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

[BugFix]Fix mtp torchair bug caused by #2719 (#3566)

Browse Source 
### What this PR does / why we need it?
Fix mtp tochair bug cuased by #2719
Since FIA need extra space for padding, we need to enforce
`self.max_num_seqs > self.scheduler_config.max_num_seqs` in KV consumer
+ MTP
This means that, `self.max_num_seqs` **>** the actual maximum requests
(`self.scheduler_config.max_num_seqs`)

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

### How was this patch tested?


- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: xuyexiong <xuyexiong@huawei.com>
This commit is contained in:
xuyexiong
2025-10-21 22:21:44 +08:00
committed by GitHub
parent 534f32d27c
commit 79821106e6
2 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm_ascend/torchair/torchair_model_runner.py
View File

@@ -467,7 +467,7 @@ class NPUTorchairModelRunner(NPUModelRunner):
# pd disaggregation scenario may incorrectly calculate the batch in mtp scenario, so we force set it to max_num_reqs # pd disaggregation scenario may incorrectly calculate the batch in mtp scenario, so we force set it to max_num_reqs
self.torchair_graph_batch_sizes = [self.max_num_reqs] self.torchair_graph_batch_sizes = [self.max_num_reqs]
logger.warning( logger.warning(
"is kv_consumer, torch_graph_batch_sizes sets to [max_num_seqs]" f"is kv_consumer, torch_graph_batch_sizes sets to [max_num_seqs] {[self.max_num_reqs]}"
) )
elif len(self.torchair_graph_batch_sizes) == 0: elif len(self.torchair_graph_batch_sizes) == 0:
self.torchair_graph_batch_sizes = [1, self.max_num_reqs] self.torchair_graph_batch_sizes = [1, self.max_num_reqs]
@@ -517,11 +517,16 @@ class NPUTorchairModelRunner(NPUModelRunner):
f"torchair_graph_batch_sizes {cur_graph_batch_size} is bigger than max_num_batched_tokens", f"torchair_graph_batch_sizes {cur_graph_batch_size} is bigger than max_num_batched_tokens",
f"{self.scheduler_config.max_num_batched_tokens} will skip this batch size." f"{self.scheduler_config.max_num_batched_tokens} will skip this batch size."
) )
new_max_num_reqs = max(new_graph_batch_sizes) new_max_num_reqs = math.ceil(
max(new_graph_batch_sizes) / self.decode_token_per_req)
if self.max_num_reqs != new_max_num_reqs: if self.max_num_reqs != new_max_num_reqs:
logger.warning(f"max_num_reqs is updated to {new_max_num_reqs}") logger.warning(f"max_num_reqs is updated to {new_max_num_reqs}")
self.max_num_reqs = new_max_num_reqs self.max_num_reqs = new_max_num_reqs
self.scheduler_config.max_num_seqs = new_max_num_reqs if not (self.decode_token_per_req > 1 and self.is_kv_consumer):
# Do not update scheduler_config.max_num_seqs in KV consumer + MTP
# Since FIA need extra space for padding
# Enforce self.max_num_seqs > self.scheduler_config.max_num_seqs in KV consumer + MTP
self.scheduler_config.max_num_seqs = new_max_num_reqs
if new_graph_batch_sizes != self.torchair_graph_batch_sizes: if new_graph_batch_sizes != self.torchair_graph_batch_sizes:
logger.warning( logger.warning(

2
vllm_ascend/worker/model_runner_v1.py
View File

@@ -2390,7 +2390,7 @@ class NPUModelRunner(LoRAModelRunnerMixin):
# for dummy run with LoRA so that the num_reqs collectively # for dummy run with LoRA so that the num_reqs collectively
# has num_tokens in total. # has num_tokens in total.
assert num_tokens <= self.scheduler_config.max_num_batched_tokens assert num_tokens <= self.scheduler_config.max_num_batched_tokens
max_num_reqs = self.scheduler_config.max_num_seqs max_num_reqs = self.max_num_reqs
if uniform_decode: if uniform_decode:
num_reqs = cdiv(num_tokens, max_query_len) num_reqs = cdiv(num_tokens, max_query_len)
num_scheduled_tokens_list = [max_query_len] * num_reqs num_scheduled_tokens_list = [max_query_len] * num_reqs