[TRTC-122][feat] Eagle3 Specdec UX improvements #10124
Conversation
|
/bot run --disable-fail-fast |
|
PR_Github #28985 [ run ] triggered by Bot. Commit: |
There was a problem hiding this comment.
calling this out: previously, removing this would fail.
but since the default is vocab_size anyways, the behavior after current changes should be identical (+ a warning noting user that vocab_size was chosen as default)
venkywonka
changed the title
|
PR_Github #28985 [ run ] completed with state |
venkywonka
changed the title
venkywonka
requested review from
Superjomn,
laikhtewari,
nv-guomingz and
symphonylyh
📝 WalkthroughWalkthroughThis pull request introduces Eagle3 as a new speculative decoding type option. Documentation updates reflect the decoding_type preference from Eagle to Eagle3. Core implementation adds Eagle3 suffix detection, default draft_vocab_size handling with warnings, and a new Eagle3DecodingConfig class with TensorRT validation that blocks Eagle3 usage on TRT backend while enabling PyTorch support. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
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: 0
🧹 Nitpick comments (3)
docs/source/features/speculative-decoding.md (1)
136-142: Optional: Add language identifier to YAML code block.The YAML code block at line 136 is missing a language identifier. While this doesn't affect functionality, adding
yamlwould improve syntax highlighting.🔎 Suggested improvement
-``` +```yaml disable_overlap_scheduler: true speculative_config: decoding_type: Eagle3tensorrt_llm/_torch/models/modeling_speculative.py (2)
180-186: Addstacklevel=2towarnings.warnfor correct caller attribution.Without
stacklevel, the warning will point to this line rather than the caller's location, making debugging harder.🔎 Proposed fix
if not hasattr(config, "draft_vocab_size"): warnings.warn( "Pretrained config does not define 'draft_vocab_size'; assuming it matches 'vocab_size'. " "If the draft head uses a different vocabulary, set 'draft_vocab_size' explicitly " - "before exporting to TensorRT-LLM.") + "before exporting to TensorRT-LLM.", + stacklevel=2) config.draft_vocab_size = config.vocab_size
312-326: Addstacklevel=2towarnings.warnfor correct caller attribution.Same issue as in
Eagle3DraftModel.__init__.🔎 Proposed fix
config = model_config.pretrained_config if not hasattr(config, "draft_vocab_size"): warnings.warn( "Pretrained config does not define 'draft_vocab_size'; assuming it matches 'vocab_size'. " "If the draft head uses a different vocabulary, set 'draft_vocab_size' explicitly " - "before exporting to TensorRT-LLM.") + "before exporting to TensorRT-LLM.", + stacklevel=2) config.draft_vocab_size = config.vocab_size
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (11)
docs/source/blogs/tech_blog/blog11_GPT_OSS_Eagle3.mddocs/source/blogs/tech_blog/blog6_Llama4_maverick_eagle_guide.mddocs/source/features/speculative-decoding.mddocs/source/features/torch_compile_and_piecewise_cuda_graph.mddocs/source/legacy/advanced/speculative-decoding.mdexamples/models/core/qwen/README.mdtensorrt_llm/_torch/models/modeling_auto.pytensorrt_llm/_torch/models/modeling_speculative.pytensorrt_llm/llmapi/llm_args.pytests/unittest/_torch/speculative/test_eagle3.pytests/unittest/llmapi/test_llm_args.py
💤 Files with no reviewable changes (1)
- tests/unittest/_torch/speculative/test_eagle3.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:
tests/unittest/llmapi/test_llm_args.pytensorrt_llm/_torch/models/modeling_speculative.pytensorrt_llm/_torch/models/modeling_auto.pytensorrt_llm/llmapi/llm_args.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:
tests/unittest/llmapi/test_llm_args.pytensorrt_llm/_torch/models/modeling_speculative.pytensorrt_llm/_torch/models/modeling_auto.pytensorrt_llm/llmapi/llm_args.py
🧠 Learnings (25)
📚 Learning: 2025-08-18T08:42:02.640Z
Learnt from: samuellees
Repo: NVIDIA/TensorRT-LLM PR: 6974
File: tensorrt_llm/serve/scripts/benchmark_dataset.py:558-566
Timestamp: 2025-08-18T08:42:02.640Z
Learning: In TensorRT-LLM's RandomDataset (tensorrt_llm/serve/scripts/benchmark_dataset.py), when using --random-token-ids option, sequence length accuracy is prioritized over semantic correctness for benchmarking purposes. The encode/decode operations should use skip_special_tokens=True and add_special_tokens=False to ensure exact target token lengths.
Applied to files:
docs/source/features/speculative-decoding.mddocs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
Repo: NVIDIA/TensorRT-LLM PR: 7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM's bench configuration, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which is a Dict[str, Any] that can contain default values including `cuda_graph_config`, making the fallback `llm_args["cuda_graph_config"]` safe to use.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
Repo: NVIDIA/TensorRT-LLM PR: 7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which can contain default `cuda_graph_config` values, so `llm_args` may already have this config before the extra options processing.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-09-04T07:33:10.618Z
Learnt from: MrGeva
Repo: NVIDIA/TensorRT-LLM PR: 7219
File: tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py:162-168
Timestamp: 2025-09-04T07:33:10.618Z
Learning: When users explicitly provide cuda_graph_batch_sizes in TorchCudagraphCompiler, respect their choices and only sanitize the values (clamp, dedupe, sort) without forcing additional batch sizes like 1 or max_batch_size. Only add commonly-used batch sizes when falling back to the heuristic.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-08T04:10:19.038Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6728
File: cpp/tensorrt_llm/plugins/mixtureOfExperts/mixtureOfExpertsPlugin.cpp:966-966
Timestamp: 2025-08-08T04:10:19.038Z
Learning: TensorRT plugins currently don't support padding functionality, and TensorRT is not getting new features (in maintenance mode). This means that duplicating parameters like mExpertHiddenSize in function calls, even with TODO comments, can be acceptable as pragmatic solutions within these constraints.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-26T09:49:04.956Z
Learnt from: pengbowang-nv
Repo: NVIDIA/TensorRT-LLM PR: 7192
File: tests/integration/test_lists/test-db/l0_dgx_b200.yml:56-72
Timestamp: 2025-08-26T09:49:04.956Z
Learning: In TensorRT-LLM test configuration files, the test scheduling system handles wildcard matching with special rules that prevent duplicate test execution even when the same tests appear in multiple yaml files with overlapping GPU wildcards (e.g., "*b200*" and "*gb200*").
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-14T15:38:01.771Z
Learnt from: MatthiasKohl
Repo: NVIDIA/TensorRT-LLM PR: 6904
File: cpp/tensorrt_llm/pybind/thop/bindings.cpp:55-57
Timestamp: 2025-08-14T15:38:01.771Z
Learning: In TensorRT-LLM Python bindings, tensor parameter collections like mla_tensor_params and spec_decoding_tensor_params are kept as required parameters without defaults to maintain API consistency, even when it might affect backward compatibility.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-15T06:46:54.897Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6767
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-15T06:46:54.897Z
Learning: In cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp addToken function, newly allocated blocks are unshared by design. The beam search path in addToken (when sequence.getNumTokens() > windowSize) is currently broken/non-functional with SWA, so the block allocation doesn't follow a shared-then-unshared pattern.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-14T21:04:50.248Z
Learnt from: thorjohnsen
Repo: NVIDIA/TensorRT-LLM PR: 6910
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-14T21:04:50.248Z
Learning: In KV cache onboarding logic during prefill in cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp, when calculating which blocks fall within the attention window, use getTokensPerBlock() to advance token indices rather than block->getUniqueTokens().size(), because the calculation needs to consider the post-prefill state where blocks will be filled to capacity, not their current token count.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-09-23T14:58:05.372Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/kernels/nccl_device/config.cu:42-49
Timestamp: 2025-09-23T14:58:05.372Z
Learning: In TensorRT-LLM NCCL device kernels (cpp/tensorrt_llm/kernels/nccl_device/), the token partitioning intentionally uses ceil-like distribution (same token_per_rank for all ranks) to ensure all ranks launch the same number of blocks. This is required for optimal NCCL device API barrier performance, even though it may launch extra blocks for non-existent tokens on later ranks. Runtime bounds checking in the kernel (blockID validation) handles the overshoot cases.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-12-12T10:07:36.866Z
Learnt from: lirundong
Repo: NVIDIA/TensorRT-LLM PR: 9725
File: tensorrt_llm/_torch/custom_ops/cuda_tile_custom_ops.py:110-178
Timestamp: 2025-12-12T10:07:36.866Z
Learning: In PyTorch custom operators registered with torch.library.custom_op, mutable operators that return None and specify mutates_args do NOT require a register_fake decorator. The mutation tracking is handled automatically without needing a FakeTensor kernel, as documented in the PyTorch tutorial on mutable Python custom operators.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-15T06:46:53.813Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6767
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-15T06:46:53.813Z
Learning: In the TensorRT-LLM KV cache manager, SWA (Sliding Window Attention) combined with beam search is currently in a broken/non-functional state and is planned for future rework. During preparatory refactoring phases, code related to SWA+beam search may intentionally remain in a non-working state until the broader rework is completed.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-14T15:43:23.107Z
Learnt from: MatthiasKohl
Repo: NVIDIA/TensorRT-LLM PR: 6904
File: tensorrt_llm/_torch/attention_backend/trtllm.py:259-262
Timestamp: 2025-08-14T15:43:23.107Z
Learning: In TensorRT-LLM's attention backend, tensor parameters in the plan() method are assigned directly without validation (dtype, device, contiguity checks). This maintains consistency across all tensor inputs and follows the pattern of trusting callers to provide correctly formatted tensors.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-07-28T17:06:08.621Z
Learnt from: moraxu
Repo: NVIDIA/TensorRT-LLM PR: 6303
File: tests/integration/test_lists/qa/examples_test_list.txt:494-494
Timestamp: 2025-07-28T17:06:08.621Z
Learning: In TensorRT-LLM testing, it's common to have both CLI flow tests (test_cli_flow.py) and PyTorch API tests (test_llm_api_pytorch.py) for the same model. These serve different purposes: CLI flow tests validate the traditional command-line workflow, while PyTorch API tests validate the newer LLM API backend. Both are legitimate and should coexist.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-11T20:09:24.389Z
Learnt from: achartier
Repo: NVIDIA/TensorRT-LLM PR: 6763
File: tests/integration/defs/triton_server/conftest.py:16-22
Timestamp: 2025-08-11T20:09:24.389Z
Learning: In the TensorRT-LLM test infrastructure, the team prefers simple, direct solutions (like hard-coding directory traversal counts) over more complex but robust approaches when dealing with stable directory structures. They accept the maintenance cost of updating tests if the layout changes.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-21T21:48:35.135Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 7104
File: cpp/tensorrt_llm/cutlass_extensions/include/cutlass_extensions/epilogue/fusion/sm90_visitor_scatter.hpp:399-417
Timestamp: 2025-08-21T21:48:35.135Z
Learning: CUTLASS extensions in TensorRT-LLM (located under cpp/tensorrt_llm/cutlass_extensions/) are designed to integrate with and extend functionality in the external CUTLASS repository. When analyzing these extensions, their consumers and functionality wiring may exist in the CUTLASS codebase rather than within TensorRT-LLM itself.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-09-23T15:12:38.312Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/thop/allreduceOp.cpp:352-446
Timestamp: 2025-09-23T15:12:38.312Z
Learning: In TensorRT-LLM NCCL device implementation, NCCL version 2.28+ requirements are handled at runtime in the nccl_device/config layer rather than with compile-time guards. This allows the allreduceOp to remain version-agnostic and delegates version compatibility validation to the appropriate lower-level components that can gracefully handle unsupported configurations.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-09-09T09:40:45.658Z
Learnt from: fredricz-20070104
Repo: NVIDIA/TensorRT-LLM PR: 7645
File: tests/integration/test_lists/qa/llm_function_core.txt:648-648
Timestamp: 2025-09-09T09:40:45.658Z
Learning: In TensorRT-LLM test lists, it's common and intentional for the same test to appear in multiple test list files when they serve different purposes (e.g., llm_function_core.txt for comprehensive core functionality testing and llm_function_core_sanity.txt for quick sanity checks). This duplication allows tests to be run in different testing contexts.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-09-24T03:31:28.908Z
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 7520
File: tensorrt_llm/_torch/pyexecutor/resource_manager.py:605-613
Timestamp: 2025-09-24T03:31:28.908Z
Learning: In TensorRT-LLM Ray orchestrator mode, ProcessGroups are initialized with both Gloo and NCCL backends (e.g., "cuda:nccl,cpu:gloo"), allowing PyTorch distributed to automatically route CPU tensors through Gloo and GPU tensors through NCCL. This eliminates the need for manual device placement when performing allreduce operations on base types.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-01T15:14:45.673Z
Learnt from: yibinl-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 6506
File: examples/models/core/mixtral/requirements.txt:3-3
Timestamp: 2025-08-01T15:14:45.673Z
Learning: In TensorRT-LLM, examples directory can have different dependency versions than the root requirements.txt file. Version conflicts between root and examples dependencies are acceptable because examples are designed to be standalone and self-contained.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.mddocs/source/legacy/advanced/speculative-decoding.md
📚 Learning: 2025-08-06T13:58:07.506Z
Learnt from: galagam
Repo: NVIDIA/TensorRT-LLM PR: 6487
File: tests/unittest/_torch/auto_deploy/unit/singlegpu/test_ad_trtllm_bench.py:1-12
Timestamp: 2025-08-06T13:58:07.506Z
Learning: In TensorRT-LLM, test files (files under tests/ directories) do not require NVIDIA copyright headers, unlike production source code files. Test files typically start directly with imports, docstrings, or code.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-09-23T15:12:38.312Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/thop/allreduceOp.cpp:352-446
Timestamp: 2025-09-23T15:12:38.312Z
Learning: In TensorRT-LLM NCCL device allreduce implementation (cpp/tensorrt_llm/thop/allreduceOp.cpp), the goto pattern in runNCCLAllReduceDeviceFusion is intentionally used for future extensibility, allowing multiple switch cases to fallback to the default handler. While not aesthetically ideal, this pattern supports adding more fusion cases later that can reuse the same fallback logic.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-17T15:07:01.420Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 6968
File: cpp/tensorrt_llm/thop/loraOp.cpp:133-141
Timestamp: 2025-08-17T15:07:01.420Z
Learning: In TensorRT-LLM's LoRA implementation, the LoraImpl::run() method handles setStream() internally in _runGemm() (line 51 in lora.cpp), along with setWorkspace(). The stream parameter flows from loraOp.cpp through LoraImpl::run() to _runGemm() where setStream() is called appropriately. Adding setStream() in loraOp.cpp would be redundant and goes against the intended architectural design.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-08-17T15:07:01.420Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 6968
File: cpp/tensorrt_llm/thop/loraOp.cpp:133-141
Timestamp: 2025-08-17T15:07:01.420Z
Learning: In TensorRT-LLM's LoRA implementation, the LoraImpl::run() method handles setStream() internally in _runGemm(), along with setWorkspace(). Both stream and workspace are passed as arguments to run(), so there's no need to call setStream() explicitly in loraOp.cpp - this avoids redundancy and follows the intended architectural separation.
Applied to files:
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
📚 Learning: 2025-11-27T09:23:18.742Z
Learnt from: fredricz-20070104
Repo: NVIDIA/TensorRT-LLM PR: 9511
File: tests/integration/defs/examples/serve/test_serve.py:136-186
Timestamp: 2025-11-27T09:23:18.742Z
Learning: In TensorRT-LLM testing, when adding test cases based on RCCA commands, the command format should be copied exactly as it appears in the RCCA case, even if it differs from existing tests. For example, some RCCA commands for trtllm-serve may omit the "serve" subcommand while others include it.
Applied to files:
docs/source/legacy/advanced/speculative-decoding.mddocs/source/blogs/tech_blog/blog6_Llama4_maverick_eagle_guide.md
🧬 Code graph analysis (2)
tests/unittest/llmapi/test_llm_args.py (1)
tensorrt_llm/llmapi/llm_args.py (17)
from_dict(198-216)from_dict(262-263)from_dict(292-293)from_dict(443-444)from_dict(471-472)from_dict(488-489)from_dict(586-595)from_dict(725-746)from_dict(802-803)from_dict(877-878)from_dict(953-954)from_dict(991-992)from_dict(1027-1028)from_dict(1043-1044)from_dict(1078-1082)from_dict(1121-1122)Eagle3DecodingConfig(931-932)
tensorrt_llm/_torch/models/modeling_speculative.py (4)
tensorrt_llm/_torch/models/modeling_llama.py (1)
config(1092-1093)tests/unittest/_torch/executor/test_pytorch_model_engine.py (1)
config(61-62)tensorrt_llm/_torch/models/modeling_utils.py (1)
config(525-526)tensorrt_llm/_torch/models/checkpoints/base_weight_mapper.py (1)
config(166-169)
🪛 LanguageTool
docs/source/features/torch_compile_and_piecewise_cuda_graph.md
[grammar] ~130-~130: Ensure spelling is correct
Context: ...atus For hot models like deepseek/qwen/lllama, we’ve already wrapped some large modul...
(QB_NEW_EN_ORTHOGRAPHY_ERROR_IDS_1)
[grammar] ~191-~191: Use a hyphen to join words.
Context: ...n for AllReduce & RMSNorm. 1. AllReduce related fusion: Fuse the following opera...
(QB_NEW_EN_HYPHEN)
[grammar] ~217-~217: Use a hyphen to join words.
Context: ...d by user config 4. Insert multi-stream related custom op: since the Fx graph ex...
(QB_NEW_EN_HYPHEN)
🪛 markdownlint-cli2 (0.18.1)
docs/source/features/speculative-decoding.md
136-136: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🪛 Ruff (0.14.10)
tests/unittest/llmapi/test_llm_args.py
143-143: DecodingBaseConfig may be undefined, or defined from star imports
(F405)
151-151: Eagle3DecodingConfig may be undefined, or defined from star imports
(F405)
159-159: Unused function argument: args
(ARG001)
159-159: Unused function argument: kwargs
(ARG001)
164-164: DecodingBaseConfig may be undefined, or defined from star imports
(F405)
173-173: TorchLlmArgs may be undefined, or defined from star imports
(F405)
180-180: DecodingBaseConfig may be undefined, or defined from star imports
(F405)
190-190: TrtLlmArgs may be undefined, or defined from star imports
(F405)
tensorrt_llm/_torch/models/modeling_speculative.py
1-1: The file is executable but no shebang is present
(EXE002)
181-181: No explicit stacklevel keyword argument found
Set stacklevel=2
(B028)
314-314: No explicit stacklevel keyword argument found
Set stacklevel=2
(B028)
tensorrt_llm/llmapi/llm_args.py
2431-2434: Avoid specifying long messages outside the exception class
(TRY003)
🔇 Additional comments (19)
docs/source/blogs/tech_blog/blog11_GPT_OSS_Eagle3.md (1)
87-87: LGTM! Documentation correctly reflects Eagle3 decoding type.The update to use
Eagle3as thedecoding_typealigns with the PR's objective to introduce Eagle3 as a distinct configuration option and improve clarity between Eagle (v1/v2) and Eagle3.docs/source/features/torch_compile_and_piecewise_cuda_graph.md (2)
53-53: LGTM! Clarification improves documentation accuracy.The addition of "Specify max capture batch size" makes it clearer what this parameter controls for generation-only CUDA graphs.
93-94: LGTM! Documentation correctly updated to Eagle3.Consistent with the PR's goal to distinguish Eagle3 from legacy Eagle configurations.
docs/source/legacy/advanced/speculative-decoding.md (1)
174-175: LGTM! Helpful clarification for users.This note provides clear guidance on the draft_vocab_size defaulting behavior introduced in this PR, helping users understand when they need to explicitly set this parameter versus when they can rely on the default.
docs/source/features/speculative-decoding.md (1)
128-133: LGTM! Documentation clearly explains Eagle3 usage and backward compatibility.The updates appropriately:
- List Eagle3 as the preferred decoding type
- Document that Eagle is accepted as a PyTorch-backend alias
- Recommend using Eagle3 for clarity
This aligns well with the PR's goal to reduce confusion between Eagle (v1/v2) and Eagle3.
tests/unittest/llmapi/test_llm_args.py (3)
142-152: LGTM! Test appropriately verifies Eagle3 config parsing.This test ensures that the new
Eagle3decoding type correctly parses to anEagle3DecodingConfiginstance, validating the from_dict dispatch mechanism introduced in the PR.
154-177: LGTM! Test correctly verifies backward compatibility warning.The test appropriately verifies that:
- Using
Eagledecoding type on PyTorch backend still works (backward compatibility)- A warning is emitted to guide users toward using
Eagle3The monkeypatch approach to capture warnings is clean and appropriate.
179-191: LGTM! Test correctly verifies TensorRT backend restriction.This test appropriately ensures that the TensorRT backend rejects
Eagle3decoding type, as the legacy TensorRT backend only supports the original Eagle implementation. The error message provides clear guidance to users.tensorrt_llm/_torch/models/modeling_auto.py (2)
17-17: LGTM! Early detection of Eagle3 suffix is correctly placed.Detecting the Eagle3 suffix before any string manipulations ensures the flag is set accurately, which is then used in the conditional logic below.
33-35: LGTM! Extended condition correctly handles Eagle3 checkpoints.The updated condition now treats models as Eagle3 when either:
- They have a
draft_vocab_sizeattribute (original check), OR- They had an "Eagle3" suffix in the architecture name (new check)
This properly handles Eagle3 checkpoints that identify themselves via naming convention, aligning with the PR's objective to improve Eagle3 checkpoint detection.
docs/source/blogs/tech_blog/blog6_Llama4_maverick_eagle_guide.md (1)
71-71: LGTM! Docker run example correctly uses Eagle3.The embedded configuration in the docker run command now correctly specifies
Eagle3as the decoding type, ensuring the guide provides accurate instructions for users.examples/models/core/qwen/README.md (2)
840-841: LGTM! Documentation correctly references Eagle3.The example configuration now uses
Eagle3as the decoding type, providing accurate guidance for users configuring Eagle3 speculative decoding with Qwen models.
858-858: LGTM! YAML example correctly uses Eagle3.Consistent with other documentation updates in this PR, the configuration example now properly specifies
Eagle3.tensorrt_llm/_torch/models/modeling_speculative.py (1)
1-1: LGTM!The
warningsimport is correctly added to support the newdraft_vocab_sizefallback warnings.tensorrt_llm/llmapi/llm_args.py (5)
728-739: LGTM!The
Eagle3dispatch entry is correctly added to thefrom_dictmapping, enabling proper deserialization of Eagle3 configs.
931-934: LGTM!Clean subclass design that reuses
EagleDecodingConfigbehavior while explicitly identifying asEagle3decoding type.
2430-2434: LGTM!The validation correctly rejects
Eagle3on the TensorRT backend with a clear, actionable error message. The check is properly ordered before theEagleDecodingConfigcheck sinceEagle3DecodingConfigis a subclass.
2436-2449: LGTM!The
EagleDecodingConfigvalidation for TensorRT backend correctly handles legacy Eagle with appropriate assertion messaging.
2953-2959: LGTM!Good use of
type(x) is EagleDecodingConfigto distinguish the base class fromEagle3DecodingConfigsubclass. The warning appropriately guides users to use the explicitEagle3type, and the assertion message correctly references "EAGLE3 weights" for the PyTorch context.
venkywonka
force-pushed
the
fix-eagle3-draft-vocab-fallback
branch
from
3888cd8 to
2218e4d
Compare
venkywonka
changed the title
venkywonka
changed the title
|
/bot run |
|
PR_Github #29603 [ run ] triggered by Bot. Commit: |
|
PR_Github #29603 [ run ] completed with state
|
venkywonka
force-pushed
the
fix-eagle3-draft-vocab-fallback
branch
from
2218e4d to
21b499a
Compare
|
/bot run |
|
PR_Github #30543 [ run ] triggered by Bot. Commit: |
Introduce speculative_config.decoding_type: Eagle3 for the PyTorch backend, warn when using Eagle as an alias, and reject Eagle3 on the TensorRT backend. Update docs/examples and add unit tests. Signed-off-by: Venky Ganesh <[email protected]>
De-duplicate draft_vocab_size fallback warning text and clarify that decoding_type: Eagle is a PyTorch-backend alias for Eagle3 (EAGLE v1/v2 draft checkpoints are incompatible). Signed-off-by: Venky Ganesh <[email protected]>
Add test_eagle3_defaults_draft_vocab_size_when_missing to explicitly test the fallback behavior when draft_vocab_size is missing from pretrained config. Restore draft_vocab_size in existing test configs (test_deepseek_eagle3, test_multi_eagle3) rather than relying on the fallback path. Update test imports to use Eagle3DecodingConfig. Signed-off-by: Venky Ganesh <[email protected]>
- Export Eagle3DecodingConfig from llmapi - Add _decoding_type_alias tracking for Eagle→Eagle3 mapping on PyTorch - Update from_dict to map 'Eagle' to Eagle3DecodingConfig on PyTorch backend - Show deprecation warning when 'Eagle' is used on PyTorch backend - Reject 'Eagle3' on TensorRT backend with clear error message - Update docs and examples to use Eagle3DecodingConfig - Update test imports to Eagle3DecodingConfig Signed-off-by: Venky Ganesh <[email protected]>
Signed-off-by: Venky Ganesh <[email protected]>
Signed-off-by: Venky Ganesh <[email protected]>
Signed-off-by: Venky Ganesh <[email protected]>
Signed-off-by: Venky Ganesh <[email protected]>
venkywonka
force-pushed
the
fix-eagle3-draft-vocab-fallback
branch
from
6fcc572 to
429b650
Compare
|
/bot run |
|
PR_Github #30817 [ run ] triggered by Bot. Commit: |
Signed-off-by: Venky Ganesh <[email protected]>
|
/bot run |
|
PR_Github #30824 [ run ] triggered by Bot. Commit: |
|
PR_Github #30824 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30876 [ run ] triggered by Bot. Commit: |
|
PR_Github #30876 [ run ] completed with state
|
Signed-off-by: Venky Ganesh <[email protected]>
|
/bot run |
|
PR_Github #30917 [ run ] triggered by Bot. Commit: |
|
PR_Github #30917 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30966 [ run ] triggered by Bot. Commit: |
|
PR_Github #30966 [ run ] completed with state
|
|
/bot run |
|
PR_Github #31032 [ run ] triggered by Bot. Commit: |
|
PR_Github #31032 [ run ] completed with state
|
8 participants
Description
1. Make
draft_vocab_sizeoptional for Eagle3, and default totarget_vocab_sizedraft_vocab_sizeas part of the nestedspeculative_decoding_configin the config.yaml.draft_vocab_size == target_vocab_sizein the default, unspecified situation without throwing an error. Following the same behavior would improve usability and keeping our API tight with vLLM for interop.2. Deprecate
decoding_type: Eaglefor pytorch backend, introducedecoding_type: Eagle3.Eagle(v1, v2) are fundamentally different fromEagle3- and should not be conflated.decoding_type: Eagleto switch over - so its non-breaking - and explicitly inform them that TRTLLM pytorch backend treatsdecoding_type: EagleasEagle3and therefore expects Eagle 3 draft checkpoint. The user must be explicitly noted that providing an Eagle draft checkpoint will result in errors - currently this is rather implicit and not documented clearly.decoding_type: Eagle3will raise an error.decoding_type: Eaglewill continue to work as usual.Summary by CodeRabbit
New Features
Documentation
Bug Fixes
Tests
✏️ Tip: You can customize this high-level summary in your review settings.
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.