{ "model": "models/gemma-4-e2b-it-4bit", "compute_dtype": "float32", "status": "cache_numerics_and_timing_probe_not_quality_benchmark", "results": [ { "state": "text", "prefix_tokens": 50, "suffix_lengths": [ 111, 99, 111, 111 ], "cached_serial_vs_uncached": { "max_logit_delta": 0.032517433166503906, "max_probability_delta": 4.89707640993764e-07, "all_winners_match": true }, "cached_batch_vs_uncached": { "max_logit_delta": 0.047650814056396484, "max_probability_delta": 9.921427801540261e-07, "all_winners_match": true }, "reversed_batch_vs_original": { "max_logit_delta": 0.0, "max_probability_delta": 0.0, "all_winners_match": true }, "prefix_unchanged": true, "branch_batch_sizes": [ 4 ], "warm_seconds": { "uncached": [ 0.24687987499055453, 0.24587425001664087, 0.24255950000951998 ], "cached_batched": [ 0.17461620797985233, 0.18541070900391787, 0.16993425000691786 ] }, "warm_median_seconds": { "uncached": 0.24587425001664087, "cached_batched": 0.17461620797985233 }, "speedup": 1.4080837790556675 }, { "state": "image", "prefix_tokens": 295, "suffix_lengths": [ 111, 99, 111, 111 ], "cached_serial_vs_uncached": { "max_logit_delta": 0.04843616485595703, "max_probability_delta": 0.0007440930557797154, "all_winners_match": true }, "cached_batch_vs_uncached": { "max_logit_delta": 0.02086162567138672, "max_probability_delta": 0.0013810108588161452, "all_winners_match": true }, "reversed_batch_vs_original": { "max_logit_delta": 0.0, "max_probability_delta": 0.0, "all_winners_match": true }, "prefix_unchanged": true, "branch_batch_sizes": [ 4 ], "warm_seconds": { "uncached": [ 2.039599666983122, 1.9174777079897467, 1.8907254159857985 ], "cached_batched": [ 0.5769222080125473, 0.5648717090080027, 0.5669595829967875 ] }, "warm_median_seconds": { "uncached": 1.9174777079897467, "cached_batched": 0.5669595829967875 }, "speedup": 3.3820359783928575 }, { "state": "speech", "prefix_tokens": 86, "suffix_lengths": [ 111, 99, 111, 111 ], "cached_serial_vs_uncached": { "max_logit_delta": 0.08770561218261719, "max_probability_delta": 0.013604034682185018, "all_winners_match": true }, "cached_batch_vs_uncached": { "max_logit_delta": 0.051482200622558594, "max_probability_delta": 0.008151936419399375, "all_winners_match": true }, "reversed_batch_vs_original": { "max_logit_delta": 0.0, "max_probability_delta": 0.0, "all_winners_match": true }, "prefix_unchanged": true, "branch_batch_sizes": [ 4 ], "warm_seconds": { "uncached": [ 0.33885545798693784, 0.34261837499798276, 0.33428337500663474 ], "cached_batched": [ 0.199704833998112, 0.19404779098113067, 0.18803491699509323 ] }, "warm_median_seconds": { "uncached": 0.33885545798693784, "cached_batched": 0.19404779098113067 }, "speedup": 1.7462474387038416 }, { "state": "video", "prefix_tokens": 328, "suffix_lengths": [ 111, 99, 111, 111 ], "cached_serial_vs_uncached": { "max_logit_delta": 0.007994651794433594, "max_probability_delta": 0.00047290616548957554, "all_winners_match": true }, "cached_batch_vs_uncached": { "max_logit_delta": 0.016923904418945312, "max_probability_delta": 0.0007754361118514908, "all_winners_match": true }, "reversed_batch_vs_original": { "max_logit_delta": 0.0, "max_probability_delta": 0.0, "all_winners_match": true }, "prefix_unchanged": true, "branch_batch_sizes": [ 4 ], "warm_seconds": { "uncached": [ 1.87490254198201, 1.9036407500097994, 2.002558832988143 ], "cached_batched": [ 0.6002716670045629, 0.5745167080021929, 0.6253891669912264 ] }, "warm_median_seconds": { "uncached": 1.9036407500097994, "cached_batched": 0.6002716670045629 }, "speedup": 3.1712986879910976 }, { "state": "video_speech", "prefix_tokens": 431, "suffix_lengths": [ 111, 99, 111, 111 ], "cached_serial_vs_uncached": { "max_logit_delta": 0.05509662628173828, "max_probability_delta": 0.005538062741424418, "all_winners_match": true }, "cached_batch_vs_uncached": { "max_logit_delta": 0.03372955322265625, "max_probability_delta": 0.002709010563853226, "all_winners_match": true }, "reversed_batch_vs_original": { "max_logit_delta": 0.0, "max_probability_delta": 0.0, "all_winners_match": true }, "prefix_unchanged": true, "branch_batch_sizes": [ 4 ], "warm_seconds": { "uncached": [ 2.368914416991174, 2.1928319169965107, 2.2464634170173667 ], "cached_batched": [ 0.7086897499975748, 0.6427362920076121, 0.8200017909985036 ] }, "warm_median_seconds": { "uncached": 2.2464634170173667, "cached_batched": 0.7086897499975748 }, "speedup": 3.169882754795105 }, { "state": "long_text", "prefix_tokens": 1250, "suffix_lengths": [ 111, 99, 111, 111 ], "cached_serial_vs_uncached": { "max_logit_delta": 0.028132915496826172, "max_probability_delta": 1.1824594636777874e-09, "all_winners_match": true }, "cached_batch_vs_uncached": { "max_logit_delta": 0.041655540466308594, "max_probability_delta": 2.3831407958141426e-09, "all_winners_match": true }, "reversed_batch_vs_original": { "max_logit_delta": 0.0, "max_probability_delta": 0.0, "all_winners_match": true }, "prefix_unchanged": true, "branch_batch_sizes": [ 4 ], "warm_seconds": { "uncached": [ 1.887113375007175, 1.7507582920079585, 1.7365583749779034 ], "cached_batched": [ 0.5613800830033142, 0.5454969170095865, 0.5390715000103228 ] }, "warm_median_seconds": { "uncached": 1.7507582920079585, "cached_batched": 0.5454969170095865 }, "speedup": 3.2094742195897523 } ] }