[doc] fix: Clarify from_hf_config limitation for export workflows (#2113)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>

Author: yaoyu-33

docs/bridge-guide.md

@@ -199,20 +199,33 @@ model = bridge.to_megatron_model()  # Uses default settings
 ```
 
 ### 3. Leverage the Parameter Streaming API
-You can stream converted weights from Megatron to HF without saving to disk. You can also use config-only loading for architecture exploration without loading weights:
+You can stream converted weights from Megatron to HF without saving to disk:
 
 ```python
 # ✅ Use streaming for large models
 for name, weight in bridge.export_hf_weights(model, cpu=True):
     process_weight(name, weight)
+```
+
+### 4. Use `from_hf_pretrained` for Export Workflows
+
+When exporting Megatron checkpoints back to 🤗 Hugging Face format, always use `from_hf_pretrained()` instead of `from_hf_config()`. The `from_hf_config()` method does not load the tokenizer and other artifacts required for saving a complete 🤗 Hugging Face checkpoint:
 
-# ✅ Use config-only loading for architecture exploration
-config = AutoConfig.from_pretrained("meta-llama/Llama-3-8B")
-bridge = AutoBridge.from_hf_config(config)
-transformer_config = bridge.transformer_config
-print(f"Hidden size: {transformer_config.hidden_size}")
+```python
+from megatron.bridge import AutoBridge
+
+# ✅ Correct: Use from_hf_pretrained for export workflows
+bridge = AutoBridge.from_hf_pretrained("meta-llama/Llama-3.2-1B")
+bridge.export_ckpt("./megatron_checkpoints/llama32_1b", "./hf_exports/llama32_1b")
+
+# ❌ Avoid: from_hf_config lacks artifacts needed for saving
+# config = AutoConfig.from_pretrained("meta-llama/Llama-3.2-1B")
+# bridge = AutoBridge.from_hf_config(config)  # Missing tokenizer, etc.
+# bridge.export_ckpt(...)  # Will fail!
 ```
 
+The `from_hf_config()` method is only suitable for architecture exploration and introspection (e.g., inspecting `transformer_config`), not for checkpoint conversion workflows.
+
 For more examples and advanced usage patterns, see the `examples/conversion/` directory in the repository.
 
 ## Convenience Workflows (Commands)
@@ -229,7 +242,7 @@ python -c "from megatron.bridge import AutoBridge; AutoBridge.import_ckpt('meta-
 ### Megatron → HF export (one call)
 
 ```bash
-python -c "from megatron.bridge import AutoBridge; from transformers import AutoConfig; cfg=AutoConfig.from_pretrained('meta-llama/Llama-3.2-1B'); b=AutoBridge.from_hf_config(cfg); b.export_ckpt('./megatron_checkpoints/llama32_1b','./hf_exports/llama32_1b')"
+python -c "from megatron.bridge import AutoBridge; b=AutoBridge.from_hf_pretrained('meta-llama/Llama-3.2-1B'); b.export_ckpt('./megatron_checkpoints/llama32_1b','./hf_exports/llama32_1b')"
 ```
 
 ### Create Megatron models and run locally