| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | from vllm import LLM, SamplingParams |
| 5 | from vllm.config import KVTransferConfig |
| 6 | |
| 7 | |
| 8 | def read_prompts(): |
| 9 | """Read prompts from output.txt.""" |
| 10 | prompts = [] |
| 11 | try: |
| 12 | with open("output.txt") as f: |
| 13 | for line in f: |
| 14 | prompts.append(line.strip()) |
| 15 | print(f"Loaded {len(prompts)} prompts from output.txt") |
| 16 | return prompts |
| 17 | except FileNotFoundError: |
| 18 | print("Error: output.txt file not found") |
| 19 | exit(-1) |
| 20 | |
| 21 | |
| 22 | def main(): |
| 23 | prompts = read_prompts() |
| 24 | sampling_params = SamplingParams(temperature=0, top_p=0.95, max_tokens=10) |
| 25 | |
| 26 | llm = LLM( |
| 27 | model="meta-llama/Llama-3.2-1B-Instruct", |
| 28 | enforce_eager=True, |
| 29 | gpu_memory_utilization=0.8, |
| 30 | max_num_batched_tokens=64, |
| 31 | max_num_seqs=16, |
| 32 | kv_transfer_config=KVTransferConfig( |
| 33 | kv_connector="ExampleConnector", |
| 34 | kv_role="kv_both", |
| 35 | kv_connector_extra_config={"shared_storage_path": "local_storage"}, |
| 36 | ), |
| 37 | ) # , max_model_len=2048, max_num_batched_tokens=2048) |
| 38 | |
| 39 | # 1ST generation (prefill instance) |
| 40 | outputs = llm.generate(prompts, sampling_params) |
| 41 | |
| 42 | print("-" * 30) |
| 43 | for output in outputs: |
| 44 | prompt = output.prompt |
| 45 | generated_text = output.outputs[0].text |
| 46 | print(f"Prompt: {prompt!r}\nGenerated text: {generated_text!r}") |
| 47 | print("-" * 30) |
| 48 | |
| 49 | |
| 50 | if __name__ == "__main__": |
| 51 | main() |