minor

Signed-off-by: Kinjal Patel <[email protected]>

Author: kinjalpatel27

examples/vllm_serve/README.md

@@ -89,3 +89,4 @@ torch.distributed.barrier()
 
 1. AWQ is not yet supported in vLLM.
 2. PTQ/QAT checkpoint doesn't work with KV Cache quantization enabled.
+3. Mixed precision checkpoint doesn't work currently.

modelopt/torch/export/unified_export_hf.py

@@ -582,6 +582,8 @@ def export_hf_checkpoint(
         dtype: the weights data type to export the unquantized layers or the default model data type if None.
         export_dir: the target export path.
         save_modelopt_state: whether to save the modelopt state_dict.
+        export_bf16_weights_amax: whether to export the bf16 weights and amax values separately. This can be used for
+                                  vLLM fakequant serving.
     """
     export_dir = Path(export_dir)
     export_dir.mkdir(parents=True, exist_ok=True)