Remove VLLM_TEST_FORCE_FP8_MARLIN to replace with linear_backend/moe_backend (#52182)

Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Misha Goin
2026-08-18 12:58:04 -04:00
committed by GitHub
co-authored by OpenAI Codex
parent 88b2bff2c6
commit bca7bea240
15 changed files with 46 additions and 66 deletions
@@ -370,13 +370,6 @@ def select_fp8_moe_backend(
backend, config, weight_key, activation_key, activation_format
)
# Handle explicit MARLIN FP8 configuration.
if envs.VLLM_TEST_FORCE_FP8_MARLIN:
backend = Fp8MoeBackend.MARLIN
return _return_or_raise(
backend, config, weight_key, activation_key, activation_format
)
# Handle explicit AITER FP8 configuration.
if envs.is_set("VLLM_ROCM_USE_AITER") or envs.is_set("VLLM_ROCM_USE_AITER_MOE"):
skip_aiter_moe = (
@@ -4,7 +4,6 @@ from enum import Enum
import torch
import vllm.envs as envs
import vllm.model_executor.layers.fused_moe.modular_kernel as mk
from vllm.config.kernel import MoEBackend
from vllm.logger import init_logger
@@ -271,12 +270,6 @@ def select_nvfp4_moe_backend(
requested_backend, config, weight_key, activation_key, activation_format
)
if envs.VLLM_TEST_FORCE_FP8_MARLIN:
backend = NvFp4MoeBackend.MARLIN
return _return_or_raise(
backend, config, weight_key, activation_key, activation_format
)
# Select kernels in order of backend.
for backend in AVAILABLE_BACKENDS:
for k_cls in backend_to_kernel_cls(backend):