Skip to content
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.

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

TensorRT-LLM backend bump to latest version + misc fixes #2791

Merged
merged 30 commits into from
Dec 13, 2024

Conversation

mfuntowicz
Copy link
Member

This PR bumps some dependencies related to TensorRT-LLM alongside rebasing Docker container against ubuntu24.04 instead of ubuntu22.04.

To support this, we need to use latest TensorRT-LLM main due to a missing import in one of the file on Nvidia side leading to gcc failure for gcc > 12 (more info here).

It also reimplement the backbone of the backend for simplicity and alow easier testing which is actively WIP as for now but will land in its own PR as it will involve some additional changes on GA side.

@mfuntowicz mfuntowicz requested a review from Hugoch December 1, 2024 23:18
* chore: Add doc and CI for TRTLLM

* chore: Add doc and CI for TRTLLM

* chore: Add doc and CI for TRTLLM

* chore: Add doc and CI for TRTLLM

* doc: Formatting
@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@mfuntowicz mfuntowicz marked this pull request as ready for review December 11, 2024 09:28
Hugoch
Hugoch previously approved these changes Dec 13, 2024
Copy link
Member

@Hugoch Hugoch left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's gooo 🚀 !!!!!!

// Define some configuration variables
executor_config.setKvCacheConfig(tle::KvCacheConfig(true));
executor_config.setEnableChunkedContext(compute_capabilities.is_at_least_ampere());
executor_config.setSchedulerConfig(tle::SchedulerConfig(tle::CapacitySchedulerPolicy::kMAX_UTILIZATION));
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Probably something we would need to have as a variable somewhere (but we need to bench the impacts first).

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What about we expose it through the CLI of the launcher mb? This way it's both CLI and ENV overridable

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah but let's make sure that it's an advanced setting (I don't users to be confused by all the CLI options)

backends/trtllm/csrc/backend.hpp Outdated Show resolved Hide resolved
@mfuntowicz mfuntowicz merged commit ea7f408 into main Dec 13, 2024
9 of 13 checks passed
@mfuntowicz mfuntowicz deleted the trtllm/cancellation branch December 13, 2024 14:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants