| 1 | [ |
| 2 | { |
| 3 | "test_name": "latency_llama8B_tp2", |
| 4 | "environment_variables": { |
| 5 | "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, |
| 6 | "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, |
| 7 | "VLLM_CPU_SGL_KERNEL": 1, |
| 8 | "VLLM_CPU_KVCACHE_SPACE": 40 |
| 9 | }, |
| 10 | "parameters": { |
| 11 | "model": "meta-llama/Llama-3.1-8B-Instruct", |
| 12 | "tensor_parallel_size": 2, |
| 13 | "dtype": "bfloat16", |
| 14 | "distributed_executor_backend": "mp", |
| 15 | "block_size": 128, |
| 16 | "trust_remote_code": "", |
| 17 | "disable_log_stats": "", |
| 18 | "enforce_eager": "", |
| 19 | "max_num_batched_tokens": 2048, |
| 20 | "max_num_seqs": 256, |
| 21 | "num_iters_warmup": 5, |
| 22 | "num_iters": 15 |
| 23 | } |
| 24 | } |
| 25 | ] |