Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
b992cdf
·
verified ·
1 Parent(s): b30cd18

Upload stage_5_zta6wl8a/meta_000198.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. stage_5_zta6wl8a/meta_000198.json +174 -0
stage_5_zta6wl8a/meta_000198.json ADDED
@@ -0,0 +1,174 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 198,
3
+ "val_bpb": 4.274172415513479,
4
+ "model_config": {
5
+ "run": "rva_6x256/stage_5",
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
+ 5
24
+ ],
25
+ "recurrent_vocab_sizes": [
26
+ 265,
27
+ 256,
28
+ 256,
29
+ 256,
30
+ 256,
31
+ 256
32
+ ],
33
+ "kla_blocks": [],
34
+ "mamba_blocks": [],
35
+ "gdn_blocks": [],
36
+ "gla_blocks": [],
37
+ "moe_blocks": [],
38
+ "d_state": 16,
39
+ "mamba_params": true,
40
+ "kla_kernel": true,
41
+ "mimo_rank": 1,
42
+ "skip_around_kla": true,
43
+ "decoder_mlp": false,
44
+ "use_reparametrisation_trick": true,
45
+ "num_experts": 4,
46
+ "moe_top_k": 2,
47
+ "moe_capacity_factor": 1.0,
48
+ "moe_aux_loss_weight": 0.01,
49
+ "moe_embedding_experts": false,
50
+ "num_iterations": -1,
51
+ "target_flops": -1.0,
52
+ "target_param_data_ratio": 20,
53
+ "data_dir": "base_data",
54
+ "device_batch_size": 64,
55
+ "total_batch_size": 524288,
56
+ "use_muon": true,
57
+ "embedding_lr": 0.3,
58
+ "unembedding_lr": 0.004,
59
+ "grad_clip": 1.0,
60
+ "weight_decay": 0.0,
61
+ "matrix_lr": 0.02,
62
+ "adam_beta1": 0.8,
63
+ "adam_beta2": 0.95,
64
+ "warmup_ratio": 0.05,
65
+ "warmdown_ratio": 0.4,
66
+ "final_lr_frac": 0.0,
67
+ "resume_from_step": -1,
68
+ "eval_every": 250,
69
+ "eval_tokens": 10485760,
70
+ "core_metric_every": -1,
71
+ "core_metric_max_per_task": 500,
72
+ "sample_every": 250,
73
+ "save_every": 1000,
74
+ "push_checkpoints_to_hub": true,
75
+ "use_profiler": false,
76
+ "profile_step": 2,
77
+ "profile_micro_step": 0,
78
+ "memory_history_max_entries": 10000,
79
+ "model_tag": "rva_6x256/stage_5",
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_5",
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
+ 5
105
+ ],
106
+ "recurrent_vocab_sizes": [
107
+ 265,
108
+ 256,
109
+ 256,
110
+ 256,
111
+ 256,
112
+ 256
113
+ ],
114
+ "kla_blocks": [],
115
+ "mamba_blocks": [],
116
+ "gdn_blocks": [],
117
+ "gla_blocks": [],
118
+ "moe_blocks": [],
119
+ "d_state": 16,
120
+ "mamba_params": true,
121
+ "kla_kernel": true,
122
+ "mimo_rank": 1,
123
+ "skip_around_kla": true,
124
+ "decoder_mlp": false,
125
+ "use_reparametrisation_trick": true,
126
+ "num_experts": 4,
127
+ "moe_top_k": 2,
128
+ "moe_capacity_factor": 1.0,
129
+ "moe_aux_loss_weight": 0.01,
130
+ "moe_embedding_experts": false,
131
+ "num_iterations": -1,
132
+ "target_flops": -1.0,
133
+ "target_param_data_ratio": 20,
134
+ "data_dir": "base_data",
135
+ "device_batch_size": 64,
136
+ "total_batch_size": 524288,
137
+ "use_muon": true,
138
+ "embedding_lr": 0.3,
139
+ "unembedding_lr": 0.004,
140
+ "grad_clip": 1.0,
141
+ "weight_decay": 0.0,
142
+ "matrix_lr": 0.02,
143
+ "adam_beta1": 0.8,
144
+ "adam_beta2": 0.95,
145
+ "warmup_ratio": 0.05,
146
+ "warmdown_ratio": 0.4,
147
+ "final_lr_frac": 0.0,
148
+ "resume_from_step": -1,
149
+ "eval_every": 250,
150
+ "eval_tokens": 10485760,
151
+ "core_metric_every": -1,
152
+ "core_metric_max_per_task": 500,
153
+ "sample_every": 250,
154
+ "save_every": 1000,
155
+ "push_checkpoints_to_hub": true,
156
+ "use_profiler": false,
157
+ "profile_step": 2,
158
+ "profile_micro_step": 0,
159
+ "memory_history_max_entries": 10000,
160
+ "model_tag": "rva_6x256/stage_5"
161
+ },
162
+ "device_batch_size": 64,
163
+ "sequence_len": 1024,
164
+ "dataloader_state_dict": {
165
+ "pq_idx": 0,
166
+ "rg_idx": 22
167
+ },
168
+ "loop_state": {
169
+ "min_val_bpb": 4.274172415513479,
170
+ "smooth_train_loss": 140.5364370329333,
171
+ "smooth_aux_loss": 137.49944244324095,
172
+ "total_training_time": 209.0471909046173
173
+ }
174
+ }