Laghu-Checkpoints / ctx32k /eval /eval_base2k.json
jiteshkprasad's picture
Add ctx32k/eval from udgamlabs/udgam3-ctx32k
8eb2fc5 verified
Raw History Blame Contribute Delete
3.94 kB
{
"ckpt": "/content/init/udgam3-base-v1/model_ema_076294.pt",
"config": {
"sequence_len": 8193,
"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": 0.8501416386046416,
"bpb2048_task": 0.634110372765497,
"bpb2048_reason": 0.6844764738368659,
"bpb2048_fresh": 0.9209135476121134,
"long_bpb_L8192": {
"0-512": 1.173462,
"512-1024": 1.035295,
"1024-2048": 1.012183,
"2048-3072": 1.082558,
"3072-4096": 1.27177,
"4096-6144": 1.366089,
"6144-8192": 1.347903,
"n_docs": 400,
"per_src_first_half_second_half": {
"pre1929": [
1.20924,
1.47077,
60
],
"gutenberg": [
1.25172,
1.53167,
112
],
"arxiv": [
0.91582,
1.12526,
35
],
"loc": [
1.41106,
1.63034,
59
],
"doab": [
0.88866,
1.10496,
54
],
"hansard_commons": [
0.93938,
1.13722,
80
]
}
},
"long_bpb_L16384": {
"0-512": 1.173462,
"512-1024": 1.035295,
"1024-2048": 1.012183,
"2048-3072": 1.082558,
"3072-4096": 1.27177,
"4096-6144": 1.366089,
"6144-8192": 1.347903,
"8192-12288": 1.368215,
"12288-16384": 1.388605,
"n_docs": 400,
"per_src_first_half_second_half": {
"pre1929": [
1.33858,
1.49199,
60
],
"gutenberg": [
1.3932,
1.5649,
112
],
"arxiv": [
1.0123,
1.13282,
35
],
"loc": [
1.52222,
1.59875,
59
],
"doab": [
0.99651,
1.11498,
54
],
"hansard_commons": [
1.0388,
1.17896,
80
]
}
},
"long_bpb_L32768": {
"0-512": 1.173462,
"512-1024": 1.035295,
"1024-2048": 1.012183,
"2048-3072": 1.082558,
"3072-4096": 1.27177,
"4096-6144": 1.366089,
"6144-8192": 1.347903,
"8192-12288": 1.368215,
"12288-16384": 1.388605,
"16384-24576": 1.403997,
"24576-32768": 1.436084,
"n_docs": 400,
"per_src_first_half_second_half": {
"pre1929": [
1.41471,
1.50851,
60
],
"gutenberg": [
1.47915,
1.60659,
112
],
"arxiv": [
1.06798,
1.26401,
35
],
"loc": [
1.56117,
1.61969,
59
],
"doab": [
1.05531,
1.19035,
54
],
"hansard_commons": [
1.10888,
1.20284,
80
]
}
},
"needle": {
"1024": {
"exact": 1.0,
"mean_logprob_key_token": -0.0224,
"distance_tokens_approx": 920
},
"2048": {
"exact": 1.0,
"mean_logprob_key_token": -0.0422,
"distance_tokens_approx": 1944
},
"4096": {
"exact": 0.0,
"mean_logprob_key_token": -2.6239,
"distance_tokens_approx": 3992
},
"8192": {
"exact": 0.0,
"mean_logprob_key_token": -3.6672,
"distance_tokens_approx": 8088
},
"12288": {
"exact": 0.0,
"mean_logprob_key_token": -3.7545,
"distance_tokens_approx": 12184
},
"16384": {
"exact": 0.0,
"mean_logprob_key_token": -3.9069,
"distance_tokens_approx": 16280
},
"24576": {
"exact": 0.0,
"mean_logprob_key_token": -4.1492,
"distance_tokens_approx": 24472
},
"32768": {
"exact": 0.0,
"mean_logprob_key_token": -4.1836,
"distance_tokens_approx": 32664
}
},
"copy": {
"2048": 0.9475,
"8192": 0.043,
"16384": 0.0225,
"32768": 0.023
},
"seconds": 597.9
}