Skip to content

Nixl infra - #3769

Closed
zackyoray wants to merge 12 commits into
NVIDIA:mainfrom
zackyoray:nixl_infra
Closed

Nixl infra#3769
zackyoray wants to merge 12 commits into
NVIDIA:mainfrom
zackyoray:nixl_infra

Conversation

@zackyoray

@zackyoray zackyoray commented Apr 22, 2025

Copy link
Copy Markdown
Collaborator

Add nixl infrastructure to trtllm

[Split from https://github.com//pull/3677]


Comment thread cpp/tensorrt_llm/batch_manager/CMakeLists.txt Outdated
Comment thread cpp/tensorrt_llm/batch_manager/CMakeLists.txt Outdated
Comment thread cpp/CMakeLists.txt Outdated

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Is this the /opt/nvidia/nvda_nixl directory? I think it needs to be modified because the development docker image does not have write permissions for this directory by default.

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

its not running inside /opt/nvidia/nvda_nixl (it is running inside the cpp/build/nixl_build)
but it will install into /opt/nvidia/nvda_nixl

Comment thread cpp/CMakeLists.txt Outdated

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

We need a more general method to locate UCX, as the path in our default development Docker image is not /usr/local/ucx.

@zackyoray zackyoray Apr 27, 2025

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

Do you have any suggested method? using ENV var? CMAK var?
I added for now cmake -DUCX_PATH=/your/custom/ucx/path

Comment thread cpp/tensorrt_llm/batch_manager/cacheFormatter.cpp Outdated
Co-authored-by: Rabia Loulou <174243936+rabial-nv@users.noreply.github.com>
Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
zackyoray and others added 4 commits April 24, 2025 05:59
Tests for the NIXL can be found in the nixl_build directory:
- `{build_dir}/nixl_build/test/nixl/agent_example`
- `{build_dir}/nixl_build/test/nixl/nixl_test`

Co-authored-by: Rabia Loulou <174243936+rabial-nv@users.noreply.github.com>
Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
Co-authored-by: Rabia Loulou <174243936+rabial-nv@users.noreply.github.com>
Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
Add temp nixl code for testing compilation

Co-authored-by: Rabia Loulou <174243936+rabial-nv@users.noreply.github.com>
Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
Signed-off-by: Shixiaowei02 <39303645+Shixiaowei02@users.noreply.github.com>
@Shixiaowei02

Copy link
Copy Markdown
Collaborator

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3301 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3301 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #2299 completed with status: 'FAILURE'

Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3304 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3304 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #2302 completed with status: 'FAILURE'

# Find NIXL
find_path(NIXL_INCLUDE_DIR nixl.h HINTS ${NIXL_ROOT}/include)

# Find all required NIXL libraries

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Here and the script above are repeated, can we simply use find_package?

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

sure. will add commit that add find_package. if you prefer i can squash it into the existing one.

if(NOT WIN32)
set_target_properties(
${BATCH_MANAGER_STATIC_TARGET}
PROPERTIES BUILD_RPATH "/opt/nvidia/nvda_nixl/lib/x86_64-linux-gnu"

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

This still uses the absolute path /opt.

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

right thanks

@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3494 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3494 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #2459 completed with status: 'FAILURE'

Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3619 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3606 [ run ] completed with state ABORTED

@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3633 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3619 [ run ] completed with state ABORTED

@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3635 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3633 [ run ] completed with state ABORTED

@zackyoray
zackyoray force-pushed the nixl_infra branch 2 times, most recently from bd7a7d4 to e9d2df7 Compare April 28, 2025 17:33
@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3639 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3635 [ run ] completed with state ABORTED

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3639 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #2573 completed with status: 'FAILURE'

@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3728 [ run ] triggered by Bot

@juney-nvidia
juney-nvidia removed the request for review from a team April 29, 2025 12:07
Signed-off-by: Yoray Zack <62789610+zackyoray@users.noreply.github.com>
@zackyoray

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3731 [ run ] triggered by Bot

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3728 [ run ] completed with state ABORTED

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #3731 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #2641 completed with status: 'FAILURE'

@pcastonguay

Copy link
Copy Markdown
Collaborator

Is this PR still relevant? @zackyoray @Shixiaowei02 ? If not, can we close? Thx.

@zackyoray zackyoray closed this Jun 3, 2025
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.

4 participants