Update README with vLLM usage instructions

README.md

@@ -65,6 +65,8 @@ pip install flash-linear-attention  # required for linear attention operators (K
 
 ## Usage
 
+### HuggingFace Transformers
+
 This model uses a custom architecture and requires `trust_remote_code=True`.
 
 ```python
@@ -101,6 +103,54 @@ response = processor.decode(output_ids[0][inputs.input_ids.shape[1]:], skip_spec
 print(response)
 ```
 
+### vLLM Acceleration
+
+For significantly faster inference, MOSAIC-4B supports vLLM via an out-of-tree monkey patch plugin.
+
+**Install the plugin:**
+
+```bash
+pip install -e .
+# or from the model directory:
+pip install -e /path/to/MOSAIC-4B
+```
+
+**Prerequisites:**
+
+| Package | Version |
+|---------|---------|
+| vllm | ≥ 0.17.0, < 0.18.0 |
+| flash-linear-attention (fla) | ≥ 0.4.2 |
+| einops | latest |
+
+**Usage:**
+
+```python
+import nas_child_vl_vllm  # Register MOSAIC-4B with vLLM (must be imported before vLLM)
+
+from vllm import LLM, SamplingParams
+
+llm = LLM(
+    model="LiAuto-DSR/MOSAIC-4B",
+    trust_remote_code=True,
+    enforce_eager=True,  # Required: FLA layers don't support CUDAGraph
+    dtype="bfloat16",
+)
+
+output = llm.generate("Hello, how are you?", SamplingParams(max_tokens=128))
+print(output)
+```
+
+> **Note:** `enforce_eager=True` is required because the FLA (Flash Linear Attention) GDN/KDA layers do not support CUDA graphs. `trust_remote_code=True` is needed to load the custom config and model classes.
+
+**What the plugin does when you `import nas_child_vl_vllm`:**
+
+1. Registers `NasChildVLConfig` with vLLM's config discovery system
+2. Maps `model_type="nas-child-vl"` to the config class
+3. Maps `NasChildVLModelForCausalLM` architecture to the vLLM model class (lazy-loaded)
+4. Registers `MambaModelConfig` for proper hybrid state management (GDN/KDA layers)
+5. Patches `rmsnorm_fn` to support sigmoid activation (needed by KDA layers)
+
 ---
 
 ## Dependencies
@@ -110,6 +160,8 @@ print(response)
 | transformers | ≥ 4.57.0 |
 | torch | ≥ 2.0 |
 | flash-linear-attention (fla) | latest |
+| vllm (optional, for acceleration) | ≥ 0.17.0, < 0.18.0 |
+| einops (optional, for vLLM) | latest |
 
 ---