config.py: Clarify that only local GGUF checkpoints are supported. (vllm-project#18623)

Signed-off-by: Mathieu Bordere <[email protected]>

Author: MathieuBordere

vllm/model_executor/layers/fused_moe/layer.py

@@ -41,6 +41,7 @@
         from .pplx_prepare_finalize import PplxPrepareAndFinalize
 else:
     fused_experts = None  # type: ignore
+    FusedMoEPermuteExpertsUnpermute = None  # type: ignore
     FusedMoEPrepareAndFinalize = None  # type: ignore
 if is_rocm_aiter_moe_enabled():
     from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (  # noqa: E501

vllm/transformers_utils/config.py

@@ -300,7 +300,10 @@ def get_config(
                 "   - For Hugging Face models: ensure the presence of a "
                 "'config.json'.\n"
                 "   - For Mistral models: ensure the presence of a "
-                "'params.json'.\n").format(model=model)
+                "'params.json'.\n"
+                "3. For GGUF: pass the local path of the GGUF checkpoint.\n"
+                "   Loading GGUF from a remote repo directly is not yet "
+                "supported.\n").format(model=model)
 
             raise ValueError(error_message) from e