Laghu-Checkpoints / ctx32k /eval /eval_ctx32k.json
jiteshkprasad's picture
Add ctx32k/eval from udgamlabs/udgam3-ctx32k
8eb2fc5 verified
Raw History Blame Contribute Delete
3.96 kB
{
"ckpt": "/content/cache/base_checkpoints/udgam3-ctx32k/model_001900.pt",
"config": {
"sequence_len": 32768,
"vocab_size": 32768,
"n_layer": 28,
"n_head": 10,
"n_kv_head": 2,
"n_embd": 640,
"head_dim": 64,
"global_head_dim": 128,
"intermediate_size": 1792,
"ple_dim": 0,
"window_pattern": "SSSL",
"sliding_window": 512,
"rope_theta_sliding": 10000.0,
"rope_theta_full": 1000000.0,
"final_logit_softcap": 15.0,
"rms_norm_eps": 1e-06,
"loop_start": -1,
"loop_end": -1,
"n_loops": 1,
"mtp": false,
"loop_loss": false,
"loop_loss_stride": 4,
"loop_loss_weight": -1.0,
"loop_reinject": false,
"abacus": 0,
"abacus_offset": 0,
"digit_ids": [
291,
292,
293,
294,
295,
296,
297,
298,
299,
300
],
"digit_space_ids": [
1150,
958,
994,
1106,
1189,
1182,
1348,
1506,
1500,
1564
],
"tie_embeddings": true
},
"bpb2048_web": 0.8535457788260455,
"bpb2048_task": 0.6393049352887717,
"bpb2048_reason": 0.6919760354978337,
"bpb2048_fresh": 0.9205743961059951,
"long_bpb_L8192": {
"0-512": 1.026994,
"512-1024": 0.935882,
"1024-2048": 0.944786,
"2048-3072": 0.930071,
"3072-4096": 0.921109,
"4096-6144": 0.928579,
"6144-8192": 0.925182,
"n_docs": 400,
"per_src_first_half_second_half": {
"pre1929": [
1.06866,
1.07843,
60
],
"gutenberg": [
1.0833,
1.06462,
112
],
"arxiv": [
0.76825,
0.6845,
35
],
"loc": [
1.21512,
1.12338,
59
],
"doab": [
0.68966,
0.69926,
54
],
"hansard_commons": [
0.75608,
0.75785,
80
]
}
},
"long_bpb_L16384": {
"0-512": 1.026994,
"512-1024": 0.935882,
"1024-2048": 0.944786,
"2048-3072": 0.930071,
"3072-4096": 0.921109,
"4096-6144": 0.928579,
"6144-8192": 0.925182,
"8192-12288": 0.928408,
"12288-16384": 0.938268,
"n_docs": 400,
"per_src_first_half_second_half": {
"pre1929": [
1.07349,
1.08106,
60
],
"gutenberg": [
1.07386,
1.08415,
112
],
"arxiv": [
0.72967,
0.6699,
35
],
"loc": [
1.16862,
1.104,
59
],
"doab": [
0.69445,
0.68398,
54
],
"hansard_commons": [
0.75697,
0.77872,
80
]
}
},
"long_bpb_L32768": {
"0-512": 1.026994,
"512-1024": 0.935882,
"1024-2048": 0.944786,
"2048-3072": 0.930071,
"3072-4096": 0.921109,
"4096-6144": 0.928579,
"6144-8192": 0.925182,
"8192-12288": 0.928408,
"12288-16384": 0.938268,
"16384-24576": 0.943074,
"24576-32768": 0.940369,
"n_docs": 400,
"per_src_first_half_second_half": {
"pre1929": [
1.07725,
1.07459,
60
],
"gutenberg": [
1.07901,
1.09185,
112
],
"arxiv": [
0.70206,
0.68107,
35
],
"loc": [
1.13573,
1.09291,
59
],
"doab": [
0.68925,
0.69537,
54
],
"hansard_commons": [
0.76784,
0.80331,
80
]
}
},
"needle": {
"1024": {
"exact": 1.0,
"mean_logprob_key_token": -0.0238,
"distance_tokens_approx": 920
},
"2048": {
"exact": 1.0,
"mean_logprob_key_token": -0.0283,
"distance_tokens_approx": 1944
},
"4096": {
"exact": 1.0,
"mean_logprob_key_token": -0.0521,
"distance_tokens_approx": 3992
},
"8192": {
"exact": 1.0,
"mean_logprob_key_token": -0.0644,
"distance_tokens_approx": 8088
},
"12288": {
"exact": 0.99,
"mean_logprob_key_token": -0.1275,
"distance_tokens_approx": 12184
},
"16384": {
"exact": 0.96,
"mean_logprob_key_token": -0.1908,
"distance_tokens_approx": 16280
},
"24576": {
"exact": 0.0,
"mean_logprob_key_token": -2.4617,
"distance_tokens_approx": 24472
},
"32768": {
"exact": 0.0,
"mean_logprob_key_token": -2.9218,
"distance_tokens_approx": 32664
}
},
"copy": {
"2048": 0.947,
"8192": 0.8895,
"16384": 0.749,
"32768": 0.192
},
"seconds": 598.7
}