Hugging Face's logo

ASzecsenyi
/
VQLM

Model card Files
xet
Community
ASzecsenyi commited on
Commit
700772f
·
verified ·
1 Parent(s): 07dc3a4

Upload gpt_2x256_kgqalj4y/meta_002870.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. gpt_2x256_kgqalj4y/meta_002870.json +196 -0
gpt_2x256_kgqalj4y/meta_002870.json ADDED
@@ -0,0 +1,196 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "step": 2870,
3
+ "val_bpb": 1.257881337768729,
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": 16384,
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
+ "use_skip_connection": true,
47
+ "clustering_init_num_samples": 0,
48
+ "max_active_tokens_in_batch": -1,
49
+ "max_active_token_fraction": 0.0,
50
+ "d_state": 16,
51
+ "mamba_params": true,
52
+ "kla_kernel": true,
53
+ "mimo_rank": 1,
54
+ "skip_around_kla": true,
55
+ "decoder_mlp": false,
56
+ "use_reparametrisation_trick": true,
57
+ "num_experts": 4,
58
+ "moe_top_k": 2,
59
+ "moe_capacity_factor": 1.0,
60
+ "moe_aux_loss_weight": 0.01,
61
+ "moe_embedding_experts": false,
62
+ "num_iterations": 2870,
63
+ "target_flops": -1.0,
64
+ "target_param_data_ratio": 20,
65
+ "data_dir": "base_data",
66
+ "device_batch_size": 256,
67
+ "total_batch_size": 524288,
68
+ "use_muon": true,
69
+ "embedding_lr": 0.3,
70
+ "unembedding_lr": 0.004,
71
+ "grad_clip": 1.0,
72
+ "weight_decay": 0.0,
73
+ "matrix_lr": 0.02,
74
+ "adam_beta1": 0.8,
75
+ "adam_beta2": 0.95,
76
+ "warmup_ratio": 0.05,
77
+ "warmdown_ratio": 0.4,
78
+ "final_lr_frac": 0.0,
79
+ "resume_from_step": -1,
80
+ "eval_every": 250,
81
+ "eval_tokens": 10485760,
82
+ "core_metric_every": -1,
83
+ "core_metric_max_per_task": 500,
84
+ "sample_every": 250,
85
+ "save_every": 1000,
86
+ "push_checkpoints_to_hub": true,
87
+ "profile_rva_codebook_usage": true,
88
+ "model_tag": "gpt_2x256",
89
+ "stage": null,
90
+ "n_layer": 2,
91
+ "n_head": 2,
92
+ "n_kv_head": 2,
93
+ "n_embd": 256
94
+ },
95
+ "user_config": {
96
+ "run": "gpt_2x256",
97
+ "wandb_group": null,
98
+ "seed": 42,
99
+ "device_type": "cuda",
100
+ "config": "gpt5m",
101
+ "depth": 2,
102
+ "aspect_ratio": 64,
103
+ "model_dim": 256,
104
+ "head_dim": 128,
105
+ "sequence_len": 1024,
106
+ "vocab_size": 16384,
107
+ "rva_blocks": [],
108
+ "mamba_blocks": [],
109
+ "gdn_blocks": [],
110
+ "gla_blocks": [],
111
+ "moe_blocks": [],
112
+ "kla_blocks": [],
113
+ "recurrent_vocab_sizes": [],
114
+ "rva_balance_loss_weight": 0.0,
115
+ "rva_balance_loss_temperature": 1.0,
116
+ "rva_entropy_loss_weight": 0.0,
117
+ "rva_use_ema": true,
118
+ "rva_ema_decay": 0.99,
119
+ "rva_ema_epsilon": 1e-05,
120
+ "rva_use_random_reset": false,
121
+ "rva_reset_threshold": 1.0,
122
+ "rva_reset_prob": 1.0,
123
+ "rva_use_l2_norm": false,
124
+ "rva_use_fsq_ae": false,
125
+ "rva_fsq_levels": [
126
+ 4,
127
+ 4,
128
+ 4,
129
+ 4,
130
+ 4,
131
+ 4,
132
+ 4
133
+ ],
134
+ "rva_recon_loss_init": 1.0,
135
+ "rva_recon_loss_final": 0.01,
136
+ "rva_recon_anneal_fraction": 0.8,
137
+ "use_skip_connection": true,
138
+ "clustering_init_num_samples": 0,
139
+ "max_active_tokens_in_batch": -1,
140
+ "max_active_token_fraction": 0.0,
141
+ "d_state": 16,
142
+ "mamba_params": true,
143
+ "kla_kernel": true,
144
+ "mimo_rank": 1,
145
+ "skip_around_kla": true,
146
+ "decoder_mlp": false,
147
+ "use_reparametrisation_trick": true,
148
+ "num_experts": 4,
149
+ "moe_top_k": 2,
150
+ "moe_capacity_factor": 1.0,
151
+ "moe_aux_loss_weight": 0.01,
152
+ "moe_embedding_experts": false,
153
+ "num_iterations": 2870,
154
+ "target_flops": -1.0,
155
+ "target_param_data_ratio": 20,
156
+ "data_dir": "base_data",
157
+ "device_batch_size": 256,
158
+ "total_batch_size": 524288,
159
+ "use_muon": true,
160
+ "embedding_lr": 0.3,
161
+ "unembedding_lr": 0.004,
162
+ "grad_clip": 1.0,
163
+ "weight_decay": 0.0,
164
+ "matrix_lr": 0.02,
165
+ "adam_beta1": 0.8,
166
+ "adam_beta2": 0.95,
167
+ "warmup_ratio": 0.05,
168
+ "warmdown_ratio": 0.4,
169
+ "final_lr_frac": 0.0,
170
+ "resume_from_step": -1,
171
+ "eval_every": 250,
172
+ "eval_tokens": 10485760,
173
+ "core_metric_every": -1,
174
+ "core_metric_max_per_task": 500,
175
+ "sample_every": 250,
176
+ "save_every": 1000,
177
+ "push_checkpoints_to_hub": true,
178
+ "profile_rva_codebook_usage": true,
179
+ "model_tag": "gpt_2x256",
180
+ "stage": null,
181
+ "pod_name": "s2027538-infk8s-job-rbwss-j7v45"
182
+ },
183
+ "stage": null,
184
+ "device_batch_size": 256,
185
+ "sequence_len": 1024,
186
+ "dataloader_state_dict": {
187
+ "pq_idx": 25,
188
+ "rg_idx": 30
189
+ },
190
+ "loop_state": {
191
+ "min_val_bpb": 1.257881337768729,
192
+ "smooth_train_loss": 3.804668989854215,
193
+ "smooth_aux_loss": 0.0,
194
+ "total_training_time": 753.8793840408325
195
+ }
196
+ }