ASzecsenyi commited on
Commit
2b1dd24
·
verified ·
1 Parent(s): ecc3355

Upload gpt_3x256_iuld0inq/meta_002000.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. gpt_3x256_iuld0inq/meta_002000.json +203 -0
gpt_3x256_iuld0inq/meta_002000.json ADDED
@@ -0,0 +1,203 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 2000,
3
+ "val_bpb": 1.5119289467762564,
4
+ "model_config": {
5
+ "run": "gpt_3x256",
6
+ "wandb_group": null,
7
+ "seed": 42,
8
+ "device_type": "cuda",
9
+ "config": "gpt5m",
10
+ "depth": 3,
11
+ "aspect_ratio": 64,
12
+ "model_dim": 256,
13
+ "head_dim": 128,
14
+ "sequence_len": 1024,
15
+ "prefill_len": 0,
16
+ "vocab_size": 265,
17
+ "rva_blocks": [],
18
+ "mamba_blocks": [],
19
+ "gdn_blocks": [],
20
+ "gla_blocks": [],
21
+ "moe_blocks": [],
22
+ "kla_blocks": [],
23
+ "recurrent_vocab_sizes": [],
24
+ "rva_balance_loss_weight": 0.0,
25
+ "rva_balance_loss_temperature": 1.0,
26
+ "rva_entropy_loss_weight": 0.0,
27
+ "rva_use_ema": true,
28
+ "rva_ema_decay": 0.99,
29
+ "rva_ema_epsilon": 1e-05,
30
+ "rva_use_random_reset": false,
31
+ "rva_reset_threshold": 1.0,
32
+ "rva_reset_prob": 1.0,
33
+ "rva_use_l2_norm": false,
34
+ "rva_use_fsq_ae": false,
35
+ "rva_fsq_levels": [
36
+ 4,
37
+ 4,
38
+ 4,
39
+ 4,
40
+ 4,
41
+ 4,
42
+ 4
43
+ ],
44
+ "rva_recon_loss_init": 1.0,
45
+ "rva_recon_loss_final": 0.01,
46
+ "rva_recon_anneal_fraction": 0.8,
47
+ "rva_unified_embed": false,
48
+ "rva_factored": true,
49
+ "rva_sub_vocab_size": 0,
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": 2870,
67
+ "target_flops": -1.0,
68
+ "target_param_data_ratio": 20,
69
+ "data_dir": "base_data",
70
+ "device_batch_size": 64,
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": "gpt_3x256",
93
+ "stage": null,
94
+ "n_layer": 3,
95
+ "n_head": 2,
96
+ "n_kv_head": 2,
97
+ "n_embd": 256
98
+ },
99
+ "user_config": {
100
+ "run": "gpt_3x256",
101
+ "wandb_group": null,
102
+ "seed": 42,
103
+ "device_type": "cuda",
104
+ "config": "gpt5m",
105
+ "depth": 3,
106
+ "aspect_ratio": 64,
107
+ "model_dim": 256,
108
+ "head_dim": 128,
109
+ "sequence_len": 1024,
110
+ "prefill_len": 0,
111
+ "vocab_size": 265,
112
+ "rva_blocks": [],
113
+ "mamba_blocks": [],
114
+ "gdn_blocks": [],
115
+ "gla_blocks": [],
116
+ "moe_blocks": [],
117
+ "kla_blocks": [],
118
+ "recurrent_vocab_sizes": [],
119
+ "rva_balance_loss_weight": 0.0,
120
+ "rva_balance_loss_temperature": 1.0,
121
+ "rva_entropy_loss_weight": 0.0,
122
+ "rva_use_ema": true,
123
+ "rva_ema_decay": 0.99,
124
+ "rva_ema_epsilon": 1e-05,
125
+ "rva_use_random_reset": false,
126
+ "rva_reset_threshold": 1.0,
127
+ "rva_reset_prob": 1.0,
128
+ "rva_use_l2_norm": false,
129
+ "rva_use_fsq_ae": false,
130
+ "rva_fsq_levels": [
131
+ 4,
132
+ 4,
133
+ 4,
134
+ 4,
135
+ 4,
136
+ 4,
137
+ 4
138
+ ],
139
+ "rva_recon_loss_init": 1.0,
140
+ "rva_recon_loss_final": 0.01,
141
+ "rva_recon_anneal_fraction": 0.8,
142
+ "rva_unified_embed": false,
143
+ "rva_factored": true,
144
+ "rva_sub_vocab_size": 0,
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": 2870,
162
+ "target_flops": -1.0,
163
+ "target_param_data_ratio": 20,
164
+ "data_dir": "base_data",
165
+ "device_batch_size": 64,
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": "gpt_3x256",
188
+ "stage": null
189
+ },
190
+ "stage": null,
191
+ "device_batch_size": 64,
192
+ "sequence_len": 1024,
193
+ "dataloader_state_dict": {
194
+ "pq_idx": 3,
195
+ "rg_idx": 48
196
+ },
197
+ "loop_state": {
198
+ "min_val_bpb": 1.5119289467762564,
199
+ "smooth_train_loss": 1.0546791909312665,
200
+ "smooth_aux_loss": 0.0,
201
+ "total_training_time": 171.4380488395691
202
+ }
203
+ }