vllm.models.mistral.mistral_large4.amd.specialized_decode.runtime.spec
¶
Side-effect-free Mistral Large 4 specialized-decode configuration.
Classes:
Functions:
-
decode_backend–The requested decode backend: ``--additional-config
-
specialized_decode_config–The deployment's specialized decode, None for vLLM's decode. Raises with
DecodeConfig
dataclass
¶
Methods:
-
refusal_reasons–Default-deny: every deployment feature the kernels do not serve.
Source code in vllm/models/mistral/mistral_large4/amd/specialized_decode/runtime/spec.py
refusal_reasons(config)
¶
Default-deny: every deployment feature the kernels do not serve.
Source code in vllm/models/mistral/mistral_large4/amd/specialized_decode/runtime/spec.py
decode_backend(config)
¶
The requested decode backend: --additional-config
'{"decode_backend": "latency_specialized"}', or --monokernel where the
kernel config has it.
Source code in vllm/models/mistral/mistral_large4/amd/specialized_decode/runtime/spec.py
specialized_decode_config(config)
¶
The deployment's specialized decode, None for vLLM's decode. Raises with every reason when it is requested for an unsupported deployment.