Skip to content

[fix] mismatch dim during capture graph if with --gpu-memory-utilization #40719

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

Closed
ir1ka wants to merge 2 commits into from
Closed

[fix] mismatch dim during capture graph if with --gpu-memory-utilization #40719

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a1df8b9
[fix] mismatch dim during capture graph if with --gpu-memory-utilization
ir1ka Apr 23, 2026
9b1997d
Add cp_kv_cache_interleave_size in may_reinitialize_input_batch to av…
ir1ka Apr 23, 2026
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
12 changes: 11 additions & 1 deletion vllm/v1/worker/gpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -658,19 +658,24 @@ def __init__(
placeholder_block_size = (
self.cache_config.block_size or CacheConfig.DEFAULT_BLOCK_SIZE
)
placeholder_max_num_blocks = None
placeholder_max_model_len = max(self.max_model_len, self.max_encoder_len)
self._init_block_sizes = [placeholder_block_size]
self._init_kernel_block_sizes = [placeholder_block_size]
self._init_max_num_blocks: list[int] | None = placeholder_max_num_blocks
self._init_max_model_len = placeholder_max_model_len
self.input_batch = InputBatch(
max_num_reqs=self.max_num_reqs,
# We need to use the encoder length for encoder-decoder
# because of KV cache for cross-attention.
max_model_len=max(self.max_model_len, self.max_encoder_len),
max_model_len=placeholder_max_model_len,
max_num_batched_tokens=self.max_num_tokens,
device=self.device,
vocab_size=self.model_config.get_vocab_size(),
block_sizes=[placeholder_block_size],
kernel_block_sizes=[placeholder_block_size],
num_spec_tokens=self.num_spec_tokens,
max_num_blocks_per_req=placeholder_max_num_blocks,
logitsprocs=build_logitsprocs(
self.vllm_config,
self.device,
Expand Down Expand Up @@ -6998,9 +7003,13 @@ def may_reinitialize_input_batch(
if (
block_sizes != self._init_block_sizes
or kernel_block_sizes != self._init_kernel_block_sizes
or max_num_blocks != self._init_max_num_blocks
or max_model_len != self._init_max_model_len
Comment thread
ir1ka marked this conversation as resolved.
):
self._init_block_sizes = block_sizes
self._init_kernel_block_sizes = kernel_block_sizes
self._init_max_num_blocks = max_num_blocks
self._init_max_model_len = max_model_len
self.input_batch = InputBatch(
max_num_reqs=self.max_num_reqs,
max_model_len=max_model_len,
Expand All @@ -7015,6 +7024,7 @@ def may_reinitialize_input_batch(
logitsprocs_need_output_token_ids=self.input_batch.logitsprocs_need_output_token_ids,
is_pooling_model=self.is_pooling_model,
reasoning_config=self.vllm_config.reasoning_config,
cp_kv_cache_interleave_size=self.parallel_config.cp_kv_cache_interleave_size,
)

assert self._init_block_sizes == block_sizes, (
Expand Down
Loading