[None][doc] Create deployment guide for Llama 3.3 70B FP8 and NVFP4 #6543
Conversation
📝 WalkthroughWalkthroughA new documentation file has been added, providing a detailed quick start guide for deploying and benchmarking the Llama 3.3-70B Instruct model with TensorRT-LLM using FP8 and NVFP4 quantization on NVIDIA GPUs. The guide covers prerequisites, setup, configuration, inference, evaluation, benchmarking, and troubleshooting. Changes
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~3 minutes Possibly related PRs
Suggested labels
Suggested reviewers
Note ⚡️ Unit Test Generation is now available in beta!Learn more here, or try it out under "Finishing Touches" below. 📜 Recent review detailsConfiguration used: .coderabbit.yaml 📒 Files selected for processing (1)
🚧 Files skipped from review as they are similar to previous changes (1)
✨ Finishing Touches🧪 Generate unit tests
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. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
Documentation and Community
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (5)
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md (5)
1-1: Tighten and standardise the document titleMinor formatting tweaks improve readability and match the naming that appears later in the doc.
-# Deployment Guide for TensorRT-LLM Llama3.3 70B FP8 and NVFP4 +# Deployment Guide for TensorRT-LLM Llama 3.3-70B (FP8 / NVFP4)
3-4: Remove the empty second-level headingThe lone
##renders as an empty heading and breaks the outline.-## -
183-184: Remove stray back-tick after the code fontThe dangling
`renders literally at the end of the sentence.-...extra_llm_api_options`.` +...`extra_llm_api_options`.
238-266: Add language identifiers to fenced code blocksSeveral blocks violate MD040. Prefix them with
shellortextto enable syntax-highlighting:-``` +```shell MODEL_PATH=nvidia/Llama-3.3-70B-Instruct-FP8 ...Do the same for:
- FP8 GSM8K command
- FP4 GSM8K command
- Sample accuracy results (
text)- Sample benchmark output (
text)Also applies to: 326-334
275-278: Rephrase and fix broken link referenceThe sentence is awkward and the Markdown link resolves to http://bench.sh instead of a local file.
-To do this first creating a wrapper [bench.sh](http://bench.sh) script. +First, create a small helper script (e.g. `bench.sh`) to wrap the benchmark command:
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md(1 hunks)
🧰 Additional context used
🧠 Learnings (2)
📓 Common learnings
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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.
📚 Learning: in tensorrt-llm testing, it's common to have both cli flow tests (test_cli_flow.py) and pytorch api ...
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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:
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
🪛 LanguageTool
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
[style] ~117-~117: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...ch** backend. #### --max_batch_size Description: The maximum number of ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~121-~121: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...rocessing. #### --max_num_of_tokens Description: The maximum total numb...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~125-~125: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...scheduled batch. #### --max_seq_len Description: The maximum possible s...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~129-~129: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...ut tokens. #### --trust_remote_code Description: Allows TensorRT-LLM to...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~144-~144: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...: Sets the data type for the KV cache. Default: auto (uses the data ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~148-~148: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...checkpoint). #### cuda_graph_config Description: A section for configur...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~158-~158: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...or which a CUDA graph will be created. Default: 0 **Rec...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~162-~162: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...-max_batch_size command-line option. batch_sizes: A specific list of ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~164-~164: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...batch sizes to create CUDA graphs for. Default: None #### `moe_conf...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~174-~174: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...The backend to use for MoE operations. Default: CUTLASS #### `atten...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~180-~180: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...end to use for attention calculations. Default: TRTLLM See the [https://g...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~305-~305: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...l,e2el" done EOF chmod +x bench.sh ``` To benchmark the FP4 model, replace --mod...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
🪛 markdownlint-cli2 (0.17.2)
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
98-98: Heading levels should only increment by one level at a time
Expected: h3; Actual: h4
(MD001, heading-increment)
240-240: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
248-248: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
326-326: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🔇 Additional comments (1)
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md (1)
25-27: Verify naming: “NVFP4” vs model tag “FP4”The bullet is labelled “NVFP4” but links to
…-FP4on Hugging Face.
Confirm whether the public model tag should actually be…-NVFP4or keep the heading consistent with the link.
| ## Configs and Parameters | ||
|
|
||
| These options are used directly on the command line when you start the `trtllm-serve` process. | ||
| #### `--tp_size` | ||
|
|
||
|  **Description:** Sets the **tensor-parallel size**. This should typically match the number of GPUs you intend to use for a single model instance. | ||
|
|
||
| #### `--ep_size` | ||
|
|
||
|  **Description:** Sets the **expert-parallel size** for Mixture-of-Experts (MoE) models. Like `tp_size`, this should generally match the number of GPUs you're using. This setting has no effect on non-MoE models. | ||
|
|
||
| #### `--kv_cache_free_gpu_memory_fraction` |
There was a problem hiding this comment.
🛠️ Refactor suggestion
Fix heading-level jump to satisfy MD001 and improve skimming
You jump from ## to #### and skip ###, triggering markdown-lint and hurting outline structure.
Apply the same level to all flag descriptions, e.g.:
-#### `--tp_size`
+### `--tp_size`
...
-#### `--ep_size`
+### `--ep_size`Repeat for every parameter subsection.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| ## Configs and Parameters | |
| These options are used directly on the command line when you start the `trtllm-serve` process. | |
| #### `--tp_size` | |
|  **Description:** Sets the **tensor-parallel size**. This should typically match the number of GPUs you intend to use for a single model instance. | |
| #### `--ep_size` | |
|  **Description:** Sets the **expert-parallel size** for Mixture-of-Experts (MoE) models. Like `tp_size`, this should generally match the number of GPUs you're using. This setting has no effect on non-MoE models. | |
| #### `--kv_cache_free_gpu_memory_fraction` | |
| ## Configs and Parameters | |
| These options are used directly on the command line when you start the `trtllm-serve` process. | |
| ### `--tp_size` | |
|  **Description:** Sets the **tensor-parallel size**. This should typically match the number of GPUs you intend to use for a single model instance. | |
| ### `--ep_size` | |
|  **Description:** Sets the **expert-parallel size** for Mixture-of-Experts (MoE) models. Like `tp_size`, this should generally match the number of GPUs you're using. This setting has no effect on non-MoE models. | |
| #### `--kv_cache_free_gpu_memory_fraction` |
🧰 Tools
🪛 markdownlint-cli2 (0.17.2)
98-98: Heading levels should only increment by one level at a time
Expected: h3; Actual: h4
(MD001, heading-increment)
🤖 Prompt for AI Agents
In examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
around lines 95 to 106, the markdown headings jump from level 2 (##) directly to
level 4 (####), skipping level 3 (###), which breaks the document outline and
triggers markdown-lint errors. To fix this, change all parameter description
headings from #### to ### so that the heading levels increase sequentially and
maintain a proper outline structure.
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
Outdated
Show resolved
Hide resolved
svc-trtllm-gh-bot
added
the
Community want to contribute
|
@jamieliNVIDIA I think you need to run
|
| @@ -0,0 +1,367 @@ | |||
| # Deployment Guide for TensorRT-LLM Llama3.3 70B FP8 and NVFP4 | |||
There was a problem hiding this comment.
Hi Could we rename this file name to deployment-guide-for-trt-llm-llama3.3-70b? The purpose of such change is to make it easy to be indexed in future.
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
Outdated
Show resolved
Hide resolved
svc-trtllm-gh-bot
added
the
Community want to contribute
examples/models/core/llama3_3/Deployment Guide for TRT-LLM + Llama3.3 70B.md
Show resolved
Hide resolved
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (6)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md (6)
1-1: Tighten the document title
Two consecutive spaces appear between “TensorRT-LLM” and “Llama3.3”. Trim the extra space for cleaner rendering.-# Deployment Guide for TensorRT-LLM Llama3.3 70B FP8 and NVFP4 +# Deployment Guide for TensorRT-LLM Llama 3.3 70B FP8 and NVFP4
98-185: Heading level jumps (#### → h4) violate MD001
Inside “Configs and Parameters” (h2) the option sections start at h4 (####). Increment only one level (use###). This also fixes markdown-lint warnings.-#### `--tp_size` +### `--tp_size` ... -#### `--ep_size` +### `--ep_size`
212-214: Add language identifier to JSON block
Markdown-lint flags missing language spec.-```json +```json
248-253: Specify language for fenced result blocks
Lines 248–253 are rendered as plain code; addtextornoneto silence MD040.-``` +```text
326-353: Benchmark output block missing language tag
Same MD040 issue as above; usetext.-``` +```text
217-223: Spell out acronym on first use
“OOM” is used without expansion. Consider “out-of-memory (OOM)” on first mention for clarity.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md(1 hunks)
🧰 Additional context used
🧠 Learnings (3)
📓 Common learnings
Learnt from: yibinl-nvidia
PR: NVIDIA/TensorRT-LLM#6506
File: examples/models/core/mixtral/requirements.txt:3-3
Timestamp: 2025-08-01T15:14:45.630Z
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.
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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.
Learnt from: CR
PR: NVIDIA/TensorRT-LLM#0
File: CODING_GUIDELINES.md:0-0
Timestamp: 2025-08-01T07:34:42.734Z
Learning: Applies to **/*.py : The code developed for TensorRT-LLM should conform to Python 3.8+.
📚 Learning: in tensorrt-llm, examples directory can have different dependency versions than the root requirement...
Learnt from: yibinl-nvidia
PR: NVIDIA/TensorRT-LLM#6506
File: examples/models/core/mixtral/requirements.txt:3-3
Timestamp: 2025-08-01T15:14:45.630Z
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:
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
📚 Learning: in tensorrt-llm testing, it's common to have both cli flow tests (test_cli_flow.py) and pytorch api ...
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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:
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
🪛 LanguageTool
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
[style] ~117-~117: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...ch** backend. #### --max_batch_size Description: The maximum number of ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~121-~121: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...r processing. #### --max_num_tokens Description: The maximum total numb...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~125-~125: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...scheduled batch. #### --max_seq_len Description: The maximum possible s...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~129-~129: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...ut tokens. #### --trust_remote_code Description: Allows TensorRT-LLM to...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~144-~144: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...: Sets the data type for the KV cache. Default: auto (uses the data ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~148-~148: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...checkpoint). #### cuda_graph_config Description: A section for configur...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~158-~158: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...or which a CUDA graph will be created. Default: 0 **Rec...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~162-~162: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...-max_batch_size command-line option. batch_sizes: A specific list of ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~164-~164: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...batch sizes to create CUDA graphs for. Default: None #### `moe_conf...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~174-~174: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...The backend to use for MoE operations. Default: CUTLASS #### `atten...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~180-~180: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...end to use for attention calculations. Default: TRTLLM See the [https://g...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~305-~305: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...l,e2el" done EOF chmod +x bench.sh ``` To benchmark the FP4 model, replace --mod...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
🪛 markdownlint-cli2 (0.17.2)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
98-98: Heading levels should only increment by one level at a time
Expected: h3; Actual: h4
(MD001, heading-increment)
240-240: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
248-248: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
326-326: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🔇 Additional comments (1)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md (1)
25-30: Inconsistent precision naming: “FP4” vs “NVFP4”
The doc introduces “NVFP4” in the title and notes (Line 29) but lists the model link as…-FP4(Line 26). Pick one term and use it consistently (TensorRT-LLM docs usually say NVFP4).
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
Outdated
Show resolved
Hide resolved
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
Show resolved
Hide resolved
There was a problem hiding this comment.
Actionable comments posted: 0
♻️ Duplicate comments (1)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md (1)
67-69: Hard-coding KV cache dtype causes mismatch for NVFP4 users
dtype: fp8forces the server to treat every checkpoint as FP8. This breaks NVFP4 flows and contradicts the later documentation that says the default should beauto.
🧹 Nitpick comments (2)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md (2)
93-100: Heading level jumps from H2 to H4 – use H3 insteadThe subsections
#### --tp_size,#### --ep_size, etc. are direct children of “## Configs and Parameters”.
Jumping two levels violates MD001 and collapses the section hierarchy in rendered HTML.
Change####to###(or add an intermediate H3).
238-246: Add language identifier to fenced code blocks for lint complianceSeveral fenced blocks are missing a language tag (
```instead of```shell/```text), triggering MD040 and reducing syntax-highlighting:
- GSM8K FP8 sample (238–246)
- GSM8K FP4 sample (246–253)
- Benchmark sample output (324–332)
Add an appropriate identifier (
text,none, etc.) to silence markdownlint and improve readability.Also applies to: 324-332
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md(1 hunks)
🧰 Additional context used
🧠 Learnings (3)
📓 Common learnings
Learnt from: yibinl-nvidia
PR: NVIDIA/TensorRT-LLM#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.
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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.
Learnt from: CR
PR: NVIDIA/TensorRT-LLM#0
File: CODING_GUIDELINES.md:0-0
Timestamp: 2025-08-01T07:34:42.734Z
Learning: Applies to **/*.py : The code developed for TensorRT-LLM should conform to Python 3.8+.
📚 Learning: in tensorrt-llm, examples directory can have different dependency versions than the root requirement...
Learnt from: yibinl-nvidia
PR: NVIDIA/TensorRT-LLM#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:
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
📚 Learning: in tensorrt-llm testing, it's common to have both cli flow tests (test_cli_flow.py) and pytorch api ...
Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#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:
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
🪛 LanguageTool
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
[style] ~115-~115: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...ch** backend. #### --max_batch_size Description: The maximum number of ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~119-~119: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...r processing. #### --max_num_tokens Description: The maximum total numb...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~123-~123: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...scheduled batch. #### --max_seq_len Description: The maximum possible s...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~127-~127: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...ut tokens. #### --trust_remote_code Description: Allows TensorRT-LLM to...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~142-~142: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...: Sets the data type for the KV cache. Default: auto (uses the data ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~146-~146: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...checkpoint). #### cuda_graph_config Description: A section for configur...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~156-~156: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...or which a CUDA graph will be created. Default: 0 **Rec...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~160-~160: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...-max_batch_size command-line option. batch_sizes: A specific list of ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~162-~162: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...batch sizes to create CUDA graphs for. Default: None #### `moe_conf...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~172-~172: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...The backend to use for MoE operations. Default: CUTLASS #### `atten...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~178-~178: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...end to use for attention calculations. Default: TRTLLM See the [https://g...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
[style] ~303-~303: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...l,e2el" done EOF chmod +x bench.sh ``` To benchmark the FP4 model, replace --mod...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
🪛 markdownlint-cli2 (0.17.2)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
96-96: Heading levels should only increment by one level at a time
Expected: h3; Actual: h4
(MD001, heading-increment)
238-238: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
246-246: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
324-324: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🔇 Additional comments (1)
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md (1)
15-18: Prerequisites claim Hopper support but NVFP4 is Blackwell-only – please clarifyLine 15 lists “NVIDIA Blackwell or Hopper Architecture”, yet line 27 states “NVFP4 is only supported on NVIDIA Blackwell”.
Either:
- Split the prerequisite bullet into two (FP8 → Hopper/Blackwell, NVFP4 → Blackwell-only), or
- Add a clarifying note to avoid confusing users who have Hopper hardware and intend to try NVFP4.
Also applies to: 27-28
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
Show resolved
Hide resolved
examples/models/core/llama3_3/deployment-guide-for-trt-llm-llama3.3-70b.md
Show resolved
Hide resolved
examples/models/core/llama3_3/ Quick Start Recipe for TensorRT-LLM Llama3.3 70B FP8 and NVFP4
Outdated
Show resolved
Hide resolved
…ck Start Recipe for TensorRT-LLM Llama3.3 70B FP8 and NVFP4 Signed-off-by: jamieliNVIDIA <[email protected]>
…-LLM Llama3.3 70B FP8 and NVFP4 Signed-off-by: Tao Li @ NVIDIA <[email protected]>
nv-guomingz
force-pushed
the
feat/llama-3.3-deployment-guide
branch
from
33d698e to
1c62b30
Compare
|
/bot skip --comment "docs only change" |
|
PR_Github #14395 [ skip ] triggered by Bot |
|
PR_Github #14395 [ skip ] completed with state |
7 participants
Summary by CodeRabbit
Description
Created deployment guide for Llama 3.3 70B FP8 and NVFP4
Test Coverage
N/A
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 [--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.