{ "size": "350m", "version": "v0.2.0", "run": "f9_slp1_uni8k_d350m_plus_all_3x", "torch": "2.14.0+cu130", "transformers": "5.18.0", "hf_load_seconds": 0.3, "hf_param_counts": { "total": 318424064, "embedding": 8192000, "head": 8192000, "non_embedding": 302040064 }, "orig_param_counts": { "total": 318424064, "embedding": 8192000, "token_embedding": 8192000, "position_embedding": 0, "value_embedding": 0, "head": 8192000, "non_embedding": 302040064 }, "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": 0.0973672866821289, "mean_abs_diff_vs_fp32_master": 0.013156784698367119, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 13.353614807128906 }, { "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": 0.1327071189880371, "mean_abs_diff_vs_fp32_master": 0.014552494511008263, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 20.497982025146484 }, { "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": 0.21717453002929688, "mean_abs_diff_vs_fp32_master": 0.02046874165534973, "argmax_agree_vs_fp32_master": 0.9921875, "logit_abs_max": 25.13204574584961 }, { "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": 0.09520530700683594, "mean_abs_diff_vs_fp32_master": 0.011608722619712353, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 16.601123809814453 }, { "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": 0.11458206176757812, "mean_abs_diff_vs_fp32_master": 0.014504635706543922, "argmax_agree_vs_fp32_master": 1.0, "logit_abs_max": 14.617786407470703 }, { "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": 0.17591071128845215, "mean_abs_diff_vs_fp32_master": 0.011679315008223057, "argmax_agree_vs_fp32_master": 0.99609375, "logit_abs_max": 17.314958572387695 } ], "logits_max_abs_diff_impl": 0.0, "logits_max_abs_diff_bf16_storage": 0.21717453002929688, "left_padding_max_abs_diff": 5.030632019042969e-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": 0.50865, "dcs_gold": 0.64282, "gita": 0.17083, "ood": 0.51883, "prose": 0.52615, "vedic": 0.72155 }, "hf_bf16_weights_fp32_compute": { "pooled": 0.5087, "dcs_gold": 0.64263, "gita": 0.17111, "ood": 0.51887, "prose": 0.52632, "vedic": 0.72173 }, "pooled_rel_diff": 9.97135358179582e-05, "seconds": 57.7 } }