TensorRT-LLMs/examples/trtllm-eval/README.md
Enwei Zhu fc7a81ceb0
test: Add LLGuidance test and refine guided decoding (#5348)
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
2025-06-25 14:12:56 +08:00

1.5 KiB

Accuracy Evaluation Tool trtllm-eval

We provide a CLI tool trtllm-eval for evaluating model accuracy. It shares the core evaluation logics with the accuracy test suite of TensorRT-LLM.

trtllm-eval is built on the offline API -- LLM API. It provides developers a unified entrypoint for accuracy evaluation. Compared with the online API trtllm-serve, offline API provides clearer error messages and simplifies the debugging workflow.

trtllm-eval follows the CLI interface of trtllm-serve.

pip install -r requirements.txt

# Evaluate Llama-3.1-8B-Instruct on MMLU
trtllm-eval --model meta-llama/Llama-3.1-8B-Instruct mmlu

# Evaluate Llama-3.1-8B-Instruct on GSM8K
trtllm-eval --model meta-llama/Llama-3.1-8B-Instruct gsm8k

# Evaluate Llama-3.3-70B-Instruct on GPQA Diamond
trtllm-eval --model meta-llama/Llama-3.3-70B-Instruct gpqa_diamond

The --model argument accepts either a Hugging Face model ID or a local checkpoint path. By default, trtllm-eval runs the model with the PyTorch backend; pass --backend tensorrt to switch to the TensorRT backend. Alternatively, the --model argument also accepts a local path to pre-built TensorRT engines; in that case, please pass the Hugging Face tokenizer path to the --tokenizer argument.

See more details by trtllm-eval --help.