This document explains how to build the ChatGLM-6B model using TensorRT-LLM and run on a single GPU, a single node with multiple GPUs or multiple nodes with multiple GPUs.
The TensorRT-LLM ChatGLM-6B implementation can be found in
tensorrt_llm/models/chatglm6b/model.py
.
The TensorRT-LLM ChatGLM-6B example code is located in
examples/chatglm6b
. There are four main files in that folder:
hf_chatglm6b_convert.py
to convert a checkpoint from the HuggingFace (HF) Transformers format to the FasterTransformer (FT) format,build.py
to build the TensorRT engine(s) needed to run the ChatGLM-6B model,run.py
to run the inference on an input text,summarize.py
to summarize the articles in the cnn_dailymail dataset using the model.
The next two sections describe how to convert the weights from the HuggingFace (HF) Transformers format to the FT format. You can skip those two sections if you already have weights in the FT format.
Note, also, that if your weights are neither in HF Transformers nor in FT
formats, you will need to convert to the FT format. The script like
hf_chatglm6b_convert.py
can serve as a starting
point.
pip install -r requirements.txt
apt-get update
apt-get install git-lfs
git clone https://huggingface.co/THUDM/chatglm-6b pyTorchModel
TensorRT-LLM can directly load weights from FT. The
hf_chatglm6b_convert.py
script allows you to
convert weights from HF Tranformers format to FT format.
# beckup the original file
cp pyTorchModel/modeling_chatglm.py pyTorchModel/modeling_chatglm.py-backup
# replace the file with our edited version for exporting the weight of LM
cp modeling_chatglm.py pyTorchModel
# export weight of LM
python3 exportLM.py
# restore the original file for the later use (for example, summarize.py)
mv pyTorchModel/modeling_chatglm.py-backup pyTorchModel/modeling_chatglm.py
python3 hf_chatglm6b_convert.py -i pyTorchModel -o ftModel --tensor-parallelism 1 --storage-type fp16
TensorRT-LLM builds TensorRT engine(s) using a checkpoint in FT format. If no checkpoint directory is specified, TensorRT-LLM will build engine(s) using dummy weights. Note that the number of TensorRT engines depends on the number of GPUs that will be used to run inference.
The build.py
script requires a single GPU to build the TensorRT
engine(s). However, if you have more than one GPU in your system (of the same
model), you can enable parallel builds to accelerate the engine building
process. For that, add the --parallel_build
argument to the build command.
Please note that for the moment, the parallel_build
feature cannot take
advantage of more than a single node.
Examples of build invocations:
# Build a single-GPU float16 engine using FT weights.
# Enable the special TensorRT-LLM ChatGLM-6B Attention plugin (--use_gpt_attention_plugin) to increase runtime performance.
python3 build.py --model_dir=./ftModel/1-gpu \
--dtype float16 \
--use_gpt_attention_plugin float16 \
--use_gemm_plugin float16
You can enable the FMHA kernels for ChatGLM-6B by adding
--enable_context_fmha
to the invocation of build.py
. Note that it is
disabled by default because of possible accuracy issues due to the use of Flash
Attention.
To run a TensorRT-LLM ChatGLM-6B model on a single GPU, you can use python3
:
# Run the ChatGLM-6B model on a single GPU.
python3 run.py
The summarization can be done using the summarize.py
script as follows:
# Run the summarization task.
python3 summarize.py --engine_dir trtModel \
--test_hf \
--batch_size 1 \
--test_trt_llm \
--hf_model_location=pyTorchModel \
--check_accuracy \
--tensorrt_llm_rouge1_threshold=14
The TensorRT-LLM ChatGLM-6B benchmark is located in benchmarks/