{"tag": "leader", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": true, "quantization": null, "enforce_eager": false, "init_s": 53.99, "rss_mb": 8470.9, "rss_peak_mb": 8494.8, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.176, "pp_tps": 2909.7, "tg_tokens": 128, "tg_s": 2.288, "tg_tps": 55.51, "raw": {"pp": [0.18308231001719832, 0.17596286674961448, 0.21423810021951795], "tg": [2.3049342604354024, 2.3430285179056227, 2.3147874921560287], "tg1": [0.017149847000837326, 0.016962516121566296, 0.019108522683382034]}}
{"tag": "leader-nosplitkv", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": true, "quantization": null, "enforce_eager": false, "init_s": 78.64, "rss_mb": 8450.2, "rss_peak_mb": 8473.2, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.1661, "pp_tps": 3082.72, "tg_tokens": 128, "tg_s": 2.0036, "tg_tps": 63.39, "raw": {"pp": [0.1672560372389853, 0.1660872590728104, 0.18195243133231997], "tg": [2.0187266608700156, 2.078219319693744, 2.063376694917679], "tg1": [0.016000438015908003, 0.01626650895923376, 0.015106705948710442]}}
{"tag": "leader-nodetok", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": false, "quantization": null, "enforce_eager": false, "init_s": 51.93, "rss_mb": 8471.4, "rss_peak_mb": 8489.3, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.1661, "pp_tps": 3082.69, "tg_tokens": 128, "tg_s": 2.0161, "tg_tps": 62.99, "raw": {"pp": [0.16608853498473763, 0.16695645125582814, 0.17180946189910173], "tg": [2.0667635132558644, 2.0320902089588344, 2.0568131166510284], "tg1": [0.016675901133567095, 0.01703286776319146, 0.015959253069013357]}}
{"tag": "leader", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": true, "quantization": null, "enforce_eager": false, "init_s": 53.03, "rss_mb": 8450.2, "rss_peak_mb": 8474.0, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.1678, "pp_tps": 3051.95, "tg_tokens": 128, "tg_s": 2.0026, "tg_tps": 63.42, "raw": {"pp": [0.1677613421343267, 0.17518142703920603, 0.17216769512742758], "tg": [2.0181680670939386, 2.045619810000062, 2.017581162042916], "tg1": [0.014953931327909231, 0.015587549190968275, 0.01536224177107215]}}
{"tag": "leader-nosplitkv", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": true, "quantization": null, "enforce_eager": false, "init_s": 52.42, "rss_mb": 8449.9, "rss_peak_mb": 8474.8, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.1495, "pp_tps": 3425.09, "tg_tokens": 128, "tg_s": 1.7717, "tg_tps": 71.68, "raw": {"pp": [0.1503864680416882, 0.1627977816388011, 0.1494850511662662], "tg": [1.7899064039811492, 1.7877749968320131, 1.7861353252083063], "tg1": [0.01460887212306261, 0.014465874060988426, 0.014931404031813145]}}
{"tag": "leader-nodetok", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": false, "quantization": null, "enforce_eager": false, "init_s": 71.6, "rss_mb": 8448.7, "rss_peak_mb": 8472.9, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.6363, "pp_tps": 804.61, "tg_tokens": 128, "tg_s": 3.9901, "tg_tps": 31.83, "raw": {"pp": [0.6363332280889153, 0.6899973549880087, 0.9538670806214213], "tg": [10.062090303748846, 11.140456527005881, 4.0063140746206045], "tg1": [0.01683283317834139, 0.01622168067842722, 0.01864417502656579]}}
{"tag": "leader", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": true, "quantization": null, "enforce_eager": false, "init_s": 51.31, "rss_mb": 8456.8, "rss_peak_mb": 8480.4, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.1517, "pp_tps": 3374.74, "tg_tokens": 128, "tg_s": 1.7671, "tg_tps": 71.87, "raw": {"pp": [0.15171556919813156, 0.15706020314246416, 0.154549116268754], "tg": [1.782004885841161, 1.8077111751772463, 1.8367386371828616], "tg1": [0.017101087141782045, 0.014878218993544579, 0.015195842832326889]}}
{"tag": "leader-nosplitkv", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": true, "quantization": null, "enforce_eager": false, "init_s": 51.44, "rss_mb": 8442.5, "rss_peak_mb": 8467.3, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.154, "pp_tps": 3324.69, "tg_tokens": 128, "tg_s": 1.7687, "tg_tps": 71.8, "raw": {"pp": [0.15424893703311682, 0.15522084012627602, 0.15399918612092733], "tg": [1.8099028407596052, 1.8291758550330997, 1.7837468180805445], "tg1": [0.015516853891313076, 0.015281836967915297, 0.015004676301032305]}}
{"tag": "leader-nodetok", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-gptq-g128", "dtype": "bfloat16", "compilation_config": null, "detokenize": false, "quantization": null, "enforce_eager": false, "init_s": 50.8, "rss_mb": 8449.5, "rss_peak_mb": 8473.7, "threads_bind": "32-55", "omp_num_threads": "24", "pp_tokens": 512, "pp_s": 0.1511, "pp_tps": 3389.54, "tg_tokens": 128, "tg_s": 1.7391, "tg_tps": 73.03, "raw": {"pp": [0.1712735160253942, 0.15105291502550244, 0.15270386030897498], "tg": [1.7530193431302905, 1.7823738330043852, 1.7983860420063138], "tg1": [0.014007281046360731, 0.014001580886542797, 0.013904284220188856]}}