{ "size": "125m", "version": "v0.1.0", "run": "f5_slp1_uni8k_d125m_plus_clean_2x", "torch": "2.14.0+cu130", "transformers": "5.18.0", "hf_load_seconds": 0.2, "hf_param_counts": { "total": 97241856, "embedding": 6144000, "head": 6144000, "non_embedding": 84953856 }, "orig_param_counts": { "total": 97241856, "embedding": 6144000, "token_embedding": 6144000, "position_embedding": 0, "value_embedding": 0, "head": 6144000, "non_embedding": 84953856 }, "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": 0.08556461334228516, "mean_abs_diff_vs_fp32_master": 0.012285877019166946, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 14.24844741821289 }, { "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": 0.08663368225097656, "mean_abs_diff_vs_fp32_master": 0.013421257957816124, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 16.642593383789062 }, { "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": 0.17093610763549805, "mean_abs_diff_vs_fp32_master": 0.014808709733188152, "argmax_agree_vs_fp32_master": 0.99609375, "logit_abs_max": 20.75664520263672 }, { "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": 0.098907470703125, "mean_abs_diff_vs_fp32_master": 0.012143973261117935, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 18.785600662231445 }, { "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": 0.09095573425292969, "mean_abs_diff_vs_fp32_master": 0.013502920046448708, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 15.030223846435547 }, { "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": 0.18518781661987305, "mean_abs_diff_vs_fp32_master": 0.01284902635961771, "argmax_agree_vs_fp32_master": 0.99609375, "logit_abs_max": 23.490028381347656 } ], "logits_max_abs_diff_impl": 0.0, "logits_max_abs_diff_bf16_storage": 0.18518781661987305, "left_padding_max_abs_diff": 2.300739288330078e-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": 0.55835, "dcs_gold": 0.6511, "gita": 0.30412, "ood": 0.56483, "prose": 0.57064, "vedic": 0.79789 }, "hf_bf16_weights_fp32_compute": { "pooled": 0.55835, "dcs_gold": 0.65141, "gita": 0.30412, "ood": 0.56481, "prose": 0.57085, "vedic": 0.79782 }, "pooled_rel_diff": -1.4596361401117072e-05, "seconds": 25.4 } }