Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
 Community
ASzecsenyi commited on
Commit
9841f5c
·
verified ·
1 Parent(s): a2b696f

Upload stage_1_etrtjgi8/meta_000498.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. stage_1_etrtjgi8/meta_000498.json +166 -0
stage_1_etrtjgi8/meta_000498.json ADDED
@@ -0,0 +1,166 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 498,
3
+ "val_bpb": 1.773050032118871,
4
+ "model_config": {
5
+ "run": "rva_6x256/stage_1",
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
+ ],
21
+ "recurrent_vocab_sizes": [
22
+ 8192,
23
+ 8192,
24
+ 8192,
25
+ 8192,
26
+ 8192,
27
+ 8192
28
+ ],
29
+ "kla_blocks": [],
30
+ "mamba_blocks": [],
31
+ "gdn_blocks": [],
32
+ "gla_blocks": [],
33
+ "moe_blocks": [],
34
+ "d_state": 16,
35
+ "mamba_params": true,
36
+ "kla_kernel": true,
37
+ "mimo_rank": 1,
38
+ "skip_around_kla": true,
39
+ "decoder_mlp": false,
40
+ "use_reparametrisation_trick": true,
41
+ "num_experts": 4,
42
+ "moe_top_k": 2,
43
+ "moe_capacity_factor": 1.0,
44
+ "moe_aux_loss_weight": 0.01,
45
+ "moe_embedding_experts": false,
46
+ "num_iterations": -1,
47
+ "target_flops": -1.0,
48
+ "target_param_data_ratio": 20,
49
+ "data_dir": "base_data",
50
+ "device_batch_size": 8,
51
+ "total_batch_size": 524288,
52
+ "use_muon": true,
53
+ "embedding_lr": 0.3,
54
+ "unembedding_lr": 0.004,
55
+ "grad_clip": 1.0,
56
+ "weight_decay": 0.0,
57
+ "matrix_lr": 0.02,
58
+ "adam_beta1": 0.8,
59
+ "adam_beta2": 0.95,
60
+ "warmup_ratio": 0.05,
61
+ "warmdown_ratio": 0.4,
62
+ "final_lr_frac": 0.0,
63
+ "resume_from_step": -1,
64
+ "eval_every": 250,
65
+ "eval_tokens": 10485760,
66
+ "core_metric_every": -1,
67
+ "core_metric_max_per_task": 500,
68
+ "sample_every": 250,
69
+ "save_every": 1000,
70
+ "push_checkpoints_to_hub": true,
71
+ "use_profiler": false,
72
+ "profile_step": 2,
73
+ "profile_micro_step": 0,
74
+ "memory_history_max_entries": 10000,
75
+ "model_tag": "rva_6x256/stage_1",
76
+ "n_layer": 6,
77
+ "n_head": 16,
78
+ "n_kv_head": 16,
79
+ "n_embd": 256
80
+ },
81
+ "user_config": {
82
+ "run": "rva_6x256/stage_1",
83
+ "wandb_group": "rva_6x256",
84
+ "seed": 42,
85
+ "device_type": "cuda",
86
+ "config": "rva5m",
87
+ "depth": 6,
88
+ "aspect_ratio": 64,
89
+ "model_dim": 256,
90
+ "head_dim": 16,
91
+ "sequence_len": 1024,
92
+ "vocab_size": 265,
93
+ "clustering_init_num_samples": 0,
94
+ "rva_blocks": [
95
+ 0,
96
+ 1
97
+ ],
98
+ "recurrent_vocab_sizes": [
99
+ 8192,
100
+ 8192,
101
+ 8192,
102
+ 8192,
103
+ 8192,
104
+ 8192
105
+ ],
106
+ "kla_blocks": [],
107
+ "mamba_blocks": [],
108
+ "gdn_blocks": [],
109
+ "gla_blocks": [],
110
+ "moe_blocks": [],
111
+ "d_state": 16,
112
+ "mamba_params": true,
113
+ "kla_kernel": true,
114
+ "mimo_rank": 1,
115
+ "skip_around_kla": true,
116
+ "decoder_mlp": false,
117
+ "use_reparametrisation_trick": true,
118
+ "num_experts": 4,
119
+ "moe_top_k": 2,
120
+ "moe_capacity_factor": 1.0,
121
+ "moe_aux_loss_weight": 0.01,
122
+ "moe_embedding_experts": false,
123
+ "num_iterations": -1,
124
+ "target_flops": -1.0,
125
+ "target_param_data_ratio": 20,
126
+ "data_dir": "base_data",
127
+ "device_batch_size": 8,
128
+ "total_batch_size": 524288,
129
+ "use_muon": true,
130
+ "embedding_lr": 0.3,
131
+ "unembedding_lr": 0.004,
132
+ "grad_clip": 1.0,
133
+ "weight_decay": 0.0,
134
+ "matrix_lr": 0.02,
135
+ "adam_beta1": 0.8,
136
+ "adam_beta2": 0.95,
137
+ "warmup_ratio": 0.05,
138
+ "warmdown_ratio": 0.4,
139
+ "final_lr_frac": 0.0,
140
+ "resume_from_step": -1,
141
+ "eval_every": 250,
142
+ "eval_tokens": 10485760,
143
+ "core_metric_every": -1,
144
+ "core_metric_max_per_task": 500,
145
+ "sample_every": 250,
146
+ "save_every": 1000,
147
+ "push_checkpoints_to_hub": true,
148
+ "use_profiler": false,
149
+ "profile_step": 2,
150
+ "profile_micro_step": 0,
151
+ "memory_history_max_entries": 10000,
152
+ "model_tag": "rva_6x256/stage_1"
153
+ },
154
+ "device_batch_size": 8,
155
+ "sequence_len": 1024,
156
+ "dataloader_state_dict": {
157
+ "pq_idx": 1,
158
+ "rg_idx": 0
159
+ },
160
+ "loop_state": {
161
+ "min_val_bpb": 1.773050032118871,
162
+ "smooth_train_loss": 2.8556261922495008,
163
+ "smooth_aux_loss": 1.6369969724834987,
164
+ "total_training_time": 790.5082960128784
165
+ }
166
+ }