{ "hardware": "4x NVIDIA RTX PRO 6000 Blackwell 96GB", "max_model_len": 524288, "model": "GLM-5.2", "mtp_speculative_tokens": 3, "prefix_cache_isolated": true, "results": [ { "completion_tokens": 64, "decode_tokens_per_s": 16.04, "end_to_end_s": 4.906, "end_to_end_tokens_per_s": 65.23, "prefill_tokens_per_s": 261.88, "prompt_tokens": 256, "total_context_tokens": 320, "ttft_s": 0.9775 }, { "completion_tokens": 64, "decode_tokens_per_s": 29.54, "end_to_end_s": 3.242, "end_to_end_tokens_per_s": 335.6, "prefill_tokens_per_s": 923.29, "prompt_tokens": 1024, "total_context_tokens": 1088, "ttft_s": 1.1091 }, { "completion_tokens": 64, "decode_tokens_per_s": 27.6, "end_to_end_s": 8.1344, "end_to_end_tokens_per_s": 511.41, "prefill_tokens_per_s": 699.93, "prompt_tokens": 4096, "total_context_tokens": 4160, "ttft_s": 5.8521 }, { "completion_tokens": 64, "decode_tokens_per_s": 18.74, "end_to_end_s": 26.378, "end_to_end_tokens_per_s": 623.55, "prefill_tokens_per_s": 711.83, "prompt_tokens": 16384, "total_context_tokens": 16448, "ttft_s": 23.0167 }, { "completion_tokens": 64, "decode_tokens_per_s": 21.55, "end_to_end_s": 99.831, "end_to_end_tokens_per_s": 657.11, "prefill_tokens_per_s": 676.28, "prompt_tokens": 65536, "total_context_tokens": 65600, "ttft_s": 96.9068 }, { "completion_tokens": 64, "decode_tokens_per_s": 17.18, "end_to_end_s": 196.5178, "end_to_end_tokens_per_s": 667.3, "prefill_tokens_per_s": 679.66, "prompt_tokens": 131072, "total_context_tokens": 131136, "ttft_s": 192.8504 }, { "completion_tokens": 64, "decode_tokens_per_s": 11.44, "end_to_end_s": 391.5731, "end_to_end_tokens_per_s": 669.63, "prefill_tokens_per_s": 679.01, "prompt_tokens": 262144, "total_context_tokens": 262208, "ttft_s": 386.0674 }, { "completion_tokens": 64, "decode_tokens_per_s": 7.59, "end_to_end_s": 770.5666, "end_to_end_tokens_per_s": 648.96, "prefill_tokens_per_s": 655.94, "prompt_tokens": 500000, "total_context_tokens": 500064, "ttft_s": 762.2693 } ], "vision": true }