Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
 Community
ASzecsenyi commited on
Commit
6bb7142
·
verified ·
1 Parent(s): 7a76795

Upload stage_2_601u8rtg/meta_000190.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. stage_2_601u8rtg/meta_000190.json +168 -0
stage_2_601u8rtg/meta_000190.json ADDED
@@ -0,0 +1,168 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 190,
3
+ "val_bpb": 2.9875989317518807,
4
+ "model_config": {
5
+ "run": "rva_6x256/stage_2",
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
+ ],
22
+ "recurrent_vocab_sizes": [
23
+ 265,
24
+ 256,
25
+ 256,
26
+ 256,
27
+ 256,
28
+ 256
29
+ ],
30
+ "kla_blocks": [],
31
+ "mamba_blocks": [],
32
+ "gdn_blocks": [],
33
+ "gla_blocks": [],
34
+ "moe_blocks": [],
35
+ "d_state": 16,
36
+ "mamba_params": true,
37
+ "kla_kernel": true,
38
+ "mimo_rank": 1,
39
+ "skip_around_kla": true,
40
+ "decoder_mlp": false,
41
+ "use_reparametrisation_trick": true,
42
+ "num_experts": 4,
43
+ "moe_top_k": 2,
44
+ "moe_capacity_factor": 1.0,
45
+ "moe_aux_loss_weight": 0.01,
46
+ "moe_embedding_experts": false,
47
+ "num_iterations": -1,
48
+ "target_flops": -1.0,
49
+ "target_param_data_ratio": 20,
50
+ "data_dir": "base_data",
51
+ "device_batch_size": 64,
52
+ "total_batch_size": 524288,
53
+ "use_muon": true,
54
+ "embedding_lr": 0.3,
55
+ "unembedding_lr": 0.004,
56
+ "grad_clip": 1.0,
57
+ "weight_decay": 0.0,
58
+ "matrix_lr": 0.02,
59
+ "adam_beta1": 0.8,
60
+ "adam_beta2": 0.95,
61
+ "warmup_ratio": 0.05,
62
+ "warmdown_ratio": 0.4,
63
+ "final_lr_frac": 0.0,
64
+ "resume_from_step": -1,
65
+ "eval_every": 250,
66
+ "eval_tokens": 10485760,
67
+ "core_metric_every": -1,
68
+ "core_metric_max_per_task": 500,
69
+ "sample_every": 250,
70
+ "save_every": 1000,
71
+ "push_checkpoints_to_hub": true,
72
+ "use_profiler": false,
73
+ "profile_step": 2,
74
+ "profile_micro_step": 0,
75
+ "memory_history_max_entries": 10000,
76
+ "model_tag": "rva_6x256/stage_2",
77
+ "n_layer": 6,
78
+ "n_head": 16,
79
+ "n_kv_head": 16,
80
+ "n_embd": 256
81
+ },
82
+ "user_config": {
83
+ "run": "rva_6x256/stage_2",
84
+ "wandb_group": "rva_6x256",
85
+ "seed": 42,
86
+ "device_type": "cuda",
87
+ "config": "rva5m",
88
+ "depth": 6,
89
+ "aspect_ratio": 64,
90
+ "model_dim": 256,
91
+ "head_dim": 16,
92
+ "sequence_len": 1024,
93
+ "vocab_size": 265,
94
+ "clustering_init_num_samples": 0,
95
+ "rva_blocks": [
96
+ 0,
97
+ 1,
98
+ 2
99
+ ],
100
+ "recurrent_vocab_sizes": [
101
+ 265,
102
+ 256,
103
+ 256,
104
+ 256,
105
+ 256,
106
+ 256
107
+ ],
108
+ "kla_blocks": [],
109
+ "mamba_blocks": [],
110
+ "gdn_blocks": [],
111
+ "gla_blocks": [],
112
+ "moe_blocks": [],
113
+ "d_state": 16,
114
+ "mamba_params": true,
115
+ "kla_kernel": true,
116
+ "mimo_rank": 1,
117
+ "skip_around_kla": true,
118
+ "decoder_mlp": false,
119
+ "use_reparametrisation_trick": true,
120
+ "num_experts": 4,
121
+ "moe_top_k": 2,
122
+ "moe_capacity_factor": 1.0,
123
+ "moe_aux_loss_weight": 0.01,
124
+ "moe_embedding_experts": false,
125
+ "num_iterations": -1,
126
+ "target_flops": -1.0,
127
+ "target_param_data_ratio": 20,
128
+ "data_dir": "base_data",
129
+ "device_batch_size": 64,
130
+ "total_batch_size": 524288,
131
+ "use_muon": true,
132
+ "embedding_lr": 0.3,
133
+ "unembedding_lr": 0.004,
134
+ "grad_clip": 1.0,
135
+ "weight_decay": 0.0,
136
+ "matrix_lr": 0.02,
137
+ "adam_beta1": 0.8,
138
+ "adam_beta2": 0.95,
139
+ "warmup_ratio": 0.05,
140
+ "warmdown_ratio": 0.4,
141
+ "final_lr_frac": 0.0,
142
+ "resume_from_step": -1,
143
+ "eval_every": 250,
144
+ "eval_tokens": 10485760,
145
+ "core_metric_every": -1,
146
+ "core_metric_max_per_task": 500,
147
+ "sample_every": 250,
148
+ "save_every": 1000,
149
+ "push_checkpoints_to_hub": true,
150
+ "use_profiler": false,
151
+ "profile_step": 2,
152
+ "profile_micro_step": 0,
153
+ "memory_history_max_entries": 10000,
154
+ "model_tag": "rva_6x256/stage_2"
155
+ },
156
+ "device_batch_size": 64,
157
+ "sequence_len": 1024,
158
+ "dataloader_state_dict": {
159
+ "pq_idx": 0,
160
+ "rg_idx": 20
161
+ },
162
+ "loop_state": {
163
+ "min_val_bpb": 2.9875989317518807,
164
+ "smooth_train_loss": 11.878451741152228,
165
+ "smooth_aux_loss": 9.735462721314915,
166
+ "total_training_time": 49.51874494552612
167
+ }
168
+ }