TensorRT-LLMs/examples/mixtral

Mixtral

This document shows how to build and run a Mixtral model in TensorRT-LLM on both single GPU, single node multi-GPU and multi-node multi-GPU.

Overview

The TensorRT-LLM Mixtral implementation is based on the LLaMA model, with Mixture of Experts enabled. The implementation can be found in tensorrt_llm/models/llama/model.py. See the LLaMA example examples/llama for details.

Build TensorRT engine(s)

Get the weights by downloading from HF https://huggingface.co/mistralai/Mixtral-8x7B-v0.1. See also https://huggingface.co/docs/transformers/main/en/model_doc/mixtral

pip install -r requirements.txt # install latest version of transformers, needed for Mixtral

git lfs install
git clone https://huggingface.co/mistralai/Mixtral-8x7B-v0.1

We use the LLaMA build.py script to build the model. TensorRT-LLM LLaMA builds TensorRT engine(s) from HF checkpoint provided by --model_dir. If no checkpoint directory is specified, TensorRT-LLM will build engine(s) with dummy weights. Note that when loading Mixtral weights you must use the --load_by_shard option

--use_inflight_batching is a shortcut for --use_gpt_attention_plugin, --remove_input_padding and --paged_kv_cache

build.py uses one GPU by default, but if you have already more GPUs available at build time, you may enable parallel builds to make the engine building process faster by adding the --parallel_build argument.

Here are some examples:

# Build Mixtral8x7B with pipeline parallelism
python ../llama/build.py --model_dir ./Mixtral-8x7B-v0.1 \
                --use_inflight_batching \
                --enable_context_fmha \
                --use_gemm_plugin \
                --world_size 2 \
                --pp_size 2 \
                --output_dir ./trt_engines/mixtral/PP


# Build Mixtral8x7B with tensor parallelism
python ../llama/build.py --model_dir ./Mixtral-8x7B-v0.1 \
                --use_inflight_batching \
                --enable_context_fmha \
                --use_gemm_plugin \
                --world_size 2 \
                --tp_size 2 \
                --output_dir ./trt_engines/mixtral/TP

Then, you can test your engine with the run.py script:

mpirun -n 2 python3 ../run.py --engine_dir ./trt_engines/mixtral/TP --tokenizer_dir ./Mixtral-8x7B-v0.1 --max_output_len 8 --input_text "I love french quiche"

For more examples see examples/llama/README.md