Arko007 commited on
Commit
16d95ac
·
verified ·
1 Parent(s): 88957ce

step 2800

Browse files
checkpoints/step_2800/metadata.json ADDED
@@ -0,0 +1,80 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 2800,
3
+ "tokens_consumed": 1468006400,
4
+ "sequences_consumed": 716800,
5
+ "config": {
6
+ "vocab_size": 129280,
7
+ "dim": 1536,
8
+ "n_layers": 16,
9
+ "n_heads": 12,
10
+ "head_dim": 128,
11
+ "rope_head_dim": 32,
12
+ "q_lora_rank": 384,
13
+ "o_groups": 4,
14
+ "o_lora_rank": 256,
15
+ "norm_eps": 1e-06,
16
+ "n_routed_experts": 12,
17
+ "n_shared_experts": 1,
18
+ "n_activated_experts": 2,
19
+ "moe_inter_dim": 1408,
20
+ "score_func": "sqrtsoftplus",
21
+ "route_scale": 2.5,
22
+ "swiglu_limit": 10.0,
23
+ "n_dense_layers": 2,
24
+ "load_balance_alpha": 0.0001,
25
+ "hc_mult": 3,
26
+ "hc_sinkhorn_iters": 8,
27
+ "hc_eps": 1e-06,
28
+ "window_size": 128,
29
+ "compress_ratio": 4,
30
+ "n_mtp_layers": 1,
31
+ "mtp_loss_weight": 0.3,
32
+ "rope_theta": 10000.0,
33
+ "rope_factor": 1.0,
34
+ "original_seq_len": 0,
35
+ "beta_fast": 32,
36
+ "beta_slow": 1,
37
+ "ctx_phase_boundaries": [
38
+ 60000,
39
+ 80000,
40
+ 100000
41
+ ],
42
+ "ctx_phase_lengths": [
43
+ 2048,
44
+ 4096,
45
+ 8192
46
+ ],
47
+ "yarn_scale": 40.0,
48
+ "yarn_alpha": 1.0,
49
+ "yarn_beta": 32.0,
50
+ "yarn_original_seq_len": 4096,
51
+ "micro_batch_size": 1,
52
+ "gradient_accumulation_steps": 32,
53
+ "max_lr": 0.00042,
54
+ "min_lr": 4.2e-05,
55
+ "warmup_steps": 2000,
56
+ "total_steps": 200000,
57
+ "decay_steps": 180000,
58
+ "weight_decay": 0.1,
59
+ "muon_beta": 0.95,
60
+ "muon_ns_steps": 5,
61
+ "adam_beta1": 0.9,
62
+ "adam_beta2": 0.95,
63
+ "grad_clip_norm": 1.0,
64
+ "init_std": 0.006,
65
+ "dtype": "bfloat16",
66
+ "fineweb_config": "default",
67
+ "approx_tokens_per_fineweb_doc": 800,
68
+ "approx_tokens_per_code_doc": 1200,
69
+ "approx_tokens_per_math_doc": 600,
70
+ "data_mix_probs": [
71
+ 0.6,
72
+ 0.25,
73
+ 0.15
74
+ ],
75
+ "save_every_steps": 400,
76
+ "log_every_steps": 10,
77
+ "push_every_steps": 400,
78
+ "checkpoint_dir": "/kaggle/working/zenyxv3checkpoints"
79
+ }
80
+ }
checkpoints/step_2800/opt_state.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7e69f2d9252378fea67694546cf4b05ce5f57ec838bdcc65fc0ad0069920f745
3
+ size 3154940698
checkpoints/step_2800/params.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b8ba4b6facd725943eb0eebca6ee1028a87f0e9f94e424c0250fec3c920beec
3
+ size 3154768094