Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
350de47
·
verified ·
1 Parent(s): f090883

Upload rva_3x256_ftmaoqyr/meta_000882.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. rva_3x256_ftmaoqyr/meta_000882.json +204 -0
rva_3x256_ftmaoqyr/meta_000882.json ADDED
@@ -0,0 +1,204 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 882,
3
+ "val_bpb": 1.4462567441149232,
4
+ "model_config": {
5
+ "run": "rva_3x256",
6
+ "wandb_group": null,
7
+ "seed": 42,
8
+ "device_type": "cuda",
9
+ "config": "rvamix_1l",
10
+ "depth": 3,
11
+ "aspect_ratio": 64,
12
+ "model_dim": 256,
13
+ "head_dim": 16,
14
+ "sequence_len": 1024,
15
+ "vocab_size": 16384,
16
+ "rva_blocks": [
17
+ 0
18
+ ],
19
+ "mamba_blocks": [],
20
+ "gdn_blocks": [],
21
+ "gla_blocks": [],
22
+ "moe_blocks": [],
23
+ "kla_blocks": [],
24
+ "recurrent_vocab_sizes": [
25
+ 16383
26
+ ],
27
+ "rva_balance_loss_weight": 0.0,
28
+ "rva_balance_loss_temperature": 1.0,
29
+ "rva_entropy_loss_weight": 0.0,
30
+ "rva_use_ema": true,
31
+ "rva_ema_decay": 0.99,
32
+ "rva_ema_epsilon": 1e-05,
33
+ "rva_use_random_reset": false,
34
+ "rva_reset_threshold": 1.0,
35
+ "rva_reset_prob": 1.0,
36
+ "rva_use_l2_norm": false,
37
+ "rva_use_fsq_ae": false,
38
+ "rva_fsq_levels": [
39
+ 4,
40
+ 4,
41
+ 4,
42
+ 4,
43
+ 4,
44
+ 4,
45
+ 4
46
+ ],
47
+ "rva_recon_loss_init": 1.0,
48
+ "rva_recon_loss_final": 0.01,
49
+ "rva_recon_anneal_fraction": 0.8,
50
+ "use_skip_connection": true,
51
+ "clustering_init_num_samples": 0,
52
+ "max_active_tokens_in_batch": -1,
53
+ "max_active_token_fraction": 0.0,
54
+ "d_state": 16,
55
+ "mamba_params": true,
56
+ "kla_kernel": true,
57
+ "mimo_rank": 1,
58
+ "skip_around_kla": true,
59
+ "decoder_mlp": false,
60
+ "use_reparametrisation_trick": true,
61
+ "num_experts": 4,
62
+ "moe_top_k": 2,
63
+ "moe_capacity_factor": 1.0,
64
+ "moe_aux_loss_weight": 0.01,
65
+ "moe_embedding_experts": false,
66
+ "num_iterations": -1,
67
+ "target_flops": -1.0,
68
+ "target_param_data_ratio": 20,
69
+ "data_dir": "base_data",
70
+ "device_batch_size": 8,
71
+ "total_batch_size": 524288,
72
+ "use_muon": true,
73
+ "embedding_lr": 0.3,
74
+ "unembedding_lr": 0.004,
75
+ "grad_clip": 1.0,
76
+ "weight_decay": 0.0,
77
+ "matrix_lr": 0.02,
78
+ "adam_beta1": 0.8,
79
+ "adam_beta2": 0.95,
80
+ "warmup_ratio": 0.05,
81
+ "warmdown_ratio": 0.4,
82
+ "final_lr_frac": 0.0,
83
+ "resume_from_step": -1,
84
+ "eval_every": 250,
85
+ "eval_tokens": 10485760,
86
+ "core_metric_every": -1,
87
+ "core_metric_max_per_task": 500,
88
+ "sample_every": 250,
89
+ "save_every": 1000,
90
+ "push_checkpoints_to_hub": true,
91
+ "profile_rva_codebook_usage": true,
92
+ "model_tag": "rva_3x256",
93
+ "stage": null,
94
+ "n_layer": 3,
95
+ "n_head": 16,
96
+ "n_kv_head": 16,
97
+ "n_embd": 256
98
+ },
99
+ "user_config": {
100
+ "run": "rva_3x256",
101
+ "wandb_group": null,
102
+ "seed": 42,
103
+ "device_type": "cuda",
104
+ "config": "rvamix_1l",
105
+ "depth": 3,
106
+ "aspect_ratio": 64,
107
+ "model_dim": 256,
108
+ "head_dim": 16,
109
+ "sequence_len": 1024,
110
+ "vocab_size": 16384,
111
+ "rva_blocks": [
112
+ 0
113
+ ],
114
+ "mamba_blocks": [],
115
+ "gdn_blocks": [],
116
+ "gla_blocks": [],
117
+ "moe_blocks": [],
118
+ "kla_blocks": [],
119
+ "recurrent_vocab_sizes": [
120
+ 16383
121
+ ],
122
+ "rva_balance_loss_weight": 0.0,
123
+ "rva_balance_loss_temperature": 1.0,
124
+ "rva_entropy_loss_weight": 0.0,
125
+ "rva_use_ema": true,
126
+ "rva_ema_decay": 0.99,
127
+ "rva_ema_epsilon": 1e-05,
128
+ "rva_use_random_reset": false,
129
+ "rva_reset_threshold": 1.0,
130
+ "rva_reset_prob": 1.0,
131
+ "rva_use_l2_norm": false,
132
+ "rva_use_fsq_ae": false,
133
+ "rva_fsq_levels": [
134
+ 4,
135
+ 4,
136
+ 4,
137
+ 4,
138
+ 4,
139
+ 4,
140
+ 4
141
+ ],
142
+ "rva_recon_loss_init": 1.0,
143
+ "rva_recon_loss_final": 0.01,
144
+ "rva_recon_anneal_fraction": 0.8,
145
+ "use_skip_connection": true,
146
+ "clustering_init_num_samples": 0,
147
+ "max_active_tokens_in_batch": -1,
148
+ "max_active_token_fraction": 0.0,
149
+ "d_state": 16,
150
+ "mamba_params": true,
151
+ "kla_kernel": true,
152
+ "mimo_rank": 1,
153
+ "skip_around_kla": true,
154
+ "decoder_mlp": false,
155
+ "use_reparametrisation_trick": true,
156
+ "num_experts": 4,
157
+ "moe_top_k": 2,
158
+ "moe_capacity_factor": 1.0,
159
+ "moe_aux_loss_weight": 0.01,
160
+ "moe_embedding_experts": false,
161
+ "num_iterations": -1,
162
+ "target_flops": -1.0,
163
+ "target_param_data_ratio": 20,
164
+ "data_dir": "base_data",
165
+ "device_batch_size": 8,
166
+ "total_batch_size": 524288,
167
+ "use_muon": true,
168
+ "embedding_lr": 0.3,
169
+ "unembedding_lr": 0.004,
170
+ "grad_clip": 1.0,
171
+ "weight_decay": 0.0,
172
+ "matrix_lr": 0.02,
173
+ "adam_beta1": 0.8,
174
+ "adam_beta2": 0.95,
175
+ "warmup_ratio": 0.05,
176
+ "warmdown_ratio": 0.4,
177
+ "final_lr_frac": 0.0,
178
+ "resume_from_step": -1,
179
+ "eval_every": 250,
180
+ "eval_tokens": 10485760,
181
+ "core_metric_every": -1,
182
+ "core_metric_max_per_task": 500,
183
+ "sample_every": 250,
184
+ "save_every": 1000,
185
+ "push_checkpoints_to_hub": true,
186
+ "profile_rva_codebook_usage": true,
187
+ "model_tag": "rva_3x256",
188
+ "stage": null,
189
+ "pod_name": "s2027538-infk8s-job-98cll-x5j6x"
190
+ },
191
+ "stage": null,
192
+ "device_batch_size": 8,
193
+ "sequence_len": 1024,
194
+ "dataloader_state_dict": {
195
+ "pq_idx": 7,
196
+ "rg_idx": 44
197
+ },
198
+ "loop_state": {
199
+ "min_val_bpb": 1.4462567441149232,
200
+ "smooth_train_loss": 4.310881630424784,
201
+ "smooth_aux_loss": 0.0,
202
+ "total_training_time": 2368.4895277023315
203
+ }
204
+ }