{ "target": "openbmb/MiniCPM5-2B", "steps": 4600, "block_size": 8, "taps": [ 1, 10, 20, 30, 39 ], "mask_token_id": 75982, "num_blocks": 4, "seq_len": 1536, "data_mix_p1": [ "HuggingFaceH4/ultrachat_200k", "open-r1/OpenR1-Math-220k", "ise-uiuc/Magicoder-Evol-Instruct-110K", "HuggingFaceTB/smoltalk", "HuggingFaceFW/fineweb-edu" ], "data_mix_p2": [ "ise-uiuc/Magicoder-Evol-Instruct-110K", "HuggingFaceTB/smoltalk", "bigcode/self-oss-instruct-sc2-exec-filter-50k", "HuggingFaceH4/ultrachat_200k", "open-r1/OpenR1-Math-220k" ], "measured": { "domains": { "coding": { "n": 6, "plain_tok_s": 17.81, "spec_tok_s": 19.36, "speedup": 1.087, "acceptance_length": 1.234, "greedy_identical": "6/6", "per_prompt": [ { "prompt_tokens": 327, "new_tokens": 256, "plain_tok_s": 17.61, "spec_tok_s": 19.83, "speedup": 1.126, "al": 1.269, "identical": true }, { "prompt_tokens": 33, "new_tokens": 256, "plain_tok_s": 17.98, "spec_tok_s": 19.58, "speedup": 1.089, "al": 1.25, "identical": true }, { "prompt_tokens": 113, "new_tokens": 256, "plain_tok_s": 17.85, "spec_tok_s": 19.21, "speedup": 1.076, "al": 1.232, "identical": true }, { "prompt_tokens": 199, "new_tokens": 256, "plain_tok_s": 17.75, "spec_tok_s": 19.74, "speedup": 1.112, "al": 1.256, "identical": true }, { "prompt_tokens": 38, "new_tokens": 256, "plain_tok_s": 17.79, "spec_tok_s": 18.6, "speedup": 1.045, "al": 1.175, "identical": true }, { "prompt_tokens": 61, "new_tokens": 256, "plain_tok_s": 17.89, "spec_tok_s": 19.25, "speedup": 1.076, "al": 1.226, "identical": true } ], "raw": { "plain_tok": 1536, "plain_t": 86.24226474761963, "spec_tok": 1536, "spec_t": 79.34601759910583, "ident": 6, "n": 6, "al_sum": 1530.0, "al_n": 1240 }, "projected": 1.005 }, "agentic": { "n": 6, "plain_tok_s": 17.98, "spec_tok_s": 19.94, "speedup": 1.109, "acceptance_length": 1.267, "greedy_identical": "5/6", "per_prompt": [ { "prompt_tokens": 24, "new_tokens": 256, "plain_tok_s": 17.96, "spec_tok_s": 18.77, "speedup": 1.045, "al": 1.192, "identical": true }, { "prompt_tokens": 26, "new_tokens": 256, "plain_tok_s": 18.05, "spec_tok_s": 20.06, "speedup": 1.111, "al": 1.262, "identical": true }, { "prompt_tokens": 40, "new_tokens": 256, "plain_tok_s": 18.01, "spec_tok_s": 19.83, "speedup": 1.101, "al": 1.256, "identical": true }, { "prompt_tokens": 64, "new_tokens": 27, "plain_tok_s": 17.38, "spec_tok_s": 22.75, "speedup": 1.309, "al": 1.444, "identical": true }, { "prompt_tokens": 30, "new_tokens": 28, "plain_tok_s": 17.9, "spec_tok_s": 22.04, "speedup": 1.231, "al": 1.421, "identical": true }, { "prompt_tokens": 34, "new_tokens": 256, "plain_tok_s": 17.97, "spec_tok_s": 20.74, "speedup": 1.154, "al": 1.335, "identical": false } ], "raw": { "plain_tok": 1079, "plain_t": 60.022462368011475, "spec_tok": 1079, "spec_t": 54.1166627407074, "ident": 5, "n": 6, "al_sum": 1073.0, "al_n": 847 }, "projected": 1.032 }, "math": { "n": 6, "plain_tok_s": 17.91, "spec_tok_s": 19.28, "speedup": 1.076, "acceptance_length": 1.226, "greedy_identical": "5/6", "per_prompt": [ { "prompt_tokens": 71, "new_tokens": 256, "plain_tok_s": 17.93, "spec_tok_s": 17.57, "speedup": 0.98, "al": 1.118, "identical": true }, { "prompt_tokens": 35, "new_tokens": 256, "plain_tok_s": 17.83, "spec_tok_s": 19.91, "speedup": 1.116, "al": 1.269, "identical": true }, { "prompt_tokens": 58, "new_tokens": 256, "plain_tok_s": 17.93, "spec_tok_s": 19.47, "speedup": 1.086, "al": 1.25, "identical": false }, { "prompt_tokens": 43, "new_tokens": 256, "plain_tok_s": 17.8, "spec_tok_s": 19.72, "speedup": 1.108, "al": 1.256, "identical": true }, { "prompt_tokens": 116, "new_tokens": 256, "plain_tok_s": 18.0, "spec_tok_s": 20.6, "speedup": 1.144, "al": 1.301, "identical": true }, { "prompt_tokens": 61, "new_tokens": 256, "plain_tok_s": 18.0, "spec_tok_s": 18.69, "speedup": 1.038, "al": 1.181, "identical": true } ], "raw": { "plain_tok": 1536, "plain_t": 85.73986506462097, "spec_tok": 1536, "spec_t": 79.68685555458069, "ident": 5, "n": 6, "al_sum": 1530.0, "al_n": 1248 }, "projected": 0.998 }, "general": { "n": 6, "plain_tok_s": 17.9, "spec_tok_s": 19.26, "speedup": 1.076, "acceptance_length": 1.228, "greedy_identical": "5/6", "per_prompt": [ { "prompt_tokens": 66, "new_tokens": 256, "plain_tok_s": 17.92, "spec_tok_s": 20.38, "speedup": 1.137, "al": 1.294, "identical": true }, { "prompt_tokens": 99, "new_tokens": 256, "plain_tok_s": 18.02, "spec_tok_s": 20.12, "speedup": 1.116, "al": 1.269, "identical": false }, { "prompt_tokens": 143, "new_tokens": 256, "plain_tok_s": 17.96, "spec_tok_s": 19.69, "speedup": 1.096, "al": 1.244, "identical": true }, { "prompt_tokens": 330, "new_tokens": 106, "plain_tok_s": 17.81, "spec_tok_s": 18.36, "speedup": 1.03, "al": 1.193, "identical": true }, { "prompt_tokens": 93, "new_tokens": 256, "plain_tok_s": 17.79, "spec_tok_s": 18.64, "speedup": 1.048, "al": 1.192, "identical": true }, { "prompt_tokens": 106, "new_tokens": 256, "plain_tok_s": 17.82, "spec_tok_s": 18.08, "speedup": 1.014, "al": 1.164, "identical": true } ], "raw": { "plain_tok": 1386, "plain_t": 77.45153427124023, "spec_tok": 1386, "spec_t": 71.95891618728638, "ident": 5, "n": 6, "al_sum": 1380.0, "al_n": 1124 }, "projected": 1.0 } }, "overall": { "n": 24, "plain_tok_s": 17.89, "spec_tok_s": 19.42, "acceptance_length": 1.236, "greedy_identical": "21/24", "speedup": 1.086, "projected": 1.006 }, "costs": { "target_decode_ms": 57.714, "target_verify8_ms": 55.433, "draft_block_ms": 15.442 }, "gates_met": false, "steps_trained": 4600, "targets": { "coding": 2.0, "agentic": 2.0, "overall": 1.5 }, "device": "Tesla T4", "validation_mode": false } }