[OCP MX ] Add back emulation to available OCP MX backends list (#46629)
Signed-off-by: Felix Marty <Felix.Marty@amd.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
co-authored by
Andreas Karatzas
parent
4b643c463e
commit
03c6d01c30
@@ -104,11 +104,6 @@ def test_gpt_oss_attention_quantization(
|
||||
|
||||
model_args = EvaluationConfig(model_name).get_model_args(tp_size)
|
||||
|
||||
# Emulation backend on MI300, MI250 is opt-in
|
||||
# following https://github.com/vllm-project/vllm/pull/45896
|
||||
if not on_gfx950():
|
||||
model_args["moe_backend"] = "emulation"
|
||||
|
||||
extra_run_kwargs = {
|
||||
"gen_kwargs": {"max_gen_toks": 8000},
|
||||
"apply_chat_template": True,
|
||||
|
||||
@@ -221,10 +221,6 @@ class AccuracyTestConfig:
|
||||
if model_max_len is not None:
|
||||
model_args["max_model_len"] = model_max_len
|
||||
|
||||
# Emulation backend on MI300, MI250 is opt-in following https://github.com/vllm-project/vllm/pull/45896
|
||||
if not on_gfx950():
|
||||
model_args["moe_backend"] = "emulation"
|
||||
|
||||
return model_args
|
||||
|
||||
|
||||
|
||||
@@ -319,6 +319,7 @@ def _get_priority_backends_for_gpt_oss() -> list[Mxfp4MoeBackend]:
|
||||
Mxfp4MoeBackend.MARLIN,
|
||||
Mxfp4MoeBackend.BATCHED_MARLIN,
|
||||
Mxfp4MoeBackend.XPU,
|
||||
Mxfp4MoeBackend.EMULATION,
|
||||
]
|
||||
return _AVAILABLE_BACKENDS
|
||||
|
||||
@@ -554,8 +555,6 @@ def select_mxfp4_moe_backend(
|
||||
f"weight_key=kMxfp4Static, activation_key={activation_key}. "
|
||||
"Native backends require specific hardware. "
|
||||
"Set `VLLM_LOGGING_LEVEL=DEBUG` to see detailed unsupported reasons. "
|
||||
"To use the emulation backend for research/debugging, pass "
|
||||
"--moe-backend emulation."
|
||||
)
|
||||
|
||||
return Mxfp4MoeBackend.NONE, None
|
||||
|
||||
Reference in New Issue
Block a user