Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
e3592a4
·
verified ·
1 Parent(s): 563161c

Upload stage_4_zsekhtxh/meta_000195.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. stage_4_zsekhtxh/meta_000195.json +172 -0
stage_4_zsekhtxh/meta_000195.json ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 195,
3
+ "val_bpb": 4.00121383479538,
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
+ 265,
26
+ 256,
27
+ 256,
28
+ 256,
29
+ 256,
30
+ 256
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": 64,
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
+ "n_layer": 6,
80
+ "n_head": 16,
81
+ "n_kv_head": 16,
82
+ "n_embd": 256
83
+ },
84
+ "user_config": {
85
+ "run": "rva_6x256/stage_4",
86
+ "wandb_group": "rva_6x256",
87
+ "seed": 42,
88
+ "device_type": "cuda",
89
+ "config": "rva5m",
90
+ "depth": 6,
91
+ "aspect_ratio": 64,
92
+ "model_dim": 256,
93
+ "head_dim": 16,
94
+ "sequence_len": 1024,
95
+ "vocab_size": 265,
96
+ "clustering_init_num_samples": 0,
97
+ "rva_blocks": [
98
+ 0,
99
+ 1,
100
+ 2,
101
+ 3,
102
+ 4
103
+ ],
104
+ "recurrent_vocab_sizes": [
105
+ 265,
106
+ 256,
107
+ 256,
108
+ 256,
109
+ 256,
110
+ 256
111
+ ],
112
+ "kla_blocks": [],
113
+ "mamba_blocks": [],
114
+ "gdn_blocks": [],
115
+ "gla_blocks": [],
116
+ "moe_blocks": [],
117
+ "d_state": 16,
118
+ "mamba_params": true,
119
+ "kla_kernel": true,
120
+ "mimo_rank": 1,
121
+ "skip_around_kla": true,
122
+ "decoder_mlp": false,
123
+ "use_reparametrisation_trick": true,
124
+ "num_experts": 4,
125
+ "moe_top_k": 2,
126
+ "moe_capacity_factor": 1.0,
127
+ "moe_aux_loss_weight": 0.01,
128
+ "moe_embedding_experts": false,
129
+ "num_iterations": -1,
130
+ "target_flops": -1.0,
131
+ "target_param_data_ratio": 20,
132
+ "data_dir": "base_data",
133
+ "device_batch_size": 64,
134
+ "total_batch_size": 524288,
135
+ "use_muon": true,
136
+ "embedding_lr": 0.3,
137
+ "unembedding_lr": 0.004,
138
+ "grad_clip": 1.0,
139
+ "weight_decay": 0.0,
140
+ "matrix_lr": 0.02,
141
+ "adam_beta1": 0.8,
142
+ "adam_beta2": 0.95,
143
+ "warmup_ratio": 0.05,
144
+ "warmdown_ratio": 0.4,
145
+ "final_lr_frac": 0.0,
146
+ "resume_from_step": -1,
147
+ "eval_every": 250,
148
+ "eval_tokens": 10485760,
149
+ "core_metric_every": -1,
150
+ "core_metric_max_per_task": 500,
151
+ "sample_every": 250,
152
+ "save_every": 1000,
153
+ "push_checkpoints_to_hub": true,
154
+ "use_profiler": false,
155
+ "profile_step": 2,
156
+ "profile_micro_step": 0,
157
+ "memory_history_max_entries": 10000,
158
+ "model_tag": "rva_6x256/stage_4"
159
+ },
160
+ "device_batch_size": 64,
161
+ "sequence_len": 1024,
162
+ "dataloader_state_dict": {
163
+ "pq_idx": 0,
164
+ "rg_idx": 20
165
+ },
166
+ "loop_state": {
167
+ "min_val_bpb": 4.00121383479538,
168
+ "smooth_train_loss": 49.02453114192581,
169
+ "smooth_aux_loss": 46.172315643858994,
170
+ "total_training_time": 195.16926193237305
171
+ }
172
+ }