{ "size": "20m", "version": "v0.1.0", "run": "tokv3_v01_d20m", "torch": "2.14.0+cu130", "transformers": "5.18.0", "hf_load_seconds": 0.2, "hf_param_counts": { "total": 27073024, "embedding": 4096000, "head": 4096000, "non_embedding": 18881024 }, "orig_param_counts": { "total": 27073024, "embedding": 4096000, "token_embedding": 4096000, "position_embedding": 0, "value_embedding": 0, "head": 4096000, "non_embedding": 18881024 }, "tied_head": false, "hf_head_equals_wte": false, "logits": [ { "input": "dcs_gold", "tokens": 19, "max_abs_diff_vs_same_bf16_weights": 0.0, "mean_abs_diff_vs_same_bf16_weights": 0.0, "max_abs_diff_vs_fp32_master": 0.050026655197143555, "mean_abs_diff_vs_fp32_master": 0.007199867628514767, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 14.390743255615234 }, { "input": "gita", "tokens": 27, "max_abs_diff_vs_same_bf16_weights": 0.0, "mean_abs_diff_vs_same_bf16_weights": 0.0, "max_abs_diff_vs_fp32_master": 0.05633831024169922, "mean_abs_diff_vs_fp32_master": 0.007845329120755196, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 14.983847618103027 }, { "input": "ood", "tokens": 512, "max_abs_diff_vs_same_bf16_weights": 0.0, "mean_abs_diff_vs_same_bf16_weights": 0.0, "max_abs_diff_vs_fp32_master": 0.09844303131103516, "mean_abs_diff_vs_fp32_master": 0.009610326960682869, "argmax_agree_vs_fp32_master": 0.998046875, "logit_abs_max": 19.00731658935547 }, { "input": "prose", "tokens": 61, "max_abs_diff_vs_same_bf16_weights": 0.0, "mean_abs_diff_vs_same_bf16_weights": 0.0, "max_abs_diff_vs_fp32_master": 0.05675983428955078, "mean_abs_diff_vs_fp32_master": 0.007296254858374596, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 17.56630516052246 }, { "input": "vedic", "tokens": 22, "max_abs_diff_vs_same_bf16_weights": 0.0, "mean_abs_diff_vs_same_bf16_weights": 0.0, "max_abs_diff_vs_fp32_master": 0.05190229415893555, "mean_abs_diff_vs_fp32_master": 0.008906053379178047, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 13.91469955444336 }, { "input": "random_512", "tokens": 512, "max_abs_diff_vs_same_bf16_weights": 0.0, "mean_abs_diff_vs_same_bf16_weights": 0.0, "max_abs_diff_vs_fp32_master": 0.15868473052978516, "mean_abs_diff_vs_fp32_master": 0.010886614210903645, "argmax_agree_vs_fp32_master": 0.990234375, "logit_abs_max": 16.144594192504883 } ], "logits_max_abs_diff_impl": 0.0, "logits_max_abs_diff_bf16_storage": 0.15868473052978516, "left_padding_max_abs_diff": 1.430511474609375e-05, "tokenizer_ids_identical": "95/100", "tokenizer_first_difference": { "set": "ood", "index": 2 }, "tokenizer_ids_identical_fence_latin_false": "100/100", "tokenizer_roundtrip_exact": "100/100", "bpb_sample": { "records_per_set": 20, "block": 512, "stride": 256, "original_fp32_master": { "pooled": 0.68017, "dcs_gold": 0.74491, "gita": 0.47254, "ood": 0.68294, "prose": 0.72442, "vedic": 0.9819 }, "hf_bf16_weights_fp32_compute": { "pooled": 0.68017, "dcs_gold": 0.74491, "gita": 0.47263, "ood": 0.68295, "prose": 0.72441, "vedic": 0.98186 }, "pooled_rel_diff": 6.287091102380568e-06, "seconds": 6.0 } }