[nvbugs/5369799] fix: Update disaggregation handling in sampler #5762
Conversation
|
/bot run |
|
PR_Github #11013 [ run ] triggered by Bot |
|
PR_Github #11013 [ run ] completed with state |
stnie
force-pushed
the
bugfix/5369799
branch
2 times, most recently
from
c5aba98 to
8570589
Compare
|
/bot run |
|
PR_Github #11947 [ run ] triggered by Bot |
|
PR_Github #11947 [ run ] completed with state |
WalkthroughThe updates modify sequence length calculations in C++ and Python components related to request and token management. Specifically, the C++ function now includes context phase generation tokens in sequence length, while the Python samplers remove special handling for an overlap flag, simplifying new token detection. Minor formatting cleanup is also included. Changes
Suggested reviewers
Poem
📜 Recent review detailsConfiguration used: .coderabbit.yaml 📒 Files selected for processing (3)
💤 Files with no reviewable changes (1)
🚧 Files skipped from review as they are similar to previous changes (2)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
✨ Finishing Touches
🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
Documentation and Community
|
|
/bot run |
|
PR_Github #12196 [ run ] triggered by Bot |
|
PR_Github #12196 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #12211 [ run ] triggered by Bot |
|
/bot run |
|
PR_Github #12217 [ run ] triggered by Bot |
|
PR_Github #12211 [ run ] completed with state |
|
PR_Github #12217 [ run ] completed with state |
- Set `is_disagg` in `TRTLLMSampler` to False by default. - Modify sequence length calculation to consider `is_disagg` when determining overlap behavior in the sampling process. - Ensure `kv_cache_transceiver` presence updates `sampler.is_disagg` in the executor instance creation. Signed-off-by: Stefan Niebler <[email protected]>
Fixed Disaggregated Serving + overlap scheduler dropping tokens in the output. CopySequenceLengths now correctly respects tokens generated in the first generation phase. Removed the line in sampler.py that masked this issue and caused bug 5367999 Signed-off-by: Stefan Niebler <[email protected]>
Signed-off-by: Stefan Niebler <[email protected]>
|
/bot run |
|
PR_Github #12294 [ run ] triggered by Bot |
|
PR_Github #12294 [ run ] completed with state |
…IA#5762) Signed-off-by: Stefan Niebler <[email protected]>
…IA#5762) Signed-off-by: Stefan Niebler <[email protected]>
…IA#5762) Signed-off-by: Stefan Niebler <[email protected]> Signed-off-by: Shreyas Misra <[email protected]>
Description
Full fix for [nvbugs/5369799] :
Using Disaggregated Serving + overlap scheduler now correctly does not drop an output token, which allows us to remove the Workaround
As a result: Using Disaggregated Serving, Overlap scheduler and Stop words, now correctly no longer outputs the last token twice.
Test Coverage
GitHub Bot Help
/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...Provide a user friendly way for developers to interact with a Jenkins server.
Run
/bot [-h|--help]to print this help message.See details below for each supported subcommand.
run [--disable-fail-fast --skip-test --stage-list "A10-1, xxx" --gpu-type "A30, H100_PCIe" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-[Post-Merge]-1, xxx"]Launch build/test pipelines. All previously running jobs will be killed.
--disable-fail-fast(OPTIONAL) : Disable fail fast on build/tests/infra failures.--skip-test(OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.--stage-list "A10-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-1, xxx". Note: Does NOT update GitHub check status.--gpu-type "A30, H100_PCIe"(OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.--add-multi-gpu-test(OPTIONAL) : Force run the multi-GPU tests. Will also run L0 pre-merge pipeline.--post-merge(OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.--extra-stage "H100_PCIe-[Post-Merge]-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-[Post-Merge]-1, xxx".For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.md.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip testing for latest commit on pull request.
--comment "Reason for skipping build/test"is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.reuse-pipeline
reuse-pipelineReuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.
Summary by CodeRabbit
Bug Fixes
Style