[None][feat] Draft: Speculative One Model: FlashInfer sampling #10284
Conversation
IzzyPutterman
changed the title
📝 WalkthroughWalkthroughThe changes add flashinfer-based sampling support to the speculative decoding pipeline. Eagle3OneModelWorker now accepts flashinfer configuration parameters (use_flashinfer, seed, offset) and passes them to sampling_batch_spec_dec_one_model. The sampling function adds a conditional path: when use_flashinfer is enabled, it invokes flashinfer's top-k/top-p sampling; otherwise, it falls back to the existing PyTorch-native implementation. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
tensorrt_llm/_torch/speculative/one_model_sampler.py (1)
4-4: Consider guarding the FlashInfer import.The import is unconditional, which will cause an
ImportErrorif FlashInfer is not installed. Consider using a try-except or lazy import pattern to make the dependency optional.🔎 Proposed fix using try-except guard
-from flashinfer.sampling import top_k_top_p_sampling_from_logits +try: + from flashinfer.sampling import top_k_top_p_sampling_from_logits +except ImportError: + top_k_top_p_sampling_from_logits = NoneThen add a runtime check in
sampling_batch_spec_dec_one_model:if use_flashinfer: if top_k_top_p_sampling_from_logits is None: raise ImportError("FlashInfer is required for use_flashinfer=True") return top_k_top_p_sampling_from_logits(logits, top_k, top_p, seed, offset)
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
tensorrt_llm/_torch/speculative/eagle3.pytensorrt_llm/_torch/speculative/one_model_sampler.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code developed for TensorRT-LLM should conform to Python 3.8+
Indent Python code with 4 spaces. Do not use tabs
Always maintain the namespace when importing in Python, even if only one class or function from a module is used
Python files should use snake_case naming:some_file.py
Python classes should use PascalCase naming:class SomeClass
Python functions and methods should use snake_case naming:def my_awesome_function():
Python local variables should use snake_case naming:my_variable = ...
Python variable names that start with a number should be prefixed with 'k':k_99th_percentile = ...
Python global variables should use upper snake_case with prefix 'G':G_MY_GLOBAL = ...
Python constants should use upper snake_case naming:MY_CONSTANT = ...
Avoid shadowing variables declared in an outer scope in Python
Initialize all externally visible members of a Python class in the constructor
For Python interfaces that may be used outside a file, prefer docstrings over comments
Python comments should be reserved for code within a function, or interfaces that are local to a file
Use Google style docstrings in Python for classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with type and description
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except to the smallest set of errors possible
When using try-except blocks in Python to handle multiple possible variable types (duck-typing), keep the body of the try as small as possible, using the else block for logic
Files:
tensorrt_llm/_torch/speculative/eagle3.pytensorrt_llm/_torch/speculative/one_model_sampler.py
**/*.{cpp,h,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the year of its latest meaningful modification
Files:
tensorrt_llm/_torch/speculative/eagle3.pytensorrt_llm/_torch/speculative/one_model_sampler.py
🧠 Learnings (2)
📚 Learning: 2025-08-27T15:03:57.149Z
Learnt from: ixlmar
Repo: NVIDIA/TensorRT-LLM PR: 7294
File: tensorrt_llm/_torch/pyexecutor/sampler.py:368-392
Timestamp: 2025-08-27T15:03:57.149Z
Learning: In TensorRT-LLM's sampler.py, int32 usage for softmax_indices and related tensor indexing is intentional and should not be changed to int64. The torch.IntTensor type hint is correct for the sample() function's softmax_indices parameter.
Applied to files:
tensorrt_llm/_torch/speculative/one_model_sampler.py
📚 Learning: 2025-12-12T03:27:18.859Z
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 9655
File: tensorrt_llm/_torch/pyexecutor/sampler.py:3031-3031
Timestamp: 2025-12-12T03:27:18.859Z
Learning: In tensorrt_llm/_torch/pyexecutor/sampler.py, when reviewing code that iterates through requests, ensure it does not convert excessive data into Python lists. Instead, the code should use torch.gather or indexing to gather only the data that will be used in the for loop before converting to Python lists. This minimizes data movement and improves performance.
Applied to files:
tensorrt_llm/_torch/speculative/one_model_sampler.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/speculative/eagle3.py (1)
tensorrt_llm/_torch/speculative/one_model_sampler.py (1)
sampling_batch_spec_dec_one_model(77-97)
⏰ 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)
- GitHub Check: Pre-commit Check
🔇 Additional comments (3)
tensorrt_llm/_torch/speculative/one_model_sampler.py (1)
76-97: Verify@torch.compilecompatibility with conditional branching.The function is decorated with
@torch.compile, and the docstring notes it should be "CUDA-graph compatible." The newuse_flashinferconditional branch may cause recompilation or graph-break issues depending on how torch.compile handles the branching. Ensure this has been tested with CUDA graphs enabled.tensorrt_llm/_torch/speculative/eagle3.py (2)
370-372: LGTM! New FlashInfer configuration attributes.The initialization of
use_flashinfer,seed, andoffsetwith sensible defaults is appropriate. These attributes are correctly scoped to the worker instance.
535-545: Clarify the seed/offset update strategy.The seed is incremented by 1 each time sampling occurs when FlashInfer is enabled, but
offsetis never updated. Typically, FlashInfer's sampling usesoffsetto track the number of samples drawn within the same seed context.Please verify this aligns with FlashInfer's expected usage pattern—usually either:
- Increment
offsetper sample call (keeping seed fixed per request), or- Increment
seedand reset offset (if starting a new RNG stream)Also, note that
sampled_tokenswill receive a tuple when FlashInfer is enabled due to the return type issue inone_model_sampler.py(see related comment there).
| if use_flashinfer: | ||
| return top_k_top_p_sampling_from_logits(logits, top_k, top_p, seed, offset) | ||
| random_sampled = forward_native(logits, top_k, top_p) | ||
| return random_sampled |
There was a problem hiding this comment.
Inconsistent return types between branches.
The function signature declares -> tuple[torch.Tensor, torch.Tensor], but:
- FlashInfer path:
return top_k_top_p_sampling_from_logits(logits, top_k, top_p, seed, offset)returns a tuple - Native path:
forward_nativereturns a singletorch.Tensor(line 19 showsreturn random_sample(probs))
Callers like _sample_tokens_for_batch in eagle3.py assign the result to sampled_tokens, expecting a tensor. When FlashInfer is enabled, they will receive a tuple instead, causing runtime errors or incorrect behavior.
🔎 Proposed fix to ensure consistent return types
logits = apply_temperature(logits, temperatures)
if use_flashinfer:
- return top_k_top_p_sampling_from_logits(logits, top_k, top_p, seed, offset)
+ sampled_tokens, _ = top_k_top_p_sampling_from_logits(logits, top_k, top_p, seed, offset)
+ return sampled_tokens
random_sampled = forward_native(logits, top_k, top_p)
return random_sampledAlso update the return type annotation to match the actual return:
-) -> tuple[torch.Tensor, torch.Tensor]:
+) -> torch.Tensor:Committable suggestion skipped: line range outside the PR's diff.
🤖 Prompt for AI Agents
In tensorrt_llm/_torch/speculative/one_model_sampler.py around lines 94 to 97,
the two branches return incompatible types (FlashInfer returns a tuple while
native returns a single torch.Tensor), causing callers that expect a tensor to
break; make the FlashInfer branch return the same single torch.Tensor as the
native branch by extracting the sampled-tokens tensor from the tuple returned by
top_k_top_p_sampling_from_logits (or by unpacking its return and returning only
the sampled tensor), and update the function return annotation to torch.Tensor
to reflect the actual single-tensor return value.
IzzyPutterman
force-pushed
the
iputterman/flashinfer-onemodel
branch
2 times, most recently
from
9913b28 to
ce4584d
Compare
IzzyPutterman
force-pushed
the
iputterman/flashinfer-onemodel
branch
from
ce4584d to
3c26234
Compare
|
/bot run |
|
PR_Github #30406 [ run ] triggered by Bot. Commit: |
|
PR_Github #30406 [ run ] completed with state
|
IzzyPutterman
force-pushed
the
iputterman/flashinfer-onemodel
branch
from
3c26234 to
373277d
Compare
|
/bot run |
|
PR_Github #30412 [ run ] triggered by Bot. Commit: |
|
PR_Github #30412 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30416 [ run ] triggered by Bot. Commit: |
|
PR_Github #30416 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30458 [ run ] triggered by Bot. Commit: |
|
PR_Github #30458 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30460 [ run ] triggered by Bot. Commit: |
|
PR_Github #30460 [ run ] completed with state
|
IzzyPutterman
force-pushed
the
iputterman/flashinfer-onemodel
branch
from
373277d to
305f737
Compare
Signed-off-by: Izzy Putterman <[email protected]>
IzzyPutterman
force-pushed
the
iputterman/flashinfer-onemodel
branch
from
305f737 to
e3d70bd
Compare
|
/bot run |
|
PR_Github #30762 [ run ] triggered by Bot. Commit: |
|
PR_Github #30762 [ run ] completed with state
|
3 participants
Summary by CodeRabbit
✏️ Tip: You can customize this high-level summary in your review settings.
Description
Ready once we have a flash infer build with : flashinfer-ai/flashinfer@18004a8
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
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.
Details
run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]Launch build/test pipelines. All previously running jobs will be killed.
--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.--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-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-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.--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline 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 in addition to running 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-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.--debug(OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.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.