Use Flashinfer TRT-LLM as Llama 4 compatible MoE backend (#11928)
This commit is contained in:
@@ -643,6 +643,7 @@ class ModelConfig:
|
||||
"petit_nvfp4",
|
||||
]
|
||||
compatible_quantization_methods = {
|
||||
"modelopt_fp8": ["modelopt"],
|
||||
"modelopt_fp4": ["modelopt"],
|
||||
"petit_nvfp4": ["modelopt"],
|
||||
"w8a8_int8": ["compressed-tensors", "compressed_tensors"],
|
||||
|
||||
Reference in New Issue
Block a user