| 1 | [ |
| 2 | { |
| 3 | "test_name": "latency_llama8B_tp1", |
| 4 | "environment_variables": { |
| 5 | "PT_HPU_LAZY_MODE": 1, |
| 6 | "VLLM_CONTIGUOUS_PA": 1, |
| 7 | "VLLM_DEFRAG": 1 |
| 8 | }, |
| 9 | "parameters": { |
| 10 | "model": "meta-llama/Meta-Llama-3.1-8B-Instruct", |
| 11 | "tensor_parallel_size": 1, |
| 12 | "load_format": "dummy", |
| 13 | "num-iters-warmup": 5, |
| 14 | "num-iters": 15, |
| 15 | "max-model-len": 256, |
| 16 | "async-scheduling": "" |
| 17 | } |
| 18 | }, |
| 19 | { |
| 20 | "test_name": "latency_llama70B_tp4", |
| 21 | "environment_variables": { |
| 22 | "PT_HPU_LAZY_MODE": 1, |
| 23 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 24 | "VLLM_CONTIGUOUS_PA": 1, |
| 25 | "VLLM_DEFRAG": 1 |
| 26 | }, |
| 27 | "parameters": { |
| 28 | "model": "meta-llama/Meta-Llama-3.1-70B-Instruct", |
| 29 | "tensor_parallel_size": 4, |
| 30 | "load_format": "dummy", |
| 31 | "num-iters-warmup": 5, |
| 32 | "num-iters": 15, |
| 33 | "max-model-len": 256, |
| 34 | "async-scheduling": "" |
| 35 | } |
| 36 | }, |
| 37 | { |
| 38 | "test_name": "latency_mixtral8x7B_tp2", |
| 39 | "environment_variables": { |
| 40 | "PT_HPU_LAZY_MODE": 1, |
| 41 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 42 | "VLLM_CONTIGUOUS_PA": 1, |
| 43 | "VLLM_DEFRAG": 1 |
| 44 | }, |
| 45 | "parameters": { |
| 46 | "model": "mistralai/Mixtral-8x7B-Instruct-v0.1", |
| 47 | "tensor_parallel_size": 2, |
| 48 | "load_format": "dummy", |
| 49 | "num-iters-warmup": 5, |
| 50 | "num-iters": 15, |
| 51 | "max-model-len": 256, |
| 52 | "async-scheduling": "" |
| 53 | } |
| 54 | }, |
| 55 | { |
| 56 | "test_name": "latency_deepseek_r1", |
| 57 | "environment_variables": { |
| 58 | "PT_HPU_LAZY_MODE": 1, |
| 59 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 60 | "VLLM_CONTIGUOUS_PA": 1, |
| 61 | "VLLM_DEFRAG": 1 |
| 62 | }, |
| 63 | "parameters": { |
| 64 | "model": "deepseek-ai/DeepSeek-R1", |
| 65 | "tensor_parallel_size": 8, |
| 66 | "load_format": "dummy", |
| 67 | "max-model-len": 2048, |
| 68 | "dtype": "bfloat16" |
| 69 | } |
| 70 | }, |
| 71 | { |
| 72 | "test_name": "latency_llama4_maverick_17b128e_instruct_fp8", |
| 73 | "environment_variables": { |
| 74 | "PT_HPU_LAZY_MODE": 1, |
| 75 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 76 | "VLLM_CONTIGUOUS_PA": 1, |
| 77 | "VLLM_DEFRAG": 1 |
| 78 | }, |
| 79 | "parameters": { |
| 80 | "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", |
| 81 | "tensor_parallel_size": 8, |
| 82 | "max-model-len": 512, |
| 83 | "max-num-seqs": 128, |
| 84 | "async-scheduling": "", |
| 85 | "gpu-memory-utilization": 0.95, |
| 86 | "enable_expert_parallel": "" |
| 87 | } |
| 88 | }, |
| 89 | { |
| 90 | "test_name": "latency_qwen3_8b", |
| 91 | "environment_variables": { |
| 92 | "PT_HPU_LAZY_MODE": 1, |
| 93 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 94 | "VLLM_CONTIGUOUS_PA": 1, |
| 95 | "VLLM_DEFRAG": 1 |
| 96 | }, |
| 97 | "parameters": { |
| 98 | "model": "Qwen/Qwen3-8B", |
| 99 | "tensor_parallel_size": 1, |
| 100 | "max-model-len": 2048, |
| 101 | "max-num-seqs": 128, |
| 102 | "dtype": "bfloat16", |
| 103 | "async-scheduling": "" |
| 104 | } |
| 105 | } |
| 106 | ] |