Skip to content

[Bugfix] Fix DeepSeek MTP crash when using TP1ModelRunner with CUDA graph due to shape mismatch #14237

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Mar 5, 2025
Merged

[Bugfix] Fix DeepSeek MTP crash when using TP1ModelRunner with CUDA graph due to shape mismatch #14237

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
637d366
When using DeepSeek MTP module with TP1 model
pyc96 Mar 4, 2025
c1e3a38
format
pyc96 Mar 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 9 additions & 4 deletions vllm/spec_decode/draft_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,7 +133,7 @@ def _gpu_advance_step(self, model_input: ModelRunnerInputBase,
def supports_gpu_multi_step(self, execute_model_req: ExecuteModelRequest):
"""Determines if draft_model_runner GPU multi-step can be used.
Currently required conditions are:
1. Only decodes
1. Only decodes
2. Only flash-attn
3. No LORA
4. No prompt_adapter_config
Expand Down Expand Up @@ -171,12 +171,12 @@ def execute_model(
num_steps: int = 1,
**kwargs,
) -> Optional[List[SamplerOutput]]:
"""Executes num_steps forward passes with advacement of input tensors
"""Executes num_steps forward passes with advacement of input tensors
on the GPU. Look at supports_gpu_multi_step(..) for pre-conditions.

Optimizations used:
1. Input tensors are updated on the GPU directly
2. Skips GPU=>CPU serialization of sampler outputs (we don't need
2. Skips GPU=>CPU serialization of sampler outputs (we don't need
them since we do batch expansion later that uses GPU outputs)
3. Reuses sampling tensors (since we run only decodes and they have
a repeating sampling logic)
Expand Down Expand Up @@ -302,7 +302,12 @@ def execute_model(
outputs.append(output)

if self.return_hidden_states and is_fallback:
output.hidden_states = hidden_states
if use_cuda_graph:
indices = model_input.sampling_metadata\
.selected_token_indices
output.hidden_states = hidden_states[:len(indices)]
else:
output.hidden_states = hidden_states

if model_input.attn_metadata.num_prefills == 0 \
and self.indices_of_seq_with_bonus_tokens is not None:
Expand Down