Skip to content

[None][fix] Remove unnecessary attention workspace memory check #9064

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
Nov 12, 2025
Merged

[None][fix] Remove unnecessary attention workspace memory check #9064

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ab18908
remove unnecessary attention workspace memory check
jiaganc Nov 11, 2025
82d7cd5
use trace log instead of warning log
jiaganc Nov 11, 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
11 changes: 1 addition & 10 deletions cpp/tensorrt_llm/thop/attentionOp.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -816,16 +816,6 @@ void attention(torch::Tensor q, std::optional<torch::Tensor> k, std::optional<to
= runner->getWorkspaceSize(*op, num_tokens, max_attention_window_size, num_gen_tokens, max_blocks_per_sequence);
TLLM_LOG_TRACE("Expected workspace size is %ld bytes", workspace_size);

if (workspace_size >= (16l << 30))
{
auto const [free_mem, total_mem] = tensorrt_llm::common::getDeviceMemoryInfo(false);
if (workspace_size >= static_cast<int64_t const>(free_mem))
{
throw std::runtime_error("attention workspace size " + std::to_string(workspace_size)
+ " bytes, exceeds available CUDA memory " + std::to_string(free_mem) + " bytes");
}
}

torch::Tensor workspace;
if (workspace_.has_value())
{
Expand All @@ -839,6 +829,7 @@ void attention(torch::Tensor q, std::optional<torch::Tensor> k, std::optional<to
}
else
{
TLLM_LOG_TRACE("Allocate new attention workspace with size %ld bytes", workspace_size);
workspace = torch::empty({workspace_size}, torch::dtype(torch::kByte).device(qkv_or_q.device()));
}

Expand Down