Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
a680ff6
·
verified ·
1 Parent(s): 5509b7b

Upload stage_3_kktznl70/meta_000193.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. stage_3_kktznl70/meta_000193.json +170 -0
stage_3_kktznl70/meta_000193.json ADDED
@@ -0,0 +1,170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 193,
3
+ "val_bpb": 3.584739008236567,
4
+ "model_config": {
5
+ "run": "rva_6x256/stage_3",
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
+ ],
23
+ "recurrent_vocab_sizes": [
24
+ 265,
25
+ 256,
26
+ 256,
27
+ 256,
28
+ 256,
29
+ 256
30
+ ],
31
+ "kla_blocks": [],
32
+ "mamba_blocks": [],
33
+ "gdn_blocks": [],
34
+ "gla_blocks": [],
35
+ "moe_blocks": [],
36
+ "d_state": 16,
37
+ "mamba_params": true,
38
+ "kla_kernel": true,
39
+ "mimo_rank": 1,
40
+ "skip_around_kla": true,
41
+ "decoder_mlp": false,
42
+ "use_reparametrisation_trick": true,
43
+ "num_experts": 4,
44
+ "moe_top_k": 2,
45
+ "moe_capacity_factor": 1.0,
46
+ "moe_aux_loss_weight": 0.01,
47
+ "moe_embedding_experts": false,
48
+ "num_iterations": -1,
49
+ "target_flops": -1.0,
50
+ "target_param_data_ratio": 20,
51
+ "data_dir": "base_data",
52
+ "device_batch_size": 64,
53
+ "total_batch_size": 524288,
54
+ "use_muon": true,
55
+ "embedding_lr": 0.3,
56
+ "unembedding_lr": 0.004,
57
+ "grad_clip": 1.0,
58
+ "weight_decay": 0.0,
59
+ "matrix_lr": 0.02,
60
+ "adam_beta1": 0.8,
61
+ "adam_beta2": 0.95,
62
+ "warmup_ratio": 0.05,
63
+ "warmdown_ratio": 0.4,
64
+ "final_lr_frac": 0.0,
65
+ "resume_from_step": -1,
66
+ "eval_every": 250,
67
+ "eval_tokens": 10485760,
68
+ "core_metric_every": -1,
69
+ "core_metric_max_per_task": 500,
70
+ "sample_every": 250,
71
+ "save_every": 1000,
72
+ "push_checkpoints_to_hub": true,
73
+ "use_profiler": false,
74
+ "profile_step": 2,
75
+ "profile_micro_step": 0,
76
+ "memory_history_max_entries": 10000,
77
+ "model_tag": "rva_6x256/stage_3",
78
+ "n_layer": 6,
79
+ "n_head": 16,
80
+ "n_kv_head": 16,
81
+ "n_embd": 256
82
+ },
83
+ "user_config": {
84
+ "run": "rva_6x256/stage_3",
85
+ "wandb_group": "rva_6x256",
86
+ "seed": 42,
87
+ "device_type": "cuda",
88
+ "config": "rva5m",
89
+ "depth": 6,
90
+ "aspect_ratio": 64,
91
+ "model_dim": 256,
92
+ "head_dim": 16,
93
+ "sequence_len": 1024,
94
+ "vocab_size": 265,
95
+ "clustering_init_num_samples": 0,
96
+ "rva_blocks": [
97
+ 0,
98
+ 1,
99
+ 2,
100
+ 3
101
+ ],
102
+ "recurrent_vocab_sizes": [
103
+ 265,
104
+ 256,
105
+ 256,
106
+ 256,
107
+ 256,
108
+ 256
109
+ ],
110
+ "kla_blocks": [],
111
+ "mamba_blocks": [],
112
+ "gdn_blocks": [],
113
+ "gla_blocks": [],
114
+ "moe_blocks": [],
115
+ "d_state": 16,
116
+ "mamba_params": true,
117
+ "kla_kernel": true,
118
+ "mimo_rank": 1,
119
+ "skip_around_kla": true,
120
+ "decoder_mlp": false,
121
+ "use_reparametrisation_trick": true,
122
+ "num_experts": 4,
123
+ "moe_top_k": 2,
124
+ "moe_capacity_factor": 1.0,
125
+ "moe_aux_loss_weight": 0.01,
126
+ "moe_embedding_experts": false,
127
+ "num_iterations": -1,
128
+ "target_flops": -1.0,
129
+ "target_param_data_ratio": 20,
130
+ "data_dir": "base_data",
131
+ "device_batch_size": 64,
132
+ "total_batch_size": 524288,
133
+ "use_muon": true,
134
+ "embedding_lr": 0.3,
135
+ "unembedding_lr": 0.004,
136
+ "grad_clip": 1.0,
137
+ "weight_decay": 0.0,
138
+ "matrix_lr": 0.02,
139
+ "adam_beta1": 0.8,
140
+ "adam_beta2": 0.95,
141
+ "warmup_ratio": 0.05,
142
+ "warmdown_ratio": 0.4,
143
+ "final_lr_frac": 0.0,
144
+ "resume_from_step": -1,
145
+ "eval_every": 250,
146
+ "eval_tokens": 10485760,
147
+ "core_metric_every": -1,
148
+ "core_metric_max_per_task": 500,
149
+ "sample_every": 250,
150
+ "save_every": 1000,
151
+ "push_checkpoints_to_hub": true,
152
+ "use_profiler": false,
153
+ "profile_step": 2,
154
+ "profile_micro_step": 0,
155
+ "memory_history_max_entries": 10000,
156
+ "model_tag": "rva_6x256/stage_3"
157
+ },
158
+ "device_batch_size": 64,
159
+ "sequence_len": 1024,
160
+ "dataloader_state_dict": {
161
+ "pq_idx": 0,
162
+ "rg_idx": 20
163
+ },
164
+ "loop_state": {
165
+ "min_val_bpb": 3.584739008236567,
166
+ "smooth_train_loss": 23.1435707259313,
167
+ "smooth_aux_loss": 20.590842054407627,
168
+ "total_training_time": 50.40851187705994
169
+ }
170
+ }