Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
 Community
ASzecsenyi commited on
Commit
737f045
·
verified ·
1 Parent(s): 774f970

Upload gpt_2x256_ux965z3n/meta_002870.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. gpt_2x256_ux965z3n/meta_002870.json +198 -0
gpt_2x256_ux965z3n/meta_002870.json ADDED
@@ -0,0 +1,198 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 2870,
3
+ "val_bpb": 1.6411318805399224,
4
+ "model_config": {
5
+ "run": "gpt_2x256",
6
+ "wandb_group": null,
7
+ "seed": 42,
8
+ "device_type": "cuda",
9
+ "config": "gpt5m",
10
+ "depth": 2,
11
+ "aspect_ratio": 64,
12
+ "model_dim": 256,
13
+ "head_dim": 128,
14
+ "sequence_len": 1024,
15
+ "vocab_size": 265,
16
+ "rva_blocks": [],
17
+ "mamba_blocks": [],
18
+ "gdn_blocks": [],
19
+ "gla_blocks": [],
20
+ "moe_blocks": [],
21
+ "kla_blocks": [],
22
+ "recurrent_vocab_sizes": [],
23
+ "rva_balance_loss_weight": 0.0,
24
+ "rva_balance_loss_temperature": 1.0,
25
+ "rva_entropy_loss_weight": 0.0,
26
+ "rva_use_ema": true,
27
+ "rva_ema_decay": 0.99,
28
+ "rva_ema_epsilon": 1e-05,
29
+ "rva_use_random_reset": false,
30
+ "rva_reset_threshold": 1.0,
31
+ "rva_reset_prob": 1.0,
32
+ "rva_use_l2_norm": false,
33
+ "rva_use_fsq_ae": false,
34
+ "rva_fsq_levels": [
35
+ 4,
36
+ 4,
37
+ 4,
38
+ 4,
39
+ 4,
40
+ 4,
41
+ 4
42
+ ],
43
+ "rva_recon_loss_init": 1.0,
44
+ "rva_recon_loss_final": 0.01,
45
+ "rva_recon_anneal_fraction": 0.8,
46
+ "rva_unified_embed": false,
47
+ "use_skip_connection": true,
48
+ "clustering_init_num_samples": 0,
49
+ "max_active_tokens_in_batch": -1,
50
+ "max_active_token_fraction": 0.0,
51
+ "d_state": 16,
52
+ "mamba_params": true,
53
+ "kla_kernel": true,
54
+ "mimo_rank": 1,
55
+ "skip_around_kla": true,
56
+ "decoder_mlp": false,
57
+ "use_reparametrisation_trick": true,
58
+ "num_experts": 4,
59
+ "moe_top_k": 2,
60
+ "moe_capacity_factor": 1.0,
61
+ "moe_aux_loss_weight": 0.01,
62
+ "moe_embedding_experts": false,
63
+ "num_iterations": 2870,
64
+ "target_flops": -1.0,
65
+ "target_param_data_ratio": 20,
66
+ "data_dir": "base_data",
67
+ "device_batch_size": 8,
68
+ "total_batch_size": 524288,
69
+ "use_muon": true,
70
+ "embedding_lr": 0.3,
71
+ "unembedding_lr": 0.004,
72
+ "grad_clip": 1.0,
73
+ "weight_decay": 0.0,
74
+ "matrix_lr": 0.02,
75
+ "adam_beta1": 0.8,
76
+ "adam_beta2": 0.95,
77
+ "warmup_ratio": 0.05,
78
+ "warmdown_ratio": 0.4,
79
+ "final_lr_frac": 0.0,
80
+ "resume_from_step": -1,
81
+ "eval_every": 250,
82
+ "eval_tokens": 10485760,
83
+ "core_metric_every": -1,
84
+ "core_metric_max_per_task": 500,
85
+ "sample_every": 250,
86
+ "save_every": 1000,
87
+ "push_checkpoints_to_hub": true,
88
+ "profile_rva_codebook_usage": true,
89
+ "model_tag": "gpt_2x256",
90
+ "stage": null,
91
+ "n_layer": 2,
92
+ "n_head": 2,
93
+ "n_kv_head": 2,
94
+ "n_embd": 256
95
+ },
96
+ "user_config": {
97
+ "run": "gpt_2x256",
98
+ "wandb_group": null,
99
+ "seed": 42,
100
+ "device_type": "cuda",
101
+ "config": "gpt5m",
102
+ "depth": 2,
103
+ "aspect_ratio": 64,
104
+ "model_dim": 256,
105
+ "head_dim": 128,
106
+ "sequence_len": 1024,
107
+ "vocab_size": 265,
108
+ "rva_blocks": [],
109
+ "mamba_blocks": [],
110
+ "gdn_blocks": [],
111
+ "gla_blocks": [],
112
+ "moe_blocks": [],
113
+ "kla_blocks": [],
114
+ "recurrent_vocab_sizes": [],
115
+ "rva_balance_loss_weight": 0.0,
116
+ "rva_balance_loss_temperature": 1.0,
117
+ "rva_entropy_loss_weight": 0.0,
118
+ "rva_use_ema": true,
119
+ "rva_ema_decay": 0.99,
120
+ "rva_ema_epsilon": 1e-05,
121
+ "rva_use_random_reset": false,
122
+ "rva_reset_threshold": 1.0,
123
+ "rva_reset_prob": 1.0,
124
+ "rva_use_l2_norm": false,
125
+ "rva_use_fsq_ae": false,
126
+ "rva_fsq_levels": [
127
+ 4,
128
+ 4,
129
+ 4,
130
+ 4,
131
+ 4,
132
+ 4,
133
+ 4
134
+ ],
135
+ "rva_recon_loss_init": 1.0,
136
+ "rva_recon_loss_final": 0.01,
137
+ "rva_recon_anneal_fraction": 0.8,
138
+ "rva_unified_embed": false,
139
+ "use_skip_connection": true,
140
+ "clustering_init_num_samples": 0,
141
+ "max_active_tokens_in_batch": -1,
142
+ "max_active_token_fraction": 0.0,
143
+ "d_state": 16,
144
+ "mamba_params": true,
145
+ "kla_kernel": true,
146
+ "mimo_rank": 1,
147
+ "skip_around_kla": true,
148
+ "decoder_mlp": false,
149
+ "use_reparametrisation_trick": true,
150
+ "num_experts": 4,
151
+ "moe_top_k": 2,
152
+ "moe_capacity_factor": 1.0,
153
+ "moe_aux_loss_weight": 0.01,
154
+ "moe_embedding_experts": false,
155
+ "num_iterations": 2870,
156
+ "target_flops": -1.0,
157
+ "target_param_data_ratio": 20,
158
+ "data_dir": "base_data",
159
+ "device_batch_size": 8,
160
+ "total_batch_size": 524288,
161
+ "use_muon": true,
162
+ "embedding_lr": 0.3,
163
+ "unembedding_lr": 0.004,
164
+ "grad_clip": 1.0,
165
+ "weight_decay": 0.0,
166
+ "matrix_lr": 0.02,
167
+ "adam_beta1": 0.8,
168
+ "adam_beta2": 0.95,
169
+ "warmup_ratio": 0.05,
170
+ "warmdown_ratio": 0.4,
171
+ "final_lr_frac": 0.0,
172
+ "resume_from_step": -1,
173
+ "eval_every": 250,
174
+ "eval_tokens": 10485760,
175
+ "core_metric_every": -1,
176
+ "core_metric_max_per_task": 500,
177
+ "sample_every": 250,
178
+ "save_every": 1000,
179
+ "push_checkpoints_to_hub": true,
180
+ "profile_rva_codebook_usage": true,
181
+ "model_tag": "gpt_2x256",
182
+ "stage": null,
183
+ "pod_name": "s2027538-infk8s-job-7pb2t-6n4pj"
184
+ },
185
+ "stage": null,
186
+ "device_batch_size": 8,
187
+ "sequence_len": 1024,
188
+ "dataloader_state_dict": {
189
+ "pq_idx": 5,
190
+ "rg_idx": 46
191
+ },
192
+ "loop_state": {
193
+ "min_val_bpb": 1.6411318805399224,
194
+ "smooth_train_loss": 1.1934989348362761,
195
+ "smooth_aux_loss": 0.0,
196
+ "total_training_time": 843.7530717849731
197
+ }
198
+ }