EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Model] Add qwen3Next support in Main (#4596)

Browse Source 
### What this PR does / why we need it?
Add Qwen3Next support in main

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

---------

Signed-off-by: SunnyLee219 <3294305115@qq.com>
This commit is contained in:
LeeWenquan
2025-12-03 14:17:37 +08:00
committed by GitHub
parent 3f81c4bb25
commit 38bd95229f
4 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.github/workflows/_e2e_test.yaml vendored
View File

@@ -286,4 +286,4 @@ jobs:
VLLM_USE_MODELSCOPE: True VLLM_USE_MODELSCOPE: True
run: | run: |
. /usr/local/Ascend/ascend-toolkit/8.3.RC2/bisheng_toolkit/set_env.sh . /usr/local/Ascend/ascend-toolkit/8.3.RC2/bisheng_toolkit/set_env.sh
#pytest -sv tests/e2e/multicard/test_qwen3_next.py pytest -sv tests/e2e/multicard/test_qwen3_next.py

3
tests/e2e/multicard/test_qwen3_next.py
View File

@@ -24,6 +24,7 @@ Run `pytest tests/e2e/multicard/test_qwen3_next.py`.
import os import os
from unittest.mock import patch from unittest.mock import patch
import pytest
from modelscope import snapshot_download # type: ignore from modelscope import snapshot_download # type: ignore
from tests.e2e.conftest import VllmRunner from tests.e2e.conftest import VllmRunner
@@ -63,6 +64,7 @@ def test_models_distributed_Qwen3_NEXT_TP4_FULL_DECODE_ONLY():
del vllm_model del vllm_model
@pytest.mark.skip
def test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY(): def test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY():
example_prompts = [ example_prompts = [
"Hello, my name is", "Hello, my name is",
@@ -113,6 +115,7 @@ def test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY():
# TODO: will conduct accuracy verification after the subsequent version becomes stable # TODO: will conduct accuracy verification after the subsequent version becomes stable
@pytest.mark.skip
@patch.dict(os.environ, {"HCCL_BUFFSIZE": "1024"}) @patch.dict(os.environ, {"HCCL_BUFFSIZE": "1024"})
def test_models_distributed_Qwen3_NEXT_W8A8DYNAMIC_WITH_EP(): def test_models_distributed_Qwen3_NEXT_W8A8DYNAMIC_WITH_EP():
example_prompts = [ example_prompts = [

3
vllm_ascend/ops/triton/mamba/casual_conv1d.py
View File

@@ -7,7 +7,7 @@
# and https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/mamba/ops/causal_conv1d.py # and https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/mamba/ops/causal_conv1d.py
# mypy: ignore-errors # mypy: ignore-errors
from typing import Optional, Union from typing import Any, Optional, Union
import torch import torch
import torch.nn.functional as F import torch.nn.functional as F
@@ -72,6 +72,7 @@ def causal_conv1d_fn(
conv_states: Optional[torch.Tensor] = None, conv_states: Optional[torch.Tensor] = None,
activation: Optional[str] = "silu", activation: Optional[str] = "silu",
pad_slot_id: int = PAD_SLOT_ID, pad_slot_id: int = PAD_SLOT_ID,
metadata: Optional[Any] = None,
): ):
""" """
x: (batch, dim, seqlen) or (dim,cu_seq_len) for varlen x: (batch, dim, seqlen) or (dim,cu_seq_len) for varlen

2
vllm_ascend/patch/worker/patch_triton.py
View File

@@ -11,4 +11,4 @@ vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_update = causal
vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_fn = causal_conv1d_fn vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_fn = causal_conv1d_fn
vllm.model_executor.layers.fla.ops.fused_recurrent.fused_recurrent_gated_delta_rule_fwd_kernel = fused_recurrent_gated_delta_rule_fwd_kernel vllm.model_executor.layers.fla.ops.fused_recurrent.fused_recurrent_gated_delta_rule_fwd_kernel = fused_recurrent_gated_delta_rule_fwd_kernel
vllm.model_executor.layers.fla.ops.layernorm_guard.LayerNormFn = LayerNormFn vllm.model_executor.layers.fla.ops.layernorm_guard.LayerNormFn = LayerNormFn
vllm.model_executor.layers.fla.ops.chunk.chunk_gated_delta_rule = chunk_gated_delta_rule vllm.model_executor.layers.fla.ops.chunk_gated_delta_rule = chunk_gated_delta_rule