[Bugfix] rename enable_flash_comm_v1 back to enable_sp (#6883)

### What this PR does / why we need it?

PR #5632 introduced a bug by replacing some branches gated by enable_sp
with enable_flash_comm_v1. As a result, when enable_shared_expert_dp is
enabled alone (i.e., VLLM_ASCEND_ENABLE_FLASHCOMM1=0 and
VLLM_ASCEND_ENABLE_FLASHCOMM=0), the behavior becomes inconsistent with
the previous logic and leads to accuracy issues. This PR restores the
original enable_sp-based branching to recover expected behavior and
accuracy.

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

No

### How was this patch tested?

#### 1. start server
``` bash
vllm serve /home/weights/DeepSeek-V2-Lite-W8A8/  \
    --port 8001 \
    --served-model-name auto \
    --max-model-len 1024 \
    --enforce-eager \
    --tensor-parallel-size 2 \
    --data-parallel-size 2 \
    --gpu-memory-utilization 0.9 \
    --enable-expert-parallel \
    --additional-config '{"enable_shared_expert_dp": true}'
```

#### 2. curl
```bash
curl -s http://localhost:8001/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
  "model": "auto",
  "messages": [
    {"role": "user", "content": "Hello. I have a question. Who are you?"}
  ],
  "max_tokens": 10,
  "temperature": 0.0,
  "ignore_eos_token": true
}'
```

- vLLM version: v0.16.0
- vLLM main:
15d76f74e2

Signed-off-by: realliujiaxu <realliujiaxu@163.com>
This commit is contained in:
realliujiaxu
2026-03-01 20:22:50 +08:00
committed by GitHub
parent 8835236181
commit 5e24b26a54
7 changed files with 24 additions and 29 deletions

View File

@@ -12,7 +12,7 @@ import vllm_ascend.envs as envs_ascend
from vllm_ascend.ascend_config import get_ascend_config from vllm_ascend.ascend_config import get_ascend_config
from vllm_ascend.utils import ( from vllm_ascend.utils import (
AscendDeviceType, AscendDeviceType,
enable_flash_comm_v1, enable_sp,
flashcomm2_enable, flashcomm2_enable,
get_ascend_device_type, get_ascend_device_type,
has_layer_idx, has_layer_idx,
@@ -92,14 +92,14 @@ def set_ascend_forward_context(
# main model and drafter model may have different architecture # main model and drafter model may have different architecture
is_context_moe_model = is_drafter_moe_model(vllm_config) if is_draft_model else is_moe_model(vllm_config) is_context_moe_model = is_drafter_moe_model(vllm_config) if is_draft_model else is_moe_model(vllm_config)
if is_context_moe_model: if is_context_moe_model:
flash_comm_v1_enabled = enable_flash_comm_v1() and num_tokens is not None flash_comm_v1_enabled = enable_sp(vllm_config) and num_tokens is not None
mmrs_fusion = False mmrs_fusion = False
elif is_draft_model: elif is_draft_model:
# TODO: for dense drafter, `sp` is redundant and is not compatible with `dp` and `graph`. # TODO: for dense drafter, `sp` is redundant and is not compatible with `dp` and `graph`.
# Disable it to avoid more problems. # Disable it to avoid more problems.
flash_comm_v1_enabled = False flash_comm_v1_enabled = False
else: else:
flash_comm_v1_enabled = enable_flash_comm_v1() and num_tokens is not None and num_tokens > 1000 flash_comm_v1_enabled = enable_sp(vllm_config) and num_tokens is not None and num_tokens > 1000
forward_context.mmrs_fusion = mmrs_fusion forward_context.mmrs_fusion = mmrs_fusion
forward_context.num_tokens = num_tokens forward_context.num_tokens = num_tokens
forward_context.flash_comm_v1_enabled = flash_comm_v1_enabled forward_context.flash_comm_v1_enabled = flash_comm_v1_enabled

View File

@@ -40,7 +40,7 @@ from vllm.model_executor.layers.quantization.base_config import QuantizationConf
from vllm.model_executor.utils import set_weight_attrs from vllm.model_executor.utils import set_weight_attrs
from vllm_ascend.ops.linear_op import get_parallel_op, get_replicated_op from vllm_ascend.ops.linear_op import get_parallel_op, get_replicated_op
from vllm_ascend.utils import enable_flash_comm_v1, maybe_trans_nz from vllm_ascend.utils import enable_sp, maybe_trans_nz
class AscendUnquantizedLinearMethod(UnquantizedLinearMethod): class AscendUnquantizedLinearMethod(UnquantizedLinearMethod):
@@ -240,7 +240,7 @@ class AscendRowParallelLinear(RowParallelLinear):
disable_tp: bool = False, disable_tp: bool = False,
): ):
# TODO(kunpengW-code): Specifying the prefix in linear layers of some models in the vLLM. # TODO(kunpengW-code): Specifying the prefix in linear layers of some models in the vLLM.
if enable_flash_comm_v1(): if enable_sp():
compilation_config = get_current_vllm_config().compilation_config compilation_config = get_current_vllm_config().compilation_config
unique_prefix = prefix unique_prefix = prefix
if prefix in compilation_config.static_forward_context: if prefix in compilation_config.static_forward_context:

View File

@@ -70,7 +70,7 @@ from vllm_ascend.ops.flashcomm2_oshard_manager import flashcomm2_oshard_manager
from vllm_ascend.utils import ( from vllm_ascend.utils import (
enable_dsa_cp, enable_dsa_cp,
enable_dsa_cp_with_layer_shard, enable_dsa_cp_with_layer_shard,
enable_flash_comm_v1, enable_sp,
flashcomm2_enable, flashcomm2_enable,
get_flashcomm2_reorgnized_batch_ids, get_flashcomm2_reorgnized_batch_ids,
get_weight_prefetch_method, get_weight_prefetch_method,
@@ -466,7 +466,7 @@ class Flashcomm2OshardQKVParallelOp(CustomColumnParallelOp):
# Matrix multiply. # Matrix multiply.
assert self.quant_method is not None assert self.quant_method is not None
if enable_flash_comm_v1(): if enable_sp():
input_ = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(input_, True) input_ = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(input_, True)
# Trigger async broadcast before matmul to overlap communication. # Trigger async broadcast before matmul to overlap communication.
@@ -649,7 +649,7 @@ def _get_column_parallel_op(
if flashcomm2_oshard_manager.flashcomm2_oshard_enable(): if flashcomm2_oshard_manager.flashcomm2_oshard_enable():
if any(p in prefix for p in ("qkv_proj", "conv1d", "query_key_value")): if any(p in prefix for p in ("qkv_proj", "conv1d", "query_key_value")):
return Flashcomm2OshardQKVParallelOp(layer) return Flashcomm2OshardQKVParallelOp(layer)
if enable_flash_comm_v1(): if enable_sp():
if "shared_expert" in prefix: if "shared_expert" in prefix:
return None return None
sp_column_prefix = [ sp_column_prefix = [
@@ -688,7 +688,7 @@ def _get_row_parallel_op(
if flashcomm2_enable(): if flashcomm2_enable():
if "o_proj" in prefix or "out_proj" in prefix: if "o_proj" in prefix or "out_proj" in prefix:
return Flashcomm2OProjRowParallelOp(layer) return Flashcomm2OProjRowParallelOp(layer)
if enable_flash_comm_v1(): if enable_sp():
if "shared_expert" in prefix: if "shared_expert" in prefix:
return None return None
sp_row_prefixes = [ sp_row_prefixes = [

View File

@@ -39,7 +39,6 @@ from vllm_ascend.utils import (
COMPRESSED_TENSORS_METHOD, COMPRESSED_TENSORS_METHOD,
AscendDeviceType, AscendDeviceType,
check_kv_extra_config, check_kv_extra_config,
enable_sp,
flashcomm2_enable, flashcomm2_enable,
get_ascend_device_type, get_ascend_device_type,
is_moe_model, is_moe_model,
@@ -48,7 +47,7 @@ from vllm_ascend.utils import (
update_aclgraph_sizes, update_aclgraph_sizes,
update_cudagraph_capture_sizes, update_cudagraph_capture_sizes,
is_310p, is_310p,
enable_flash_comm_v1, enable_sp,
) )
if TYPE_CHECKING: if TYPE_CHECKING:
@@ -402,7 +401,7 @@ class NPUPlatform(Platform):
) )
vllm_config.parallel_config.cp_kv_cache_interleave_size = cache_config.block_size vllm_config.parallel_config.cp_kv_cache_interleave_size = cache_config.block_size
if enable_flash_comm_v1(): if enable_sp(vllm_config):
assert not is_vl_model(vllm_config), """Flash Comm V1 is not supported for VL models. \ assert not is_vl_model(vllm_config), """Flash Comm V1 is not supported for VL models. \
Please disable it by setting VLLM_ASCEND_ENABLE_FLASHCOMM1=0. \ Please disable it by setting VLLM_ASCEND_ENABLE_FLASHCOMM1=0. \
For optimal performance with VL models, we recommend enabling Sequence Parallelism \ For optimal performance with VL models, we recommend enabling Sequence Parallelism \

View File

@@ -719,15 +719,6 @@ def matmul_allreduce_enable() -> bool:
return envs_ascend.VLLM_ASCEND_ENABLE_MATMUL_ALLREDUCE return envs_ascend.VLLM_ASCEND_ENABLE_MATMUL_ALLREDUCE
def enable_flash_comm_v1():
return (
envs_ascend.VLLM_ASCEND_ENABLE_FLASHCOMM1
# Flash comm 1 should be enabled by env VLLM_ASCEND_ENABLE_FLASHCOMM1
# We retain the env VLLM_ASCEND_ENABLE_FLASHCOMM here for backward compatibility.
or bool(int(os.getenv("VLLM_ASCEND_ENABLE_FLASHCOMM", "0")))
)
def enable_sp_by_pass(vllm_config: VllmConfig): def enable_sp_by_pass(vllm_config: VllmConfig):
return not vllm_config.model_config.enforce_eager and vllm_config.compilation_config.pass_config.enable_sp return not vllm_config.model_config.enforce_eager and vllm_config.compilation_config.pass_config.enable_sp
@@ -739,7 +730,12 @@ def enable_sp(vllm_config=None, enable_shared_expert_dp: bool = False) -> bool:
from vllm.config import get_current_vllm_config from vllm.config import get_current_vllm_config
vllm_config = get_current_vllm_config() vllm_config = get_current_vllm_config()
_ENABLE_SP = enable_sp_by_pass(vllm_config) or enable_flash_comm_v1() _ENABLE_SP = (
envs_ascend.VLLM_ASCEND_ENABLE_FLASHCOMM1
# Flash comm 1 should be enabled by env VLLM_ASCEND_ENABLE_FLASHCOMM1
# We retain the env VLLM_ASCEND_ENABLE_FLASHCOMM here for backward compatibility.
or bool(int(os.getenv("VLLM_ASCEND_ENABLE_FLASHCOMM", "0")))
)
if not _ENABLE_SP and enable_shared_expert_dp: if not _ENABLE_SP and enable_shared_expert_dp:
_ENABLE_SP = True _ENABLE_SP = True
@@ -1104,7 +1100,7 @@ def enable_dsa_cp() -> bool:
is_ds_v32 = hasattr(vllm_config.model_config, "hf_text_config") and hasattr( is_ds_v32 = hasattr(vllm_config.model_config, "hf_text_config") and hasattr(
vllm_config.model_config.hf_text_config, "index_topk" vllm_config.model_config.hf_text_config, "index_topk"
) )
return bool(is_ds_v32 and enable_flash_comm_v1()) return bool(is_ds_v32 and enable_sp())
@lru_cache(maxsize=1) @lru_cache(maxsize=1)

View File

@@ -113,8 +113,8 @@ from vllm_ascend.spec_decode.medusa_proposer import MedusaProposer
from vllm_ascend.spec_decode.mtp_proposer import MtpProposer from vllm_ascend.spec_decode.mtp_proposer import MtpProposer
from vllm_ascend.utils import ( from vllm_ascend.utils import (
check_gdn_layer, check_gdn_layer,
enable_flash_comm_v1,
enable_sp, enable_sp,
enable_sp_by_pass,
is_drafter_moe_model, is_drafter_moe_model,
is_moe_model, is_moe_model,
lmhead_tp_enable, lmhead_tp_enable,
@@ -1745,7 +1745,7 @@ class NPUModelRunner(GPUModelRunner):
# Pad tokens to multiple of tensor_parallel_size when # Pad tokens to multiple of tensor_parallel_size when
# enabled collective fusion for SP # enabled collective fusion for SP
tp_size = self.vllm_config.parallel_config.tensor_parallel_size tp_size = self.vllm_config.parallel_config.tensor_parallel_size
if enable_sp(self.vllm_config): if enable_sp(self.vllm_config) or enable_sp_by_pass(self.vllm_config):
return round_up(num_scheduled_tokens, tp_size) return round_up(num_scheduled_tokens, tp_size)
return num_scheduled_tokens return num_scheduled_tokens
@@ -2300,7 +2300,7 @@ class NPUModelRunner(GPUModelRunner):
# tp_size; otherwise, on non-first PP ranks it would effectively perform an extra all-gather, leading # tp_size; otherwise, on non-first PP ranks it would effectively perform an extra all-gather, leading
# to incorrect memory estimation and potentially causing OOM. # to incorrect memory estimation and potentially causing OOM.
intermediate_tokens = num_tokens_padded intermediate_tokens = num_tokens_padded
if enable_flash_comm_v1(): if enable_sp():
tp_size = get_tensor_model_parallel_world_size() tp_size = get_tensor_model_parallel_world_size()
intermediate_tokens = (num_tokens_padded + tp_size - 1) // tp_size intermediate_tokens = (num_tokens_padded + tp_size - 1) // tp_size
if self.intermediate_tensors is None: if self.intermediate_tensors is None:

View File

@@ -55,7 +55,7 @@ from vllm_ascend.ops.triton.triton_utils import init_device_properties_triton
from vllm_ascend.utils import ( from vllm_ascend.utils import (
AscendDeviceType, AscendDeviceType,
check_ascend_device_type, check_ascend_device_type,
enable_flash_comm_v1, enable_sp,
get_ascend_device_type, get_ascend_device_type,
register_ascend_customop, register_ascend_customop,
) )
@@ -376,7 +376,7 @@ class NPUWorker(WorkerBase):
if forward_pass and not get_pp_group().is_first_rank: if forward_pass and not get_pp_group().is_first_rank:
# If flashcomm1 is used, this all_gather_group parameter needs to be removed, otherwise # If flashcomm1 is used, this all_gather_group parameter needs to be removed, otherwise
# it will conflict with the all-gather operation in flashcomm1. # it will conflict with the all-gather operation in flashcomm1.
if enable_flash_comm_v1(): if enable_sp():
all_gather_group = None all_gather_group = None
else: else:
all_gather_group = get_tp_group() all_gather_group = get_tp_group()
@@ -393,7 +393,7 @@ class NPUWorker(WorkerBase):
assert parallel_config.distributed_executor_backend != ("external_launcher") and not get_pp_group().is_last_rank assert parallel_config.distributed_executor_backend != ("external_launcher") and not get_pp_group().is_last_rank
# If flashcomm1 is used, this all_gather_group parameter needs to be removed, otherwise # If flashcomm1 is used, this all_gather_group parameter needs to be removed, otherwise
# it will conflict with the all-gather operation in flashcomm1. # it will conflict with the all-gather operation in flashcomm1.
if enable_flash_comm_v1(): if enable_sp():
all_gather_group = None all_gather_group = None
else: else:
all_gather_group = get_tp_group() all_gather_group = get_tp_group()