docs/source/torch.md (1)

4-4:Re-phrase for a smoother reading flow

“launches a new backend” sounds like a one-off event. “introduces” (or “adds”) better reflects the documentation’s timeless nature.

-To enhance the usability of the system and improve developer efficiency, TensorRT-LLM launches a new backend based on PyTorch.+To enhance usability and developer efficiency, TensorRT-LLM introduces a new backend based on PyTorch.

docs/source/advanced/speculative-decoding.md (1)

171-171:Minor grammar & spacing tidy-up

Remove the redundant “of”, add the missing space, and swap the en-dash for a hyphen to stay consistent.

-Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft tokens acceptance and draft token generation are performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported).+Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model so that logits prediction, draft-token acceptance, and draft-token generation are performed inside the TensorRT engine (EAGLE-1 and EAGLE-2 are both supported).

examples/models/core/llama/README.md (2)

679-679:Capitalise sentence start & tighten wording

-Note: use FP8 GEMV to optimize performance in FP8 small-batch-size cases.+Note: Use FP8 GEMV to optimise performance in small-batch-size FP8 scenarios.

---

697-697:Polish long explanatory note for readability

A few micro-fixes improve clarity:

-**Note**: FP8 gemv plugin uses CUDA cores to compute, by contrast to Tensor Core gemm kernel within cuBLAS. Over last year, as cuBLAS have improved their performance by a lot under small M case for Hopper(sm90), FP8 gemv kernel may or may not surpass cuBLAS, depending on specific gemm problem shape. Nonetheless, we still strongly recommend FP8 gemv kernel for Ada (sm89) as cuBLAS still falls behind gemv on it.+**Note**: The FP8 GEMV plugin runs on CUDA cores, whereas cuBLAS uses Tensor-Core GEMM kernels. Over the last year cuBLAS performance for small-M cases on Hopper (SM90) has improved substantially, so FP8 GEMV may or may not outperform cuBLAS depending on the exact GEMM shape. We still strongly recommend FP8 GEMV on Ada (SM89), where cuBLAS continues to lag behind.

docs/source/architecture/model-weights-loader.md (1)

252-252:Clarify “by default” clause

The current wording is slightly ambiguous about when the loader is active.

-The weights loader is enabled for LLaMA family models and Qwen models by default with TensorRT flow only.+By default, the weights loader is enabled for LLaMA-family and Qwen models when using the TensorRT flow.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

docs/source/advanced/speculative-decoding.md (1)

171-171:Tighten wording and fix minor grammar issues

The sentence contains a few stylistic hiccups:
• “inside of” → “inside”
• missing space before the parenthesis after “engine”
• plural-singular mismatch in “draft tokens acceptance”
• superfluous comma after “Please”

Proposed tweak:

-Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft tokens acceptance and draft token generation are performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported). Please, visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.+Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft token acceptance, and draft token generation are performed inside the TensorRT engine (EAGLE-1 and EAGLE-2 are both supported). Please visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

docs/source/advanced/speculative-decoding.md (2)

171-171:Tighten wording and fix minor formatting issues
The phrase “inside of the TensorRT engine” is redundant, there’s a missing space before the parenthesis, and “draft tokens acceptance” should be singular.

-Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft tokens acceptance and draft token generation are performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported). Please, visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.+Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft-token acceptance, and draft-token generation are performed inside the TensorRT engine (EAGLE-1 and EAGLE-2 are both supported). For details on building and running the model, see the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md).

---

169-170:Correct typo and clarify sentences
“asssembled” has an extra “s”, and both sentences read more naturally with slight re-phrasing.

-In the EAGLE-1 decoding tree needs to be known during the decoding. In the EAGLE-2 this tree is asssembled during the execution by searching for the most probable hypothesis along the beam.+In EAGLE-1, the decoding tree must be defined before decoding begins. In EAGLE-2, this tree is assembled during execution by searching for the most probable hypothesis along the beam.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

docs/source/advanced/speculative-decoding.md (1)

171-171:Polish grammar & spacing for clarity

Minor wording and punctuation tweaks improve readability:

-Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft tokens acceptance and draft token generation are performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported). Please, visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.+Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logit prediction, draft-token acceptance, and draft-token generation are performed inside the TensorRT engine (EAGLE-1 and EAGLE-2 are both supported). Please visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.

Key fixes:
• “inside the” instead of “inside of the”.
• Add space before the parenthesis.
• Oxford comma after “acceptance”.
• Singular “logit” and hyphenate compound nouns.
• Remove superfluous comma after “Please”.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

docs/source/advanced/speculative-decoding.md (1)

171-171:Drop “of” after “inside” and fix missing space before parenthesis

Small wording/formatting tweaks improve readability.

-... performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported).+... performed inside the TensorRT engine (both EAGLE-1 and EAGLE-2 are supported).

examples/disaggregated/README.md (1)

112-116:Fix typo in YAML keyrefresh_interval
refersh_interval is miss-spelled. Anyone copying this sample will hit a configuration error.

-  refersh_interval: 10.0+  refresh_interval: 10.0

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

Learnt from: yibinl-nvidiaPR: NVIDIA/TensorRT-LLM#6506File: examples/models/core/mixtral/requirements.txt:3-3Timestamp: 2025-08-01T15:14:45.673ZLearning: 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.

Learnt from: yechank-nvidiaPR: NVIDIA/TensorRT-LLM#6254File: tensorrt_llm/_torch/pyexecutor/model_engine.py:1201-1204Timestamp: 2025-07-22T09:22:14.726ZLearning: In TensorRT-LLM's multimodal processing pipeline, shared tensor recovery using `from_shared_tensor()` is only needed during the context phase. Generation requests reuse the already-recovered tensor data and only need to call `strip_for_generation()` to remove unnecessary multimodal data while preserving the recovered tensors. This avoids redundant tensor recovery operations during generation.

Learnt from: moraxuPR: NVIDIA/TensorRT-LLM#6303File: tests/integration/test_lists/qa/examples_test_list.txt:494-494Timestamp: 2025-07-28T17:06:08.621ZLearning: 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.

docs/source/advanced/speculative-decoding.md (1)

171-171:Tighten grammar & spacing for clarity

Minor wording polish:
• “inside of” → “inside” (redundant “of”).
• Insert Oxford comma after “acceptance”.
• Add space before the opening parenthesis.
• Drop comma after “Please”.

-Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft tokens acceptance and draft token generation are performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported). Please, visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.+Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft token acceptance, and draft token generation are performed inside the TensorRT engine (EAGLE-1 and EAGLE-2 are both supported). Please visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.

examples/disaggregated/README.md (2)

110-116:Fix typo in key name –refresh_interval

refersh_interval will confuse users who copy-paste the YAML and may break config loaders that validate keys.

-  refersh_interval: 10.0+  refresh_interval: 10.0

---

181-183:Correct section title – “Known Issues”

Minor wording nit:

-## Know Issues+## Known Issues

This keeps terminology consistent across the docs.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

Learnt from: yibinl-nvidiaPR: NVIDIA/TensorRT-LLM#6506File: examples/models/core/mixtral/requirements.txt:3-3Timestamp: 2025-08-01T15:14:45.673ZLearning: 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.

Learnt from: yechank-nvidiaPR: NVIDIA/TensorRT-LLM#6254File: tensorrt_llm/_torch/pyexecutor/model_engine.py:1201-1204Timestamp: 2025-07-22T09:22:14.726ZLearning: In TensorRT-LLM's multimodal processing pipeline, shared tensor recovery using `from_shared_tensor()` is only needed during the context phase. Generation requests reuse the already-recovered tensor data and only need to call `strip_for_generation()` to remove unnecessary multimodal data while preserving the recovered tensors. This avoids redundant tensor recovery operations during generation.

Learnt from: moraxuPR: NVIDIA/TensorRT-LLM#6303File: tests/integration/test_lists/qa/examples_test_list.txt:494-494Timestamp: 2025-07-28T17:06:08.621ZLearning: 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.

docs/source/advanced/speculative-decoding.md (1)

171-171:Remove redundant “of” and clean up punctuation

Minor wording and punctuation polish for clarity and consistency.

-Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft tokens acceptance and draft token generation are performed inside of the TensorRT engine(EAGLE-1 and EAGLE-2 are both supported). Please, visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.+Similarly to ReDrafter, TensorRT-LLM implements the EAGLE model such that logits prediction, draft token acceptance, and draft token generation are performed inside the TensorRT engine (both EAGLE-1 and EAGLE-2 are supported). Please visit the [EAGLE README](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/eagle/README.md) for information about building and running the model.

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

Learnt from: yibinl-nvidiaPR: NVIDIA/TensorRT-LLM#6506File: examples/models/core/mixtral/requirements.txt:3-3Timestamp: 2025-08-01T15:14:45.673ZLearning: 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.

Learnt from: yechank-nvidiaPR: NVIDIA/TensorRT-LLM#6254File: tensorrt_llm/_torch/pyexecutor/model_engine.py:1201-1204Timestamp: 2025-07-22T09:22:14.726ZLearning: In TensorRT-LLM's multimodal processing pipeline, shared tensor recovery using `from_shared_tensor()` is only needed during the context phase. Generation requests reuse the already-recovered tensor data and only need to call `strip_for_generation()` to remove unnecessary multimodal data while preserving the recovered tensors. This avoids redundant tensor recovery operations during generation.

Learnt from: moraxuPR: NVIDIA/TensorRT-LLM#6303File: tests/integration/test_lists/qa/examples_test_list.txt:494-494Timestamp: 2025-07-28T17:06:08.621ZLearning: 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.