Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
a366d04
·
verified ·
1 Parent(s): 0b6aefc

Upload stage_4_5v5k43wq/meta_001435.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. stage_4_5v5k43wq/meta_001435.json +175 -0
stage_4_5v5k43wq/meta_001435.json ADDED
@@ -0,0 +1,175 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 1435,
3
+ "val_bpb": 2.3888145393171993,
4
+ "model_config": {
5
+ "run": "rva_6x256/stage_4",
6
+ "wandb_group": "rva_6x256",
7
+ "seed": 42,
8
+ "device_type": "cuda",
9
+ "config": "rva5m",
10
+ "depth": 6,
11
+ "aspect_ratio": 64,
12
+ "model_dim": 256,
13
+ "head_dim": 16,
14
+ "sequence_len": 1024,
15
+ "vocab_size": 265,
16
+ "clustering_init_num_samples": 0,
17
+ "rva_blocks": [
18
+ 0,
19
+ 1,
20
+ 2,
21
+ 3,
22
+ 4
23
+ ],
24
+ "recurrent_vocab_sizes": [
25
+ 8192,
26
+ 8192,
27
+ 8192,
28
+ 8192,
29
+ 8192,
30
+ 8192
31
+ ],
32
+ "kla_blocks": [],
33
+ "mamba_blocks": [],
34
+ "gdn_blocks": [],
35
+ "gla_blocks": [],
36
+ "moe_blocks": [],
37
+ "d_state": 16,
38
+ "mamba_params": true,
39
+ "kla_kernel": true,
40
+ "mimo_rank": 1,
41
+ "skip_around_kla": true,
42
+ "decoder_mlp": false,
43
+ "use_reparametrisation_trick": true,
44
+ "num_experts": 4,
45
+ "moe_top_k": 2,
46
+ "moe_capacity_factor": 1.0,
47
+ "moe_aux_loss_weight": 0.01,
48
+ "moe_embedding_experts": false,
49
+ "num_iterations": -1,
50
+ "target_flops": -1.0,
51
+ "target_param_data_ratio": 20,
52
+ "data_dir": "base_data",
53
+ "device_batch_size": 8,
54
+ "total_batch_size": 524288,
55
+ "use_muon": true,
56
+ "embedding_lr": 0.3,
57
+ "unembedding_lr": 0.004,
58
+ "grad_clip": 1.0,
59
+ "weight_decay": 0.0,
60
+ "matrix_lr": 0.02,
61
+ "adam_beta1": 0.8,
62
+ "adam_beta2": 0.95,
63
+ "warmup_ratio": 0.05,
64
+ "warmdown_ratio": 0.4,
65
+ "final_lr_frac": 0.0,
66
+ "resume_from_step": -1,
67
+ "eval_every": 250,
68
+ "eval_tokens": 10485760,
69
+ "core_metric_every": -1,
70
+ "core_metric_max_per_task": 500,
71
+ "sample_every": 250,
72
+ "save_every": 1000,
73
+ "push_checkpoints_to_hub": true,
74
+ "use_profiler": false,
75
+ "profile_step": 2,
76
+ "profile_micro_step": 0,
77
+ "memory_history_max_entries": 10000,
78
+ "model_tag": "rva_6x256/stage_4",
79
+ "stage": 4,
80
+ "n_layer": 6,
81
+ "n_head": 16,
82
+ "n_kv_head": 16,
83
+ "n_embd": 256
84
+ },
85
+ "user_config": {
86
+ "run": "rva_6x256/stage_4",
87
+ "wandb_group": "rva_6x256",
88
+ "seed": 42,
89
+ "device_type": "cuda",
90
+ "config": "rva5m",
91
+ "depth": 6,
92
+ "aspect_ratio": 64,
93
+ "model_dim": 256,
94
+ "head_dim": 16,
95
+ "sequence_len": 1024,
96
+ "vocab_size": 265,
97
+ "clustering_init_num_samples": 0,
98
+ "rva_blocks": [
99
+ 0,
100
+ 1,
101
+ 2,
102
+ 3,
103
+ 4
104
+ ],
105
+ "recurrent_vocab_sizes": [
106
+ 8192,
107
+ 8192,
108
+ 8192,
109
+ 8192,
110
+ 8192,
111
+ 8192
112
+ ],
113
+ "kla_blocks": [],
114
+ "mamba_blocks": [],
115
+ "gdn_blocks": [],
116
+ "gla_blocks": [],
117
+ "moe_blocks": [],
118
+ "d_state": 16,
119
+ "mamba_params": true,
120
+ "kla_kernel": true,
121
+ "mimo_rank": 1,
122
+ "skip_around_kla": true,
123
+ "decoder_mlp": false,
124
+ "use_reparametrisation_trick": true,
125
+ "num_experts": 4,
126
+ "moe_top_k": 2,
127
+ "moe_capacity_factor": 1.0,
128
+ "moe_aux_loss_weight": 0.01,
129
+ "moe_embedding_experts": false,
130
+ "num_iterations": -1,
131
+ "target_flops": -1.0,
132
+ "target_param_data_ratio": 20,
133
+ "data_dir": "base_data",
134
+ "device_batch_size": 8,
135
+ "total_batch_size": 524288,
136
+ "use_muon": true,
137
+ "embedding_lr": 0.3,
138
+ "unembedding_lr": 0.004,
139
+ "grad_clip": 1.0,
140
+ "weight_decay": 0.0,
141
+ "matrix_lr": 0.02,
142
+ "adam_beta1": 0.8,
143
+ "adam_beta2": 0.95,
144
+ "warmup_ratio": 0.05,
145
+ "warmdown_ratio": 0.4,
146
+ "final_lr_frac": 0.0,
147
+ "resume_from_step": -1,
148
+ "eval_every": 250,
149
+ "eval_tokens": 10485760,
150
+ "core_metric_every": -1,
151
+ "core_metric_max_per_task": 500,
152
+ "sample_every": 250,
153
+ "save_every": 1000,
154
+ "push_checkpoints_to_hub": true,
155
+ "use_profiler": false,
156
+ "profile_step": 2,
157
+ "profile_micro_step": 0,
158
+ "memory_history_max_entries": 10000,
159
+ "model_tag": "rva_6x256/stage_4",
160
+ "stage": 4
161
+ },
162
+ "stage": 4,
163
+ "device_batch_size": 8,
164
+ "sequence_len": 1024,
165
+ "dataloader_state_dict": {
166
+ "pq_idx": 3,
167
+ "rg_idx": 0
168
+ },
169
+ "loop_state": {
170
+ "min_val_bpb": 2.3888145393171993,
171
+ "smooth_train_loss": 433.0383763626778,
172
+ "smooth_aux_loss": 431.352965352343,
173
+ "total_training_time": 3895.701017856598
174
+ }
175
+ }