Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
7d70ad3
·
verified ·
1 Parent(s): d6fdc01

Upload gpt_6x768_6uon5wbo/meta_001638.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. gpt_6x768_6uon5wbo/meta_001638.json +163 -0
gpt_6x768_6uon5wbo/meta_001638.json ADDED
@@ -0,0 +1,163 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 1638,
3
+ "val_bpb": 1.2348509239666234,
4
+ "model_config": {
5
+ "run": "gpt_6x768",
6
+ "wandb_group": null,
7
+ "seed": 42,
8
+ "device_type": "cuda",
9
+ "config": "gpt30m",
10
+ "depth": 6,
11
+ "aspect_ratio": 64,
12
+ "model_dim": 768,
13
+ "head_dim": 128,
14
+ "sequence_len": 1024,
15
+ "vocab_size": 265,
16
+ "clustering_init_num_samples": 0,
17
+ "rva_blocks": [],
18
+ "recurrent_vocab_sizes": [
19
+ 8192,
20
+ 8192,
21
+ 8192,
22
+ 8192,
23
+ 8192,
24
+ 8192
25
+ ],
26
+ "kla_blocks": [],
27
+ "mamba_blocks": [],
28
+ "gdn_blocks": [],
29
+ "gla_blocks": [],
30
+ "moe_blocks": [],
31
+ "d_state": 16,
32
+ "mamba_params": true,
33
+ "kla_kernel": true,
34
+ "mimo_rank": 1,
35
+ "skip_around_kla": true,
36
+ "decoder_mlp": false,
37
+ "use_reparametrisation_trick": true,
38
+ "num_experts": 4,
39
+ "moe_top_k": 2,
40
+ "moe_capacity_factor": 1.0,
41
+ "moe_aux_loss_weight": 0.01,
42
+ "moe_embedding_experts": false,
43
+ "num_iterations": -1,
44
+ "target_flops": -1.0,
45
+ "target_param_data_ratio": 20,
46
+ "data_dir": "base_data",
47
+ "device_batch_size": 64,
48
+ "total_batch_size": 524288,
49
+ "use_muon": true,
50
+ "embedding_lr": 0.3,
51
+ "unembedding_lr": 0.004,
52
+ "grad_clip": 1.0,
53
+ "weight_decay": 0.0,
54
+ "matrix_lr": 0.02,
55
+ "adam_beta1": 0.8,
56
+ "adam_beta2": 0.95,
57
+ "warmup_ratio": 0.05,
58
+ "warmdown_ratio": 0.4,
59
+ "final_lr_frac": 0.0,
60
+ "resume_from_step": -1,
61
+ "eval_every": 250,
62
+ "eval_tokens": 10485760,
63
+ "core_metric_every": -1,
64
+ "core_metric_max_per_task": 500,
65
+ "sample_every": 250,
66
+ "save_every": 1000,
67
+ "push_checkpoints_to_hub": true,
68
+ "use_profiler": false,
69
+ "profile_step": 2,
70
+ "profile_micro_step": 0,
71
+ "memory_history_max_entries": 10000,
72
+ "model_tag": "gpt_6x768",
73
+ "stage": null,
74
+ "n_layer": 6,
75
+ "n_head": 6,
76
+ "n_kv_head": 6,
77
+ "n_embd": 768
78
+ },
79
+ "user_config": {
80
+ "run": "gpt_6x768",
81
+ "wandb_group": null,
82
+ "seed": 42,
83
+ "device_type": "cuda",
84
+ "config": "gpt30m",
85
+ "depth": 6,
86
+ "aspect_ratio": 64,
87
+ "model_dim": 768,
88
+ "head_dim": 128,
89
+ "sequence_len": 1024,
90
+ "vocab_size": 265,
91
+ "clustering_init_num_samples": 0,
92
+ "rva_blocks": [],
93
+ "recurrent_vocab_sizes": [
94
+ 8192,
95
+ 8192,
96
+ 8192,
97
+ 8192,
98
+ 8192,
99
+ 8192
100
+ ],
101
+ "kla_blocks": [],
102
+ "mamba_blocks": [],
103
+ "gdn_blocks": [],
104
+ "gla_blocks": [],
105
+ "moe_blocks": [],
106
+ "d_state": 16,
107
+ "mamba_params": true,
108
+ "kla_kernel": true,
109
+ "mimo_rank": 1,
110
+ "skip_around_kla": true,
111
+ "decoder_mlp": false,
112
+ "use_reparametrisation_trick": true,
113
+ "num_experts": 4,
114
+ "moe_top_k": 2,
115
+ "moe_capacity_factor": 1.0,
116
+ "moe_aux_loss_weight": 0.01,
117
+ "moe_embedding_experts": false,
118
+ "num_iterations": -1,
119
+ "target_flops": -1.0,
120
+ "target_param_data_ratio": 20,
121
+ "data_dir": "base_data",
122
+ "device_batch_size": 64,
123
+ "total_batch_size": 524288,
124
+ "use_muon": true,
125
+ "embedding_lr": 0.3,
126
+ "unembedding_lr": 0.004,
127
+ "grad_clip": 1.0,
128
+ "weight_decay": 0.0,
129
+ "matrix_lr": 0.02,
130
+ "adam_beta1": 0.8,
131
+ "adam_beta2": 0.95,
132
+ "warmup_ratio": 0.05,
133
+ "warmdown_ratio": 0.4,
134
+ "final_lr_frac": 0.0,
135
+ "resume_from_step": -1,
136
+ "eval_every": 250,
137
+ "eval_tokens": 10485760,
138
+ "core_metric_every": -1,
139
+ "core_metric_max_per_task": 500,
140
+ "sample_every": 250,
141
+ "save_every": 1000,
142
+ "push_checkpoints_to_hub": true,
143
+ "use_profiler": false,
144
+ "profile_step": 2,
145
+ "profile_micro_step": 0,
146
+ "memory_history_max_entries": 10000,
147
+ "model_tag": "gpt_6x768",
148
+ "stage": null
149
+ },
150
+ "stage": null,
151
+ "device_batch_size": 64,
152
+ "sequence_len": 1024,
153
+ "dataloader_state_dict": {
154
+ "pq_idx": 3,
155
+ "rg_idx": 19
156
+ },
157
+ "loop_state": {
158
+ "min_val_bpb": 1.2348509239666234,
159
+ "smooth_train_loss": 0.8891634502511613,
160
+ "smooth_aux_loss": 0.0,
161
+ "total_training_time": 1545.3724329471588
162
+ }
163
+ }