sansar-60m / eval /verification.json
ks2812's picture
v0.1.0: F0 (ex-Gita 0.6434)
bcd28bc verified
Raw History Blame Contribute Delete
3.58 kB
{
"size": "60m",
"version": "v0.1.0",
"run": "f0_slp1_uni8k_d60m",
"torch": "2.14.0+cu130",
"transformers": "5.18.0",
"hf_load_seconds": 0.2,
"hf_param_counts": {
"total": 63173376,
"embedding": 6537216,
"head": 0,
"non_embedding": 56636160
},
"orig_param_counts": {
"total": 63173376,
"embedding": 6537216,
"token_embedding": 6144000,
"position_embedding": 393216,
"value_embedding": 0,
"head": 0,
"non_embedding": 56636160
},
"tied_head": true,
"hf_head_is_tied": true,
"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": 0.08428049087524414,
"mean_abs_diff_vs_fp32_master": 0.010810306295752525,
"argmax_agree_vs_fp32_master": 1.0,
"logit_abs_max": 15.046829223632812
},
{
"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": 0.08698654174804688,
"mean_abs_diff_vs_fp32_master": 0.010311635211110115,
"argmax_agree_vs_fp32_master": 1.0,
"logit_abs_max": 15.367975234985352
},
{
"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": 0.11462783813476562,
"mean_abs_diff_vs_fp32_master": 0.009786239825189114,
"argmax_agree_vs_fp32_master": 1.0,
"logit_abs_max": 19.014877319335938
},
{
"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": 0.0740823745727539,
"mean_abs_diff_vs_fp32_master": 0.009775633923709393,
"argmax_agree_vs_fp32_master": 1.0,
"logit_abs_max": 17.23141098022461
},
{
"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": 0.1070852279663086,
"mean_abs_diff_vs_fp32_master": 0.01093593705445528,
"argmax_agree_vs_fp32_master": 1.0,
"logit_abs_max": 16.507715225219727
},
{
"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": 0.15094518661499023,
"mean_abs_diff_vs_fp32_master": 0.009178726002573967,
"argmax_agree_vs_fp32_master": 0.99609375,
"logit_abs_max": 21.30445098876953
}
],
"logits_max_abs_diff_impl": 0.0,
"logits_max_abs_diff_bf16_storage": 0.15094518661499023,
"left_padding_max_abs_diff": 3.910064697265625e-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": 0.57871,
"dcs_gold": 0.67942,
"gita": 0.26099,
"ood": 0.58167,
"prose": 0.64115,
"vedic": 1.11792
},
"hf_bf16_weights_fp32_compute": {
"pooled": 0.57873,
"dcs_gold": 0.67953,
"gita": 0.26107,
"ood": 0.58167,
"prose": 0.6412,
"vedic": 1.11819
},
"pooled_rel_diff": 3.266520318733796e-05,
"seconds": 11.0
}
}