config class names, file paths, or code identifiers.

.claude/skills/xtuner-sync-supported-models/scripts/scan_model_configs.py

@@ -0,0 +1,71 @@
+#!/usr/bin/env python3
+"""Scan xtuner/v1/model for all Config classes and output model info as JSON."""
+
+import json
+import re
+import sys
+from pathlib import Path
+
+# We care about configs that are part of the supported model hierarchy
+RELEVANT_BASES = {
+    "TransformerConfig",
+    "MoEConfig",
+    "BaseComposeConfig",
+    "XTunerBaseModelConfig",
+    # Known intermediate/family bases
+    "Qwen2DenseConfig",
+    "Qwen3DenseConfig",
+    "Qwen3MoEConfig",
+    "Qwen3_5_VLTextMoEConfig",
+    "GptOssConfig",
+    "DeepSeekV3Config",
+    "Qwen3VLBaseConfig",
+    "Qwen3_5_BaseConfig",
+    "InternVLBaseConfig",
+    "InternS1BaseConfig",
+}
+
+
+def scan_file(path: Path):
+    text = path.read_text()
+    # Match class definitions like: class FooConfig(BarConfig):
+    pattern = r"^class\s+(\w+Config)\s*\(([^)]+)\):"
+    results = []
+    for m in re.finditer(pattern, text, re.MULTILINE):
+        class_name = m.group(1)
+        parents = [p.strip() for p in m.group(2).split(",")]
+        results.append({"class": class_name, "parents": parents, "file": str(path)})
+    return results
+
+
+def main():
+    root = Path(sys.argv[1]) if len(sys.argv) > 1 else Path(".")
+    model_dir = root / "xtuner" / "v1" / "model"
+    if not model_dir.exists():
+        print(f"Model directory not found: {model_dir}", file=sys.stderr)
+        sys.exit(1)
+
+    all_configs = []
+    for py_file in sorted(model_dir.rglob("*.py")):
+        all_configs.extend(scan_file(py_file))
+
+    # Build parent -> children map
+    children: dict[str, list[str]] = {}
+    for cfg in all_configs:
+        for p in cfg["parents"]:
+            if p in RELEVANT_BASES or p.endswith("Config"):
+                children.setdefault(p, []).append(cfg["class"])
+
+    # Deduplicate
+    for k in children:
+        children[k] = sorted(set(children[k]))
+
+    output = {
+        "configs": all_configs,
+        "children": children,
+    }
+    print(json.dumps(output, indent=2, ensure_ascii=False))
+
+
+if __name__ == "__main__":
+    main()

docs/en/pretrain_sft/advanced_tutorial/model.md

@@ -1,3 +1,110 @@
 # Model
 
-Coming soon...
+XTuner v1's `TrainEngine` supports a variety of Transformer architectures through different `TransformerConfig` subclasses. The documentation below summarizes the currently supported models (RL-related configs are excluded).
+
+## Base Config Classes
+
+The following table lists the **base config classes** that define each model family. They provide the `from_hf` interface for loading pretrained weights from HuggingFace.
+
+| Base Config Class | Model Family | Architecture Type | HuggingFace Counterpart |
+|---|---|---|---|
+| `Qwen2DenseConfig` | Qwen2 Dense | Dense | `Qwen2ForCausalLM` |
+| `Qwen3DenseConfig` | Qwen3 Dense | Dense | `Qwen3ForCausalLM` |
+| `DeepSeekV3Config` | DeepSeek-V3 | MoE | `DeepseekV3ForCausalLM` |
+| `GptOssConfig` | GPT-OSS | MoE | `GptOssForCausalLM` |
+| `Qwen3MoEConfig` | Qwen3 MoE | MoE | `Qwen3MoeForCausalLM` |
+
+## Concrete Model Configs
+
+The following table lists the **concrete model configs** that inherit from the base classes above. Each config corresponds to a specific model scale or variant.
+
+| Config Class | Base Class / Family | Architecture Type | Scale / Notes |
+|---|---|---|---|
+| `Qwen2Dense7BConfig` | `Qwen2DenseConfig` | Dense | ~7B parameters |
+| `Qwen3Dense8BConfig` | `Qwen3DenseConfig` | Dense | ~8B parameters |
+| `Qwen3Dense4BConfig` | `Qwen3DenseConfig` | Dense | ~4B parameters |
+| `Qwen3Dense0P6BConfig` | `Qwen3DenseConfig` | Dense | ~0.6B parameters |
+| `Qwen3VLTextDense4BConfig` | `Qwen3DenseConfig` | Dense (VL backbone) | ~4B parameters, for multimodal |
+| `Qwen3VLTextDense8BConfig` | `Qwen3DenseConfig` | Dense (VL backbone) | ~8B parameters, for multimodal |
+| `DeepSeekV3Config` | — | MoE | ~671B total / ~37B activated |
+| `GptOss21BA3P6Config` | `GptOssConfig` | MoE | ~21B total / ~3.6B activated |
+| `GptOss117BA5P8Config` | `GptOssConfig` | MoE | ~117B total / ~5.8B activated |
+| `Qwen3MoE30BA3Config` | `Qwen3MoEConfig` | MoE | ~30B total / ~3B activated |
+| `Qwen3MoE235BA22Config` | `Qwen3MoEConfig` | MoE | ~235B total / ~22B activated |
+| `Qwen3MoEFoPEConfig` | `Qwen3MoEConfig` | MoE | FoPE (Frequency-based Position Embedding) variant |
+| `Qwen3VLTextMoE30BA3Config` | `Qwen3MoEConfig` | MoE (VL backbone) | ~30B total, for multimodal |
+| `Qwen3VLTextMoE235BA22Config` | `Qwen3MoEConfig` | MoE (VL backbone) | ~235B total, for multimodal |
+| `Qwen3_5_VLTextMoE35BA3BConfig` | `Qwen3_5_VLTextMoEConfig` | MoE (VL backbone) | ~35B total / ~3B activated, for multimodal |
+
+## Compose Models
+
+In addition to pure text models, XTuner also supports **multimodal compose models** that combine a vision encoder, a projector, and a language model. These configs inherit from `BaseComposeConfig` rather than `TransformerConfig` directly, but they wrap the text configs listed above.
+
+### Compose Base Config Classes
+
+| Base Config Class | Model Family | Modality | Description |
+|---|---|---|---|
+| `Qwen3VLBaseConfig` | Qwen3-VL | Image / Video + Text | VL model based on Qwen3 text backbone |
+| `InternVLBaseConfig` | InternVL | Image + Text | VL model based on InternViT + Qwen3 |
+| `InternS1BaseConfig` | InternS1 | Image + Text | Science multimodal model based on InternViT + Qwen3 |
+
+### Concrete Compose Model Configs
+
+| Config Class | Compose Base / Family | Text Config | Scale / Notes |
+|---|---|---|---|
+| `Qwen3VLMoE30BA3Config` | `Qwen3VLBaseConfig` | `Qwen3VLTextMoE30BA3Config` | ~30B total, MoE VL |
+| `Qwen3VLMoE235BA22Config` | `Qwen3VLBaseConfig` | `Qwen3VLTextMoE235BA22Config` | ~235B total, MoE VL |
+| `Qwen3VLDense4BConfig` | `Qwen3VLBaseConfig` | `Qwen3VLTextDense4BConfig` | ~4B parameters, Dense VL |
+| `Qwen3VLDense8BConfig` | `Qwen3VLBaseConfig` | `Qwen3VLTextDense8BConfig` | ~8B parameters, Dense VL |
+| `Qwen3_5_VLMoE35BA3Config` | `Qwen3_5_BaseConfig` | `Qwen3_5_VLTextMoE35BA3BConfig` | ~35B total / ~3B activated, MoE VL |
+| `InternVL3P5Dense8BConfig` | `InternVLBaseConfig` | `Qwen3Dense8BConfig` | ~8B parameters, Dense VL |
+| `InternVL3P5MoE30BA3Config` | `InternVLBaseConfig` | `Qwen3MoE30BA3Config` | ~30B total, MoE VL |
+| `InternVL3P5Dense1BConfig` | `InternVLBaseConfig` | `Qwen3Dense0P6BConfig` | ~1B parameters, Dense VL |
+| `InternS1Config` | `InternS1BaseConfig` | `Qwen3MoE235BA22Config` | ~235B total, MoE multimodal |
+| `InternS1MiniConfig` | `InternS1BaseConfig` | `Qwen3Dense8BConfig` | ~8B parameters, Dense multimodal |
+
+## Inheritance Hierarchy
+
+The following diagram shows the complete inheritance hierarchy of all config classes supported by `TrainEngine`, including both `TransformerConfig` and `BaseComposeConfig` branches.
+
+```text
+XTunerBaseModelConfig
+├── TransformerConfig
+│   ├── Dense Models
+│   │   ├── Qwen2DenseConfig
+│   │   │   └── Qwen2Dense7BConfig
+│   │   └── Qwen3DenseConfig
+│   │       ├── Qwen3Dense8BConfig
+│   │       ├── Qwen3Dense4BConfig
+│   │       ├── Qwen3Dense0P6BConfig
+│   │       ├── Qwen3VLTextDense4BConfig
+│   │       └── Qwen3VLTextDense8BConfig
+│   └── MoE Models (via MoEConfig)
+│       ├── DeepSeekV3Config
+│       ├── GptOssConfig
+│       │   ├── GptOss21BA3P6Config
+│       │   └── GptOss117BA5P8Config
+│       ├── Qwen3MoEConfig
+│       │   ├── Qwen3MoE30BA3Config
+│       │   ├── Qwen3MoE235BA22Config
+│       │   ├── Qwen3MoEFoPEConfig
+│       │   ├── Qwen3VLTextMoE30BA3Config
+│       │   └── Qwen3VLTextMoE235BA22Config
+│       └── Qwen3_5_VLTextMoEConfig
+│           └── Qwen3_5_VLTextMoE35BA3BConfig
+└── BaseComposeConfig
+    ├── Qwen3VLBaseConfig
+    │   ├── Qwen3VLMoE30BA3Config
+    │   ├── Qwen3VLMoE235BA22Config
+    │   ├── Qwen3VLDense4BConfig
+    │   ├── Qwen3VLDense8BConfig
+    │   └── Qwen3_5_BaseConfig
+    │       └── Qwen3_5_VLMoE35BA3Config
+    ├── InternVLBaseConfig
+    │   ├── InternVL3P5Dense8BConfig
+    │   ├── InternVL3P5MoE30BA3Config
+    │   └── InternVL3P5Dense1BConfig
+    └── InternS1BaseConfig
+        ├── InternS1Config
+        └── InternS1MiniConfig
+```