[Mini-vLLM] SM86-only cut: models, backends, Rust, tests

- Keep only Qwen3.5, Qwen3.5-MTP and Qwen4Exp model paths; trim the model registry to 41 entries
- Delete idle attention backends (MLA, FlashInfer, ROCm-only, quant backends) and dead GDN CuTeDSL kernels
- Remove inc/quark quant loaders; keep compressed-tensors and Marlin
- Drop the Rust frontend, pooling entrypoints, and idle test suites
- Rebuild kernels for sm_80;sm_86 and strip the extensions
- Verified: full import sweep clean (optional deps only) and a dummy-load Qwen3.5 engine run on RTX 3090 via the Triton GDN path

Co-authored-by: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
2026-10-18 19:19:56 +07:00
co-authored by Hinata
parent c5ab42bc0f
commit e61dcb9143
2728 changed files with 159 additions and 950401 deletions
@@ -19,7 +19,7 @@ from vllm.model_executor.kernels.linear import (
init_mxfp8_linear_kernel,
init_nvfp4_linear_kernel,
)
from vllm.model_executor.layers.attention import Attention, MLAAttention
from vllm.model_executor.layers.attention import Attention
from vllm.model_executor.layers.fused_moe import (
FusedMoEConfig,
FusedMoEMethodBase,
@@ -228,6 +228,8 @@ class ModelOptQuantConfigBase(QuantizationConfig):
self, layer: torch.nn.Module, prefix: str
) -> "QuantizeMethodBase | None":
# handle kv-cache first so we can focus only on weight quantization thereafter
from vllm.model_executor.layers.attention.mla_attention import MLAAttention
if isinstance(layer, (Attention, MLAAttention)):
return self.KVCacheMethodCls(self)