[Bugfix] Qwen3Next support FlashComm1 (#6830)
### What this PR does / why we need it?
Support FlashComm1 for Qwen3-Next. Fix some padding problems in Sequence
Parallel (SP)
and resolve precision problems in shared_out when both FlashComm1 is
enabled.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
CI
- vLLM version: v0.15.0
- vLLM main:
83b47f67b1
---------
Signed-off-by: zhaojiangjiang <zhaojiangjiang1@h-partners.com>
Co-authored-by: zhaojiangjiang <zhaojiangjiang1@h-partners.com>
This commit is contained in:
@@ -73,3 +73,38 @@ def test_qwen3_next_w8a8dynamic_distributed_tp4_ep():
|
||||
quantization="ascend",
|
||||
) as vllm_model:
|
||||
vllm_model.generate_greedy(example_prompts, max_tokens)
|
||||
|
||||
|
||||
@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_FLASHCOMM1": "1"})
|
||||
@patch.dict(os.environ, {"HCCL_BUFFSIZE": "1024"})
|
||||
def test_qwen3_next_distributed_mp_flash_comm_tp4():
|
||||
example_prompts = [
|
||||
"Hello, my name is",
|
||||
] * 4
|
||||
max_tokens = 5
|
||||
with VllmRunner("Qwen/Qwen3-Next-80B-A3B-Instruct",
|
||||
tensor_parallel_size=4,
|
||||
max_model_len=4096,
|
||||
gpu_memory_utilization=0.7,
|
||||
distributed_executor_backend="mp",
|
||||
enable_expert_parallel=True,
|
||||
enforce_eager=True) as vllm_model:
|
||||
vllm_model.generate_greedy(example_prompts, max_tokens)
|
||||
del vllm_model
|
||||
|
||||
|
||||
@patch.dict(os.environ, {"HCCL_BUFFSIZE": "1024"})
|
||||
def test_qwen3_next_distributed_mp_graph_mode_tp4():
|
||||
example_prompts = [
|
||||
"Hello, my name is",
|
||||
] * 4
|
||||
max_tokens = 5
|
||||
with VllmRunner("Qwen/Qwen3-Next-80B-A3B-Instruct",
|
||||
tensor_parallel_size=4,
|
||||
max_model_len=4096,
|
||||
gpu_memory_utilization=0.7,
|
||||
distributed_executor_backend="mp",
|
||||
enable_expert_parallel=True,
|
||||
enforce_eager=False) as vllm_model:
|
||||
vllm_model.generate_greedy(example_prompts, max_tokens)
|
||||
del vllm_model
|
||||
Reference in New Issue
Block a user