LLM Generate Distributed
Source https://github.com/NVIDIA/TensorRT-LLM/tree/main/examples/llm-api/llm_generate_distributed.py.
1### Distributed LLM Generation
2from tensorrt_llm import LLM, SamplingParams
3
4
5def main():
6 # model could accept HF model name or a path to local HF model.
7 llm = LLM(
8 model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
9 # Distributed settings
10 tensor_parallel_size=2,
11 )
12
13 # Sample prompts.
14 prompts = [
15 "Hello, my name is",
16 "The president of the United States is",
17 "The capital of France is",
18 "The future of AI is",
19 ]
20
21 # Create a sampling params.
22 sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
23
24 for output in llm.generate(prompts, sampling_params):
25 print(
26 f"Prompt: {output.prompt!r}, Generated text: {output.outputs[0].text!r}"
27 )
28
29 # Got output like
30 # Prompt: 'Hello, my name is', Generated text: '\n\nJane Smith. I am a student pursuing my degree in Computer Science at [university]. I enjoy learning new things, especially technology and programming'
31 # Prompt: 'The president of the United States is', Generated text: 'likely to nominate a new Supreme Court justice to fill the seat vacated by the death of Antonin Scalia. The Senate should vote to confirm the'
32 # Prompt: 'The capital of France is', Generated text: 'Paris.'
33 # Prompt: 'The future of AI is', Generated text: 'an exciting time for us. We are constantly researching, developing, and improving our platform to create the most advanced and efficient model available. We are'
34
35
36# Due to the requirement of the underlying mpi4py, for multi-gpu, the main function must be placed inside the
37# `if __name__ == '__main__':` block.
38# Refer to https://mpi4py.readthedocs.io/en/stable/mpi4py.futures.html#mpipoolexecutor
39if __name__ == '__main__':
40 main()