cp: [docs] Add supported models page (980) into r0.2.0

docs/broken_links_false_positives.json

@@ -0,0 +1 @@
+{"uri": "https://www.llama.com/models/llama-3/"}

docs/index.md

@@ -22,6 +22,14 @@ bridge-guide.md
 bridge-tech-details.md
 ```
 
+```{toctree}
+:caption: Supported Models
+:hidden:
+
+models/llm/index.md
+models/vlm/index.md
+```
+
 ```{toctree}
 :caption: Training and Customization
 :hidden:

docs/models/llm/index.md

@@ -0,0 +1,10 @@
+# Large Language Models
+
+This section documents Large Language Models supported by Megatron Bridge, with examples for converting to/from 🤗 Hugging Face and links to training recipes.
+
+```{toctree}
+:hidden:
+
+llama3.md
+qwen.md
+```

docs/models/llm/llama3.md

@@ -0,0 +1,43 @@
+# Llama 3
+
+[Meta’s Llama](https://www.llama.com/models/llama-3/) builds on the general transformer decoder framework with some key additions such as pre-normalization, SwiGLU activations, and Rotary Positional Embeddings (RoPE). More information is available in the companion paper [“Llama: Open and Efficient Foundation Language Models”](https://arxiv.org/abs/2302.13971). With a wide variety of model sizes - Llama has options for every inference budget.
+
+Llama family models are supported via the Bridge system with auto-detected configuration and weight mapping.
+
+## Conversion with 🤗 Hugging Face
+
+### Load HF → Megatron
+```python
+from megatron.bridge import AutoBridge
+
+# Example: Llama 3.1 8B
+bridge = AutoBridge.from_hf_pretrained("meta-llama/Llama-3.1-8B-Instruct")
+provider = bridge.to_megatron_provider()
+
+# Configure parallelism before instantiating the model
+provider.tensor_model_parallel_size = 8
+provider.pipeline_model_parallel_size = 1
+
+model = provider.provide_distributed_model(wrap_with_ddp=False)
+```
+
+### Export Megatron → HF
+```python
+# Convert from a Megatron checkpoint directory to HF format
+bridge.export_ckpt(
+    megatron_path="/results/llama3_8b/checkpoints/iter_00002000",
+    hf_path="./llama-hf-export",
+)
+```
+
+## Examples
+- Checkpoint import/export: [examples/conversion/convert_checkpoints.py](https://github.com/NVIDIA-NeMo/Megatron-Bridge/blob/main/examples/conversion/convert_checkpoints.py)
+- Generate text (HF→Megatron): [examples/conversion/hf_to_megatron_generate_text.py](https://github.com/NVIDIA-NeMo/Megatron-Bridge/blob/main/examples/conversion/hf_to_megatron_generate_text.py)
+
+## Pretrain recipes
+- See: [bridge.recipes.llama.llama3](../../apidocs/bridge/bridge.recipes.llama.llama3.md)
+
+## Related docs
+- Recipe usage: [Recipe usage](../../recipe-usage.md)
+- Customizing the training recipe configuration: [Configuration overview](../../training/config-container-overview.md)
+- Training entry points: [Entry points](../../training/entry-points.md)

docs/models/llm/qwen.md

@@ -0,0 +1,62 @@
+# Qwen
+
+Qwen2/2.5/3 models are supported via the Bridge with QK layernorm handling (Qwen3) and bias in QKV (Qwen2).
+
+## Conversion with 🤗 Hugging Face
+
+### Load HF → Megatron
+```python
+from megatron.bridge import AutoBridge
+
+# Example: Qwen3 7B
+bridge = AutoBridge.from_hf_pretrained("Qwen/Qwen3-7B")
+provider = bridge.to_megatron_provider()
+
+provider.tensor_model_parallel_size = 8
+model = provider.provide_distributed_model(wrap_with_ddp=False)
+```
+
+### Export Megatron → HF
+```python
+bridge.save_hf_pretrained(model, "./qwen-hf-export")
+# or convert a checkpoint directory
+bridge.export_ckpt(
+    megatron_path="/results/qwen3_8b/checkpoints/iter_00002000",
+    hf_path="./qwen-hf-export",
+)
+```
+
+### Examples
+- Checkpoint import/export: `examples/conversion/convert_checkpoints.py`
+- Generate text (HF→Megatron): `examples/conversion/hf_to_megatron_generate_text.py`
+
+## Pretrain recipes
+- Example usage (Qwen3 8B)
+```python
+from megatron.bridge.recipes.qwen import qwen3_8b_pretrain_config
+
+cfg = qwen3_8b_pretrain_config(
+    hf_path="Qwen/Qwen3-8B",
+    data_paths=["/path/to/dataset.nvjsonl"],
+    dir="/results/qwen3_8b",
+)
+```
+
+- API reference for Qwen recipes:
+  - Qwen recipes overview: [bridge.recipes.qwen](../../apidocs/bridge/bridge.recipes.qwen.md)
+  - Qwen2 recipes: [bridge.recipes.qwen.qwen2](../../apidocs/bridge/bridge.recipes.qwen.qwen2.md)
+  - Qwen3 recipes: [bridge.recipes.qwen.qwen3](../../apidocs/bridge/bridge.recipes.qwen.qwen3.md)
+  - Qwen3 MoE recipes: [bridge.recipes.qwen.qwen3_moe](../../apidocs/bridge/bridge.recipes.qwen.qwen3_moe.md)
+
+## Finetuning recipes
+- Coming soon
+
+## Hugging Face model cards
+- Qwen2: `https://huggingface.co/Qwen/Qwen2-7B`
+- Qwen2.5: `https://huggingface.co/Qwen/Qwen2.5-7B`
+- Qwen3: `https://huggingface.co/Qwen/Qwen3-7B`
+
+## Related docs
+- Recipe usage and customization: [Recipe usage](../../recipe-usage.md)
+- Training configuration: [Configuration overview](../../training/config-container-overview.md)
+- Training entry points: [Entry points](../../training/entry-points.md)

docs/models/vlm/index.md

@@ -0,0 +1,8 @@
+# Vision Language Models
+
+This section documents Vision Language Models supported by Megatron Bridge.
+
+```{toctree}
+:hidden:
+
+```

docs/recipe-usage.md

@@ -3,6 +3,13 @@
 Megatron Bridge provides production-ready training recipes for several popular models. You can find an overview of supported recipes and 🤗 HuggingFace bridges [here](index.md#supported-models).
 This guide will cover the next steps to make use of a training recipe, including how to [override configuration](#overriding-configuration) and how to [launch a job](#launch-methods).
 
+## Overview
+
+- **Coverage**: We provide recipes across select model families and sizes, including Llama, Qwen, DeepSeek, and Nemotron-H (Mamba-based).
+- **Defaults**: Each recipe sets defaults meant for convergence and performance across parallelisms, precision data types, and optimizer & scheduler choices. These recipes can be used as a high-quality starting point. 
+- **Integration**: Recipes return a single `ConfigContainer` that plugs directly into our training [entry points](training/entry-points.md) (see the published docs as well: https://docs.nvidia.com/nemo/megatron-bridge/latest/training/entry-points.html).
+- **Customization**: You can override any part of the recipe (Python, YAML, CLI) to adapt to your data, scale, and objectives.
+
 ## Overriding configuration
 
 Recipes are provided through a {py:class}`~bridge.training.config.ConfigContainer` object. This is a dataclass that holds all configuration objects needed for training. You can find a more detailed overview of the `ConfigContainer` [here](training/config-container-overview.md).