[TRTLLM-5990][doc] trtllm-serve doc improvement.

[nv-guomingz]

This PR made below changes

• Modifying the Quickstart.md by cherry-picking doc: Include NGC release containers in quick-start-guide.md #5334 and adding explicit instructions on launch docker container
• Adjust the trtllm-serve part
  • Remove the benchmark part(using genai) from original trtllm-serve.rst
  • Add the e2e example for using trtllm-serve plus benchmarking_serving.py to benchmark llama 3.1 70B model
  • Add the brief introduction for extra_llm_api_options usage.

Summary by CodeRabbit

• Documentation
  • Added a comprehensive guide for benchmarking the Llama 3.1 70B model using the OpenAI-compatible API.
  • Updated the Quick Start Guide with detailed Docker usage instructions and clarified deployment steps for online serving.
  • Streamlined documentation by removing in-depth Model Definition API instructions and directing users to dedicated serve command documentation.
  • Improved navigation and organization by updating references and linking to new benchmarking resources.

[Copilot]

Pull Request Overview

This PR enhances the TensorRT-LLM documentation by clarifying Quick Start steps, refining the trtllm-serve section, and introducing a dedicated benchmarking guide.

• Added explicit Docker container launch instructions in the Quick Start Guide
• Updated trtllm-serve doc to focus on online serving and link to benchmarking
• Created trtllm-serve-bench.md for an end-to-end performance benchmarking workflow

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
docs/source/quick-start-guide.md	Added Docker run example, corrected trtllm-serve instructions, and cleaned up old content
docs/source/index.rst	Updated toctree entry to point at commands/trtllm-serve/index
docs/source/commands/trtllm-serve/trtllm-serve.rst	Removed inline benchmark section and added a note linking to the new bench doc
docs/source/commands/trtllm-serve/trtllm-serve-bench.md	New benchmarking guide for Llama 3.1 70B online serving performance

Comments suppressed due to low confidence (9)

docs/source/quick-start-guide.md:37

• The command name is inconsistent ("trtllm-server" vs "trtllm-serve"); update to use the correct "trtllm-serve" command.

> If you are running `trtllm-server` inside a Docker container, you have two options for sending API requests:

docs/source/quick-start-guide.md:23

• Empty link text is used. Provide descriptive link text, for example [LLM API docs](llm-api/index) and [LLM API examples](examples/llm_api_examples).

To learn more about the LLM API, check out the [](llm-api/index) and [](examples/llm_api_examples).

docs/source/commands/trtllm-serve/trtllm-serve.rst:192

• Backtick usage inside a link breaks syntax. Use [Performance Benchmarking with trtllm-serve](<URL>) without nested backticks.

Please refer to `Performance Benchmarking with `trtllm-serve` <https://github.com/NVIDIA/TensorRT-LLM/blob/main/docs/source/commands/trtllm-serve/trtllm-serve-bench.md>` for more details.

docs/source/commands/trtllm-serve/trtllm-serve-bench.md:48

• Replace the placeholder "max_num_tokens" with a concrete numeric example or clearly denote it as a placeholder (e.g., <max_num_tokens>).

    --max_num_tokens max_num_tokens \

docs/source/commands/trtllm-serve/trtllm-serve-bench.md:49

• Use an explicit integer for --max_seq_len (e.g., 2048) instead of shorthand "2k" to avoid ambiguity.

    --max_seq_len 2k \

docs/source/commands/trtllm-serve/trtllm-serve-bench.md:125

• Add a space after the period: "CUDA graphs. Default value is 0."

   * cuda_graph_max_batch_size: Maximum batch size for CUDA graphs.Default value is `0`.

docs/source/commands/trtllm-serve/trtllm-serve-bench.md:127

• Insert a space after the period and make "Value" lowercase: "enabled. Default value is True."

   * autotuner_enabled: Enable autotuner only when torch compile is enabled.Default Value is `True`.

docs/source/commands/trtllm-serve/trtllm-serve-bench.md:129

• Add a space after the colon: "attention_backend: Attention backend to use. Default value is TRTLLM."

   * attention_backend:Attention backend to use. Default value is `TRTLLM`.

docs/source/commands/trtllm-serve/trtllm-serve-bench.md:103

• The redirect syntax &< is invalid; likely intended &> output_bench.log to capture both stdout and stderr.

bash -x bench.sh &< output_bench.log

[nv-guomingz]

/bot run --stage-list "A10-Build_Docs"

[tensorrt-cicd]

PR_Github #13979 [ run ] triggered by Bot

[LinPoly]

LGTM

[tensorrt-cicd]

PR_Github #13979 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #10531 (Partly Tested) completed with status: 'SUCCESS'

[nv-guomingz]

/bot skip --comment "doc build phase already pass"

[tensorrt-cicd]

PR_Github #14080 [ skip ] triggered by Bot

[tensorrt-cicd]

PR_Github #14080 [ skip ] completed with state SUCCESS
Skipping testing for commit ed88f62