Hugging Face's logo

iskhare
/
iclr-debug

Text Generation
English
research
Model card Files
xet
 Community
iskhare commited on
Commit
e354dc0
·
verified ·
1 Parent(s): 55f7176

Back up iclr-debug-ablation-no-ema-13p5b-20260919 decoder-mstok step 10000

Browse files
Files changed (18) hide show
  1. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/config.yaml +229 -0
  2. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/EVAL_DONE.json +5 -0
  3. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-0.json +0 -0
  4. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-0.samples.json +0 -0
  5. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-1.json +0 -0
  6. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-1.samples.json +0 -0
  7. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-2.json +0 -0
  8. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-2.samples.json +0 -0
  9. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-3.json +0 -0
  10. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-3.samples.json +0 -0
  11. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-4.json +0 -0
  12. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-4.samples.json +0 -0
  13. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/summary.json +35 -0
  14. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/ncp.pt +3 -0
  15. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/vqvae-teacher.pt +3 -0
  16. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/vqvae.pt +3 -0
  17. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/milestone-iter-10000.pt +3 -0
  18. ablations/iclr-debug-ablation-no-ema-13p5b-20260919/manifest.json +528 -0
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/config.yaml ADDED
@@ -0,0 +1,229 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ task: mstok-next-concept
2
+ experiment: iclr-debug-ablation-no-ema-13p5b-20260919-decoder-mstok
3
+ experiment_dir: /home/ubuntu/mstok-results/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok
4
+ dataset:
5
+ train_source: /home/ubuntu/data/full_owt/train_gpt2.bin
6
+ validate_source: /home/ubuntu/data/full_owt/valid_gpt2.bin
7
+ pad_token_id: 50257
8
+ pre_tokenizer:
9
+ name: hf
10
+ tokenizer: hf
11
+ model_id: gpt2
12
+ special_tokens:
13
+ eos_token: <|endoftext|>
14
+ pad_token: <|pad|>
15
+ codec:
16
+ n_layers: 6
17
+ context_length: 256
18
+ embed_dim: 384
19
+ in_vocab_size: 50304
20
+ vqvae_vocab_size:
21
+ - 16384
22
+ - 16384
23
+ - 16384
24
+ - 16384
25
+ - 16384
26
+ - 16384
27
+ - 16384
28
+ - 16384
29
+ - 16384
30
+ - 16384
31
+ - 16384
32
+ - 16384
33
+ - 16384
34
+ - 16384
35
+ - 16384
36
+ - 16384
37
+ compression_factor: 1
38
+ dropout: 0.1
39
+ pre_quant_groupnorm: 4
40
+ pre_quant_dropout: 0.2
41
+ vector_quantizer_config:
42
+ clss: multiscale_residual_vector_quantizer
43
+ decay: 0.99
44
+ epsilon: 1.0e-05
45
+ commitment_cost: 0.25
46
+ learned_l1_sampling: true
47
+ learned_all_sampling: false
48
+ quant_resi:
49
+ enabled: true
50
+ ratio: 0.5
51
+ share_mode: 0
52
+ learnable_ratio: false
53
+ levels:
54
+ use_manual_levels: true
55
+ manual_levels:
56
+ - 1
57
+ - 4
58
+ - 9
59
+ - 16
60
+ - 25
61
+ - 36
62
+ - 49
63
+ - 64
64
+ - 81
65
+ - 100
66
+ - 121
67
+ - 144
68
+ - 169
69
+ - 196
70
+ - 225
71
+ - 256
72
+ aux:
73
+ fine_drop:
74
+ prob: 0.5
75
+ min_keep: 1
76
+ generator:
77
+ n_layer: 12
78
+ n_head: 12
79
+ bias: true
80
+ dropout: 0.1
81
+ n_embd: 768
82
+ context_length: 256
83
+ vocab_size:
84
+ - 16384
85
+ - 16384
86
+ - 16384
87
+ - 16384
88
+ - 16384
89
+ - 16384
90
+ - 16384
91
+ - 16384
92
+ - 16384
93
+ - 16384
94
+ - 16384
95
+ - 16384
96
+ - 16384
97
+ - 16384
98
+ - 16384
99
+ attn_pattern: block_diagonal
100
+ use_positional_encoding: true
101
+ use_level_encoding: true
102
+ prefix_len: 0
103
+ use_rope: true
104
+ rope_base: 10000
105
+ use_qk_norm: true
106
+ post_upsample_conv:
107
+ enabled: true
108
+ kernel_size: 3
109
+ use_relu2: true
110
+ use_flex_attention: true
111
+ shared_output_head: true
112
+ shared_head_per_level_bias: false
113
+ shared_head_per_level_scale: false
114
+ shared_head_adapter_rank: 0
115
+ objectives:
116
+ codec_weight: 1.0
117
+ ncp_weight: 1.0
118
+ soft_assignment_temperature: 1.0
119
+ prediction_temperature: 1.0
120
+ mstok_weight: 0.25
121
+ residual_weight: 1.0
122
+ reconstruction_weight: 1.0
123
+ teacher_ema_decay: 0.0
124
+ teacher_ema_warmup_steps: 1000
125
+ mstok_warmup_steps: 500
126
+ optimization:
127
+ codec_lr: 0.001
128
+ codec_min_lr: 0.0001
129
+ codec_warmup_iters: 0
130
+ codec_lr_decay_iters: 171662
131
+ generator_lr: 0.0005
132
+ generator_min_lr: 1.0e-05
133
+ generator_warmup_iters: 300
134
+ generator_lr_decay_iters: 171662
135
+ beta_1: 0.9
136
+ codec_beta_2: 0.99
137
+ generator_beta_2: 0.99
138
+ weight_decay: 0.1
139
+ max_grad_norm: 1.0
140
+ level_loss_alpha: 1.0
141
+ grad_accumulation_steps: 12
142
+ corruption:
143
+ mode: per_level
144
+ per_level_probs:
145
+ - 0.85
146
+ - 0.8321428571
147
+ - 0.8142857143
148
+ - 0.7964285714
149
+ - 0.7785714286
150
+ - 0.7607142857
151
+ - 0.7428571429
152
+ - 0.725
153
+ - 0.7071428571
154
+ - 0.6892857143
155
+ - 0.6714285714
156
+ - 0.6535714286
157
+ - 0.6357142857
158
+ - 0.6178571429
159
+ - 0.6
160
+ skip_level0: true
161
+ training:
162
+ log_interval: 10
163
+ eval_interval: 1000
164
+ checkpoint_interval: 1000
165
+ eval_batch_size: 4
166
+ val_iters: 25
167
+ keep_last: 3
168
+ seed: 55
169
+ codec_initialization_seed: 42
170
+ generator_initialization_seed: 55
171
+ total_iters: 171662
172
+ batch_size: 32
173
+ expected_world_size: 8
174
+ expected_global_batch_size: 3072
175
+ resume_checkpoint: null
176
+ milestone_steps:
177
+ - 1272
178
+ - 6358
179
+ - 10000
180
+ - 17167
181
+ generation_eval_steps:
182
+ - 10000
183
+ - 17167
184
+ torch_compile:
185
+ enable: true
186
+ scope: loss
187
+ mode: default
188
+ dynamic: false
189
+ fullgraph: true
190
+ backend: inductor
191
+ mixed_precision:
192
+ enable: true
193
+ wandb:
194
+ entity: mstok
195
+ project: iclr-debug
196
+ group: full-owt-ctx256-16sq-135b-v1
197
+ enable: true
198
+ id: iclr-debug-ablation-no-ema-13p5b-20260919-decoder-mstok
199
+ resume: allow
200
+ gradients_and_params:
201
+ enable: false
202
+ log: all
203
+ log_freq: 1000
204
+ export:
205
+ vqvae_config_template: /home/ubuntu/mstok-no-ema/config/repro-ctx256/vqvae.yaml
206
+ ncp_config_template: /home/ubuntu/mstok-no-ema/config/repro-ctx256/ncp-sharedhead.yaml
207
+ semantic:
208
+ enabled: true
209
+ mode: eostok
210
+ weight: 0.5
211
+ warmup_steps: 500
212
+ teacher_id: FacebookAI/roberta-base
213
+ teacher_revision: e2da8e2f811d1448a5b465c236feacd80ffbac7b
214
+ tokenizer_revision: 607a30d783dfa663caf39e06633721c8d4cfcd7e
215
+ teacher_dim: 768
216
+ feature_layer: 6
217
+ projector_dim: 2048
218
+ projector_seed: 56
219
+ alignment_site: decoder
220
+ iclr_debug:
221
+ stage: decoder-mstok
222
+ target_positions: 135000000000
223
+ lr_horizon_positions: 135000000000
224
+ stop_step: 17167
225
+ pilot: false
226
+ tokenizer_checkpoint: null
227
+ tokenizer_steps: null
228
+ budget_unit: input positions including padding; log non-padding tokens separately
229
+ ablation: no-ema
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/EVAL_DONE.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "step": 10000,
3
+ "reference_dtype": "float32",
4
+ "sampling": "supplied-level0-untruncated"
5
+ }
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-0.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-0.samples.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-1.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-1.samples.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-2.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-2.samples.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-3.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-3.samples.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-4.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/random-seed-4.samples.json ADDED
The diff for this file is too large to render. See raw diff
 
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/evaluation/step-10000/summary.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "protocol": {
3
+ "sampling": "random",
4
+ "temperature": 1.0,
5
+ "top_k": 0,
6
+ "top_p": 1.0,
7
+ "n_provided_levels": 1,
8
+ "n_levels": 16,
9
+ "document_aware_validation_sampling": true
10
+ },
11
+ "reference_model": "gpt2-large",
12
+ "checkpoint": "/home/ubuntu/mstok-results/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/ncp.pt",
13
+ "checkpoint_step": 10000,
14
+ "seeds": [
15
+ 0,
16
+ 1,
17
+ 2,
18
+ 3,
19
+ 4
20
+ ],
21
+ "samples_per_seed": 128,
22
+ "total_scored_samples": 640,
23
+ "total_scored_tokens": 159664,
24
+ "per_seed_mean_ppl": {
25
+ "0": 515.7421922074567,
26
+ "1": 491.861932795588,
27
+ "2": 495.9538945625993,
28
+ "3": 490.08621246773805,
29
+ "4": 450.5377560331705
30
+ },
31
+ "mean_gen_ppl": 488.8363976133105,
32
+ "gen_ppl_se": 10.61150882947165,
33
+ "mean_of_seed_median_ppl": 478.26988525390624,
34
+ "mean_entropy_nats": 4.539152341481356
35
+ }
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/ncp.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f431e66d03915e3f0e76e6612ff3c13ca933af843790b405c46d35bf3743f7bf
3
+ size 427469662
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/vqvae-teacher.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6957b55f6f04abd008f2e706417d73229a1f8dde22b39f2a83ec36ef52f442ff
3
+ size 1300865189
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/exports/step-10000/vqvae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52cd2f32483b8846b2356d8a749c6d5371407b8de78d10a31ca648ea91a3e0e9
3
+ size 1300864149
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok/milestone-iter-10000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df5bde94e964a58385f2b317a6b5f720a4fab2d1af79ecc326bb70c2eb303423
3
+ size 4879454945
ablations/iclr-debug-ablation-no-ema-13p5b-20260919/manifest.json ADDED
@@ -0,0 +1,528 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "iclr-debug-production-v1",
3
+ "variant": "decoder-mstok",
4
+ "configs": {
5
+ "decoder-mstok": {
6
+ "task": "mstok-next-concept",
7
+ "experiment": "iclr-debug-ablation-no-ema-13p5b-20260919-decoder-mstok",
8
+ "experiment_dir": "/home/ubuntu/mstok-results/iclr-debug-ablation-no-ema-13p5b-20260919/decoder-mstok",
9
+ "dataset": {
10
+ "train_source": "/home/ubuntu/data/full_owt/train_gpt2.bin",
11
+ "validate_source": "/home/ubuntu/data/full_owt/valid_gpt2.bin",
12
+ "pad_token_id": 50257
13
+ },
14
+ "pre_tokenizer": {
15
+ "name": "hf",
16
+ "tokenizer": "hf",
17
+ "model_id": "gpt2",
18
+ "special_tokens": {
19
+ "eos_token": "<|endoftext|>",
20
+ "pad_token": "<|pad|>"
21
+ }
22
+ },
23
+ "codec": {
24
+ "n_layers": 6,
25
+ "context_length": 256,
26
+ "embed_dim": 384,
27
+ "in_vocab_size": 50304,
28
+ "vqvae_vocab_size": [
29
+ 16384,
30
+ 16384,
31
+ 16384,
32
+ 16384,
33
+ 16384,
34
+ 16384,
35
+ 16384,
36
+ 16384,
37
+ 16384,
38
+ 16384,
39
+ 16384,
40
+ 16384,
41
+ 16384,
42
+ 16384,
43
+ 16384,
44
+ 16384
45
+ ],
46
+ "compression_factor": 1,
47
+ "dropout": 0.1,
48
+ "pre_quant_groupnorm": 4,
49
+ "pre_quant_dropout": 0.2,
50
+ "vector_quantizer_config": {
51
+ "clss": "multiscale_residual_vector_quantizer",
52
+ "decay": 0.99,
53
+ "epsilon": 1e-05,
54
+ "commitment_cost": 0.25,
55
+ "learned_l1_sampling": true,
56
+ "learned_all_sampling": false,
57
+ "quant_resi": {
58
+ "enabled": true,
59
+ "ratio": 0.5,
60
+ "share_mode": 0,
61
+ "learnable_ratio": false
62
+ },
63
+ "levels": {
64
+ "use_manual_levels": true,
65
+ "manual_levels": [
66
+ 1,
67
+ 4,
68
+ 9,
69
+ 16,
70
+ 25,
71
+ 36,
72
+ 49,
73
+ 64,
74
+ 81,
75
+ 100,
76
+ 121,
77
+ 144,
78
+ 169,
79
+ 196,
80
+ 225,
81
+ 256
82
+ ]
83
+ },
84
+ "aux": {
85
+ "fine_drop": {
86
+ "prob": 0.5,
87
+ "min_keep": 1
88
+ }
89
+ }
90
+ }
91
+ },
92
+ "generator": {
93
+ "n_layer": 12,
94
+ "n_head": 12,
95
+ "bias": true,
96
+ "dropout": 0.1,
97
+ "n_embd": 768,
98
+ "context_length": 256,
99
+ "vocab_size": [
100
+ 16384,
101
+ 16384,
102
+ 16384,
103
+ 16384,
104
+ 16384,
105
+ 16384,
106
+ 16384,
107
+ 16384,
108
+ 16384,
109
+ 16384,
110
+ 16384,
111
+ 16384,
112
+ 16384,
113
+ 16384,
114
+ 16384
115
+ ],
116
+ "attn_pattern": "block_diagonal",
117
+ "use_positional_encoding": true,
118
+ "use_level_encoding": true,
119
+ "prefix_len": 0,
120
+ "use_rope": true,
121
+ "rope_base": 10000,
122
+ "use_qk_norm": true,
123
+ "post_upsample_conv": {
124
+ "enabled": true,
125
+ "kernel_size": 3
126
+ },
127
+ "use_relu2": true,
128
+ "use_flex_attention": true,
129
+ "shared_output_head": true,
130
+ "shared_head_per_level_bias": false,
131
+ "shared_head_per_level_scale": false,
132
+ "shared_head_adapter_rank": 0
133
+ },
134
+ "objectives": {
135
+ "codec_weight": 1.0,
136
+ "ncp_weight": 1.0,
137
+ "soft_assignment_temperature": 1.0,
138
+ "prediction_temperature": 1.0,
139
+ "mstok_weight": 0.25,
140
+ "residual_weight": 1.0,
141
+ "reconstruction_weight": 1.0,
142
+ "teacher_ema_decay": 0.0,
143
+ "teacher_ema_warmup_steps": 1000,
144
+ "mstok_warmup_steps": 500
145
+ },
146
+ "optimization": {
147
+ "codec_lr": 0.001,
148
+ "codec_min_lr": 0.0001,
149
+ "codec_warmup_iters": 0,
150
+ "codec_lr_decay_iters": 171662,
151
+ "generator_lr": 0.0005,
152
+ "generator_min_lr": 1e-05,
153
+ "generator_warmup_iters": 300,
154
+ "generator_lr_decay_iters": 171662,
155
+ "beta_1": 0.9,
156
+ "codec_beta_2": 0.99,
157
+ "generator_beta_2": 0.99,
158
+ "weight_decay": 0.1,
159
+ "max_grad_norm": 1.0,
160
+ "level_loss_alpha": 1.0,
161
+ "grad_accumulation_steps": 12,
162
+ "corruption": {
163
+ "mode": "per_level",
164
+ "per_level_probs": [
165
+ 0.85,
166
+ 0.8321428571,
167
+ 0.8142857143,
168
+ 0.7964285714,
169
+ 0.7785714286,
170
+ 0.7607142857,
171
+ 0.7428571429,
172
+ 0.725,
173
+ 0.7071428571,
174
+ 0.6892857143,
175
+ 0.6714285714,
176
+ 0.6535714286,
177
+ 0.6357142857,
178
+ 0.6178571429,
179
+ 0.6
180
+ ],
181
+ "skip_level0": true
182
+ }
183
+ },
184
+ "training": {
185
+ "log_interval": 10,
186
+ "eval_interval": 1000,
187
+ "checkpoint_interval": 1000,
188
+ "eval_batch_size": 4,
189
+ "val_iters": 25,
190
+ "keep_last": 3,
191
+ "seed": 55,
192
+ "codec_initialization_seed": 42,
193
+ "generator_initialization_seed": 55,
194
+ "total_iters": 171662,
195
+ "batch_size": 32,
196
+ "expected_world_size": 8,
197
+ "expected_global_batch_size": 3072,
198
+ "resume_checkpoint": null,
199
+ "milestone_steps": [
200
+ 1272,
201
+ 6358,
202
+ 10000,
203
+ 17167
204
+ ],
205
+ "generation_eval_steps": [
206
+ 10000,
207
+ 17167
208
+ ]
209
+ },
210
+ "torch_compile": {
211
+ "enable": true,
212
+ "scope": "loss",
213
+ "mode": "default",
214
+ "dynamic": false,
215
+ "fullgraph": true,
216
+ "backend": "inductor"
217
+ },
218
+ "mixed_precision": {
219
+ "enable": true
220
+ },
221
+ "wandb": {
222
+ "entity": "mstok",
223
+ "project": "iclr-debug",
224
+ "group": "full-owt-ctx256-16sq-135b-v1",
225
+ "enable": true,
226
+ "id": "iclr-debug-ablation-no-ema-13p5b-20260919-decoder-mstok",
227
+ "resume": "allow",
228
+ "gradients_and_params": {
229
+ "enable": false,
230
+ "log": "all",
231
+ "log_freq": 1000
232
+ }
233
+ },
234
+ "export": {
235
+ "vqvae_config_template": "/home/ubuntu/mstok-no-ema/config/repro-ctx256/vqvae.yaml",
236
+ "ncp_config_template": "/home/ubuntu/mstok-no-ema/config/repro-ctx256/ncp-sharedhead.yaml"
237
+ },
238
+ "semantic": {
239
+ "enabled": true,
240
+ "mode": "eostok",
241
+ "weight": 0.5,
242
+ "warmup_steps": 500,
243
+ "teacher_id": "FacebookAI/roberta-base",
244
+ "teacher_revision": "e2da8e2f811d1448a5b465c236feacd80ffbac7b",
245
+ "tokenizer_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e",
246
+ "teacher_dim": 768,
247
+ "feature_layer": 6,
248
+ "projector_dim": 2048,
249
+ "projector_seed": 56,
250
+ "alignment_site": "decoder"
251
+ },
252
+ "iclr_debug": {
253
+ "stage": "decoder-mstok",
254
+ "target_positions": 135000000000,
255
+ "lr_horizon_positions": 135000000000,
256
+ "stop_step": 17167,
257
+ "pilot": false,
258
+ "tokenizer_checkpoint": null,
259
+ "tokenizer_steps": null,
260
+ "budget_unit": "input positions including padding; log non-padding tokens separately",
261
+ "ablation": "no-ema"
262
+ }
263
+ }
264
+ },
265
+ "sources": {
266
+ "config/repro-ctx256/alignment-generator.yaml": "f62f36d078a3a373a19ea716a3ebc7eda5ae2fb7d670c5b935c734826bb76895",
267
+ "config/repro-ctx256/alignment-joint.yaml": "bb511950d20335e4237e406464ab26760a7fd9c9c9472c0507addd88d33d5178",
268
+ "config/repro-ctx256/alignment-tokenizer.yaml": "2d32955296f7d15990f284328c1cf8f52071abdba28033a65e272bb49cf5dd56",
269
+ "config/repro-ctx256/alignment.yaml": "3a486fa2ed28eca4fc406232669399b0625d274376ac43271f4ccf187b1e9077",
270
+ "config/repro-ctx256/joint.yaml": "8b4889111687a52e2d7d71995bda45a7f01f2777674c88191f338dca4c1732d2",
271
+ "config/repro-ctx256/mstok-semantic-eostok-matched10ep.yaml": "ae8a72c0c9f1cb5d2e301d83a1cdc08985a1f895bf627a5f75b1b3ce627f5e58",
272
+ "config/repro-ctx256/mstok-semantic-eostok.yaml": "2c9c48b21325c0e479990701b28116d78f611101724846747d173e52145fe15a",
273
+ "config/repro-ctx256/mstok-semantic-gear-matched10ep.yaml": "164cb8ac39fa407bb337f559ad8a9c62a34708fb900b8829d3ccf182961f2d0c",
274
+ "config/repro-ctx256/mstok-semantic-gear.yaml": "8cb8c998551daef0bf7fd9ec77b13b34c06662028d8820767e1432b5a9e8be54",
275
+ "config/repro-ctx256/mstok-semantic.yaml": "e0b00dbfa3eb838103f657c19bde20e36375c7fbe692623452ea093003f1ac98",
276
+ "config/repro-ctx256/mstok-w1-pilot.yaml": "21ee22eeccaaea2563c92ccc84f6df3871ea5da6a5ce980656088ea577dc1438",
277
+ "config/repro-ctx256/mstok-w1.yaml": "b6bf6079114b3c997fab89e85a53318bb2ec5dae55678513599d84152eab701b",
278
+ "config/repro-ctx256/mstok.yaml": "d9bb216f4eeb58857d72be399d5f8847d197ce70d0e504879f0788045483dc87",
279
+ "config/repro-ctx256/ncp-sharedhead.yaml": "86c9efbcb2b4f74b7253aa3bc4393de4a6a05e20ea30da573d1ca094ec600645",
280
+ "config/repro-ctx256/ncp.yaml": "93663654c4ce5ad0931db052edf507fdcb7c0ff9fc9722f4018820deee5c5acf",
281
+ "config/repro-ctx256/substitution-generator.yaml": "064f55887b9b2b8f3b0b52c094d4a193510dc1148ad7a5c7f53e317971f87c7c",
282
+ "config/repro-ctx256/substitution-joint.yaml": "b1576f4ecfe6fc6029a0cd868271b80a1d988b7f3667c6b1a8d24a6a4015df3b",
283
+ "config/repro-ctx256/substitution-tokenizer.yaml": "e29be5046f0fafbfe4c06973be4cd036a004e845348911cc0e6e05a7fca5d425",
284
+ "config/repro-ctx256/substitution.yaml": "ebba1c7e664de0e84913bdfd38e794837781c42c1b48cde90135ff9417856a95",
285
+ "config/repro-ctx256/vqvae.yaml": "05fde46e57c3e6b411e8a6afcac251f178e4b8a0856a0b99332bb284fb4710ce",
286
+ "data/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
287
+ "data/serialized_dataset.py": "5622c2d94236cd1b9b4ca18e8d5c272a86b54d193d99048fb9b7cd2687e8dbc6",
288
+ "data/tinysentences/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
289
+ "data/tinysentences/clean.py": "918a9a91aae484a5e06d6d212c2c0974946a4032e061c62a7b65851735bfc8dd",
290
+ "data/tinysentences/clean_parsed.py": "009a087f967e171bc6098484dbfd58635302900f6b07629629898f1d00fa70a3",
291
+ "data/tinysentences/generate_meta.py": "662714f1370e05567d412add168a7c3a05f4b1c9adb047fad374900bc1936a7e",
292
+ "data/tinysentences/generate_metadata_char.py": "afd3336ea03ff0653ccaf41c121d9b93debf7a380db6cd881b62b308ee6ee68c",
293
+ "data/tinysentences/generate_training_data_word.py": "82e2edcb7ea4db0609ccf2b047a0b207e7b8fcff4fc0b4ceee521ff278e8a8e8",
294
+ "data/tinysentences/parse_summaries.py": "2317159364dbbe7206f252d2939cab5d375395e0a536be0a1f60c5df88b31702",
295
+ "data/tinysentences/prepare.py": "35263b287f93337fda4207ed1991e9e8a820ab75139411bb7dd5f0a8a25bcbd9",
296
+ "data/tinysentences/sentences_only.py": "5e4d72bf612df27dad46cf606a40402606d897043ae24274dc2dfa88ec0b3e79",
297
+ "data/tinysentences/stats.py": "2665e9e6fac4a8f6c3e5bd8b07cbfd941ba6359a43981d89bdef58c136fdd3a9",
298
+ "data/tinysentences/summarize/merge_summaries_char.py": "bce94f4d9a824c2d25e45448023d59b7fbc42aba22b33efbc067cdbba93e1934",
299
+ "data/tinysentences/summarize/merge_summaries_word.py": "f6079110332cc38b2dd3571c30bedcca97098c39498f45eec1ef6e60cce7ac81",
300
+ "data/tinysentences/summarize/summarize_gemini.py": "282a438efce0dee351fd9833a59c56e4480bed4314dae40844201794ada3bd96",
301
+ "data/tinysentences/summarize/summarize_gpt.py": "fc08cbc37d6161e5b0057a1ede43272d96f7274524917d6c4d42c22f505c6e09",
302
+ "data/tinysentences/summarize/summarize_llama.py": "7f3f8230652a5b88ea2856b273d497cd5b633580b430c909e4f996cd0232832d",
303
+ "data/tinysentences/summarize/summarize_qwen.py": "b07e3c226ab0fdf5b21b025263adfa7c2625d640f0fac3c787bc4a2c9e32107f",
304
+ "data/tinysentences/summarize/summarize_together.py": "8c26ea052f2a64c2653d39f1ea5c13983dfbc9547f21cd7d7ea9d9b39033bffe",
305
+ "data/tinysentences/test.py": "9749244da77c078f550a78751664b45bb92067682f3aa27c2338419f1186c844",
306
+ "data/tinysentences/test_regex.py": "46652e189a0d5f5c0c7a2c90b857082004bf6727ef1d3ad42b631ac7bb0aa88f",
307
+ "data/tinysentences/tinysentences_char_dataset.py": "152ab9c30aec669a9db14df9e90412597870105eff9cfb737b876aad2323a206",
308
+ "data/tinysentences/tinysentences_dataset.py": "f90251d21871f6a7ad5ebd2ad0a6e59cd0b02ca245f7e46f93e6d970bf657bc7",
309
+ "data/tinysentences/tinysentences_vae.py": "30f7adde5413348faf5066448706747fe26a7fedd510ffea720aa0bf8283f6f8",
310
+ "data/tinysentences/to_csv.py": "c06c74648ed95b4494835c631c94fe1a00a9ff47a272f5e38546a7f94ff9be8b",
311
+ "data/tinysentences/tokenize/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
312
+ "data/tinysentences/tokenize/common.py": "e877c74df2a48d15991492b126cd0c5fe26731d61f3aa36d83ebf2cb9bda34a0",
313
+ "data/tinysentences/tokenize/tokenizer.py": "c16abc50b92a163f6978542b36b03c9a14aa00153c73117c4d52084d5de25c2a",
314
+ "data/tinysentences/tokenize/train_bpe.py": "fa8e55978dd5ca90117200648b468fb22822bf6981f15f0394c815e72eae1bdb",
315
+ "data/tinystories/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
316
+ "data/tinystories/download.py": "5afce37e912f6965b6722f0bc028076a655fea78304f1a213e4862dbe223cdcb",
317
+ "data/tinystories/stats.py": "216650fe19b78e8431c4512ff87a5c84af64d9e56ab8fdaf3952ca035725ea1b",
318
+ "data/wikitext/parse_wikitext.py": "ab583e0328f82afc3ba242410e8076be42ef1a16a51d66f3eb46a713cca591d0",
319
+ "evaluation/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
320
+ "evaluation/eval_random_gen_ppl_checkpoint.py": "ab5ffe43a59e8ad6942994250986071327894441312deb8a16e9241dc5d51df9",
321
+ "evaluation/gen_ppl.py": "e6853b14d27f93407505eb7607a1bd9fed49a7add1511ceca2b4a3076d533859",
322
+ "evaluation/gen_ppl_ncp.py": "96d68adee372e5186d0470659171eb03829e3d37f611e8f5917b45fb61e3a4ad",
323
+ "evaluation/gen_ppl_ntp.py": "b2638622bc041fa88e9d538be4229b3723ebec19c5b99fd6e5b73db4776dc952",
324
+ "evaluation/guess.py": "c27803225dd21b2deff1e7e494b9e3666180b93776340164ce6e953dacc50656",
325
+ "evaluation/ppl.py": "f3b581cb4849b0b2aa518729b8338838c7055697c3a26028dedb819c335ac838",
326
+ "evaluation/run_genppl_multiseed.py": "f8c9ac982f4caa4335b5f4d1a22d70f56dc11a706106a3dfb31d344eb8c93c2e",
327
+ "evaluation/run_genppl_release_ts.py": "d1bc5c00089118d9cf6c05c77459329ad40c2f8bc347b5aeedda8f5318de8796",
328
+ "evaluation/run_genppl_sweep.py": "ade4f39e373125848fdd32938e8d7bf86a0ec5108cca9bd75f0a966d092a798e",
329
+ "evaluation/show_8k_samples.py": "30ad825d5a24375e178750dfaaac45bc486bc3587f2e7d6d305d53bf37f8d226",
330
+ "evaluation/show_low_sample.py": "d696f1e54ca67e87565dce5864e0e6e2ad4194108cfb1c9c1e06eebd97008da6",
331
+ "evaluation/summarize_gen_ppl.py": "307f3d0d85884879295d5268aaf7f399f654b1b330aaa5728d6824ad24c6d15a",
332
+ "evaluation/sweep_gen_ppl_checkpoints.py": "3a9141e27d84b49c5493194dfc5a70bfa0a915cdf5ceb7a46afcf27cc6f33acb",
333
+ "evaluation/sweep_gen_ppl_ncp.py": "d91c2959cfed51d224ccb1c0d0f2c3b9a35c4027cc9ab8c65aa7603d587de761",
334
+ "evaluation/sweep_gen_ppl_ncp_scaling.py": "05249c1e00c5eb05042b6ea5b1ca008ecb8f0e4a77537ee755a27c153c25a17c",
335
+ "evaluation/sweep_gen_ppl_ncp_yolo.py": "4b26ef25b27af2f7fcbcb5ff3c312d47d2a4746ba759155d28802e978a681755",
336
+ "evaluation/sweep_gen_ppl_ntp.py": "14a04074741ee44580ec6557870d42534dd172d26c5be3e943ed5baf1b727303",
337
+ "evaluation/sweep_random_llama_yolo.py": "4062b54289fcd9407d15feb1ffd241b8722fd717c700350c22472fecec5ae46c",
338
+ "evaluation/val_oracle_ppl.py": "1a23f3c2897b6a6a3d239ed533979de35352d303bae094f4006bef4156f935c0",
339
+ "evaluation/val_vqvae_ppl.py": "4c293c502d3faf982d4ad66566b7a037faa7d28191d929de1621cb67f5ae8b5f",
340
+ "models/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
341
+ "models/common.py": "99aa0da356963a282936413d738ee1da72a3e3f277fecc28eddb553051582a4b",
342
+ "models/masked_lm.py": "80279614d74efc7b4c25306789d74e2f71045c54e4e4f485ee365f664705d7f6",
343
+ "models/next_concept.py": "743667aac3bdceb7bb7476a580700185b2638ddd9575e6357db57acc7d699339",
344
+ "models/next_token.py": "d0d292b9f5c032a351cbca1a8f0a6d41c3062890cb3e3f5b67d4e23a7cffeb77",
345
+ "models/quant.py": "b7b73abd4e55e10454cb680dea82124c38c222d37236c4c76b5c7d21123ac1b1",
346
+ "models/semantic_input.py": "b06658d4bc641397cbb973255eadfd2d6087fde244d801d3c70d29e3ce841bf2",
347
+ "models/vae.py": "02e2213a8878920374baf27dd10dfc03847dce866557c97e816c427d230c9ded",
348
+ "models/vqvae.py": "8eb3503eb1d990121b7d6138a4f8216aecfc72246d4a92bbecf98025c8a3d504",
349
+ "scripts/benchmark_iclr_debug.py": "1fb00ab1ec4e615c52b5f8415c22600a16c5db9d790ba8eaba3dae2206e08491",
350
+ "scripts/benchmark_mstok_compile.py": "aad909da6f05ba25ac3d8f6b3cf450103bed55d0b93b1490764c0ef54f35b606",
351
+ "scripts/eval_owt_ckpts_gpt2_large.py": "4977a2d614698f60032fb6fd7fef614eb503d5b3883f499146a4fc048214a509",
352
+ "scripts/evaluate_mstok_sidecar.py": "9ba33e865fa0fad34bd2ab375d7fe38f37c441e2d96348f0e389e29dd9643455",
353
+ "scripts/evaluations/sweep_gen_ppl_shared_vocab_ncp.py": "2311ac71f9beba282426faa21390a5898c7d183a95f529755d6e1d6e8fc24050",
354
+ "scripts/neurips/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
355
+ "scripts/neurips/eval_gen_ppl_ncp_owt.py": "b92eff1ce198eb4ac8c88da7113b41d095c12acad92db541685175993db3535c",
356
+ "scripts/neurips/sweep_gen_ppl_owt_gpt2.py": "02adb1629cfa40ff51a75152e4da9c9ee00ead42b7ecda9e78cf7101e6826a67",
357
+ "scripts/preflight_owtsmall_mstok_ctx256.py": "a4f80921435b6cd136d089ec39bfa5442359d0dffbeefaaa1019178b8364657a",
358
+ "scripts/prepare_full_owt.py": "b1cb6b74397625c2b0643c4ee744ca3c89dfa6bbb7eeb7e9681e3a36f91a2c24",
359
+ "scripts/r1_lab/diagnose_at_40k.py": "b1582dcc29ccebc9255eaa77a7d56a7dbdef589ec5d5a9b0af150dede98874ab",
360
+ "scripts/r1_lab/inspect_summaries.py": "4154e8a6c1b3f68df5c623a57a1a7632b594c46c5f6d1d85cb08c476d6aa9cea",
361
+ "scripts/r1_lab/overnight_pipeline.py": "53f4e03ce1a5fb43fbca3ae455288f45ae8350b0a16f3d5318763011de6a8f96",
362
+ "scripts/r1_lab/prime_rope_qknorm_score_cache.py": "a791cdec03b415bfccdfd8d89c7e659aff8a2a5696d16ab7b4350ff608d0391b",
363
+ "scripts/r1_lab/sweep_gen_ppl_r1lab.py": "c7c0acf9aa6f07bb0a985cf0d5e5692c696e658ebb0470b7c92d9332565f9c06",
364
+ "scripts/r1_lab/sweep_gen_ppl_r1lab_v15.py": "524aa8680aa1b9fea2689509848c19e5a74a3ad72e78b0388bbca7acd43be22e",
365
+ "scripts/r1_lab/sweep_gen_ppl_rope_qknorm.py": "7822df057355b0a3a71ea5145742107c6e30f906859096237ff61d765fe36d50",
366
+ "scripts/r1_lab/train_variant.py": "9d0a2091784609ac9728226afd82f56fd4c1ac2dc08e89e1b17b60d34bb3d55e",
367
+ "scripts/run_alignment.py": "6312cd9d0976899763589fdcbd258c7681f6e9e96d7396cf35ff022d4695b862",
368
+ "scripts/run_iclr_debug.py": "26514d8e40a790d3c3336c951f8626e1091fb02cf305844d82a94bf368415c04",
369
+ "scripts/run_mstok_semantic.py": "8a7a38a50dd6694950bfa0e7b79718ec653c7e70be99fe13cbf2ee05c28f46a0",
370
+ "scripts/run_mstok_w1_pilot.py": "35224e462d0d11eb3e4326b0daa8bad7c99e1e30cdc6383310f8e2c8b5bb708f",
371
+ "scripts/run_substitution.py": "2cd22eaa58dab8b5231984eee340a3a2bdd0b88cc83333f7133447471e7f73f1",
372
+ "scripts/smoke_test_masked_l0.py": "d24f1270b3219a31212b68bcbb786f649c7c72f96e53f602d87ad6c17f3f73e5",
373
+ "scripts/test0_diagnostics.py": "ed389c666beef015ccac004f314a5a1428def303b46ac0e442f0236337344549",
374
+ "scripts/test0_r1_similarity.py": "4a486878bebd418a3b5454ddbf02585726ccbacc2700fdf591aa524766f9c9c9",
375
+ "scripts/test0_r1_uniqueness.py": "3ee59d4397d5b268151dbb8ca779c7f9bf8461d129d97fcd1a517c48ec2a69a1",
376
+ "scripts/test0_residual_norms.py": "3c4f68e5960e84796dc0bca4fd4583bf3b598552654f9b565fd5fcabd420f9e1",
377
+ "scripts/test_gen_equiv.py": "ae0e9c513422ee58bdc050ed6218ee54fc555dfb15ea8968c9ffc62a7a7ea99a",
378
+ "scripts/training_budget.py": "bb94c19c6dfbdd542946b10476b8d6958b8a189abd141cf62288fdb5fa9a2d38",
379
+ "scripts/vqvae_utilization.py": "5f90327f1507a5042b2f77b119fe4fc0df4e2131e5bbc21069dffd1fadc059c5",
380
+ "scripts/xsum/codec_ceiling.py": "ddb751e9152d5c88966e588d89cdceee4c5d91c4139aaedf369f65a45ecf2276",
381
+ "scripts/xsum/eval_merge.py": "6e313d6ed25227d0418b7422d58e09ceda4fc6dfae7d0fd663417b306241135c",
382
+ "scripts/xsum/eval_q0.py": "82fe04dccce28d7fc9c6ec8b4c9c91404a49527f9437913f44f51a6853ba9abc",
383
+ "scripts/xsum/eval_shard.py": "07192f6ba6219da1e8123ceeb99748c86da85322859ed5fa25cebcc3816f7ea8",
384
+ "scripts/xsum/eval_shard_cfg.py": "95ac4a8e2e3a73a21fa4303c62df80c086f23bcfe5ab34eb07ae90b585052ecb",
385
+ "scripts/xsum/eval_test.py": "7f92d578c31893e788e5fa7d9eef3098ba5602b3642873ce2ce031342c8be868",
386
+ "scripts/xsum/eval_uncond_rep.py": "ba4740b29bc091d866e2dfd91881e9a7d5c46a1bfa993d13eb63c0467397ec28",
387
+ "scripts/xsum/make_warmstart.py": "78793d6ef11fbf658c2837ef12cb843fc4817add8ba73c822b5e408901de3318",
388
+ "scripts/xsum/mbr_fast.py": "3db6e21d2dcce2e3bcf6f22474535d7eba0232b071b4271d2b172f787aed5c85",
389
+ "scripts/xsum/mbr_select.py": "107a061014c62395b7f5fd986fba9e55dd738054e2474022005cb7e9e0c9b8fb",
390
+ "scripts/xsum/mbr_shard_cfg.py": "3e7162375028c391276149eff288e9de40c874a3108ac53de0248198f4d08791",
391
+ "scripts/xsum/pool_append.py": "e50252f48c4d8d905be0de3eb56ca1d720f8f8545c6fdcb58ed27f0360c4e2ad",
392
+ "scripts/xsum/prep_xsum_le48.py": "3688ed1abc49d18f75d34f8382ff68a5afbd753b36cefdc0652d0bffb6a5f347",
393
+ "scripts/xsum/prep_xsum_mix30.py": "4bd47e52637c4b6bca5fd4956c1787b506a7f51b8a5bb0b68f19f728c4d48afb",
394
+ "scripts/xsum/prep_xsum_mix_split.py": "ec2fe74f08068ae8fda4ed59a86537a1d3f48276d205099bbfe16f8428ca36d5",
395
+ "scripts/xsum/prep_xsum_pass_uncond.py": "82ea50248d811e5182e63d0b4dfd6f995dd2f47eb963b5bb1c5d09bc533db7c1",
396
+ "scripts/xsum/prep_xsum_test_all.py": "7389fbdf5612ba524801c5b04109f810b139152ca50101b60acbb9b1e767b256",
397
+ "scripts/xsum/score_checkpoints.py": "41b412c06cd92cbd4caa6b586da7ddc7e913607a47c1a3aac0ee0a36a20a36be",
398
+ "tokenization/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
399
+ "tokenization/bpe_tokenizer.py": "0c8af84868a6aa40f981768f8ab49d28e501d55cce444edb1ad315de53c3445a",
400
+ "tokenization/build_word_vocab.py": "778f9666f4e7c642dc471773ee81c92c26975db37de1ebfffb605cae211351b5",
401
+ "tokenization/byte_tokenizer.py": "eb567a1772ae69738884ed9b94330aa39387c49e54d76e4b9752e67674fad810",
402
+ "tokenization/compression_ratio.py": "ed07ab1cc38d6176362631ce655bc974dd3623b87693f51279d5047e6bc6890f",
403
+ "tokenization/hf_tokenizer.py": "2ca8fae097f9c640bcb52b4788c37de8e0d9bb2fa971e30a5abb39f90ff97d4c",
404
+ "tokenization/tokenize.py": "f8ab54338613b6bd84513fdda85be173fd32274c35d97ef39a5c17d584f8108f",
405
+ "tokenization/tokenize_doc_level.py": "a06c81cf088fb81d2317ed1c3a1417a0e48ae0ea04962824f13f5d3beddc8902",
406
+ "tokenization/tokenize_vqvae.py": "14730eec3da87f6528e5fbd1166d5da79e10ac1700f7cd0be73bc40716bdebfc",
407
+ "tokenization/tokenizer.py": "e527d47e4ed8634c673d48a872a07c9cc8828d30988331de2d4bd3b7c21ddcb7",
408
+ "tokenization/train_bpe.py": "bf4e2a29eeba874024ed95f32fe720ae2320f72cb978361692db73e71be98683",
409
+ "tokenization/vqvae_tokenizer.py": "d7fa9d595865aa3e0b0d185bfd04721eb297fe5f657c4fca4ce69fc96aca308f",
410
+ "tokenization/word_tokenizer.py": "a405621bf35ed6cbb29abadd9d1d651e0783e6801fc7645b35e799d5de381ac2",
411
+ "train_mstok.py": "077b7c99c12daad2bbd986918d0f487411779de05a472af71124098da399b79d",
412
+ "trainer/alignment_trainer.py": "54f8ee7cca4d407223f2ce5146e6fc561ad748e20f2c702beb12ae040cd290c4",
413
+ "trainer/corruption.py": "f0f24c94652e6e9cf8e4ef30ae1a302e3232d64e765c931696d454db96e8b57d",
414
+ "trainer/joint_trainer.py": "8c8ebb55d21b6e223be1f7595c51f1b9257ccac40f8cff28260c1088656e2e49",
415
+ "trainer/mstok_trainer.py": "90712e396472d5dac9679b62af4f0b62afa6390a839d4d9bd9dae57406fe76a4",
416
+ "trainer/ncp_trainer.py": "73fc23f78f403dc39800b924dfb18ad42ffeab847fed30e931183c3d3acaee68",
417
+ "trainer/ntp_trainer.py": "5076d7aa29fac70869e1e309a15af6c4d3a91d6de8e00ac113a586841a9866e1",
418
+ "trainer/onpolicy.py": "6ff1fe4103d71001afe2d73e574d20c7242381e420929db14020141464ec6821",
419
+ "trainer/plain_two_stage.py": "4c5973ed87100b1b478aa1df0bb16cd3d1eb5844dffe5894003dd67c463f6c9a",
420
+ "trainer/scheduled_sampling.py": "f53515e23c2cec1231265f21a341ff77d11fa8a6129e0e382d540cc85571f5ed",
421
+ "trainer/semantic_mstok_trainer.py": "ebc02e80d279a445c99d8f09505cdfaf7632f1a661d6880f09cc1bd0fa90ef30",
422
+ "trainer/semantic_teacher.py": "7983b5da86d25954ed8267eee03d144c421305967705e0a242f4f025794ceeae",
423
+ "trainer/substitution_trainer.py": "33c7833b2e8dfabbb9ee56b258708055d3194b8d7453a1b8d93f59b44ff91465",
424
+ "utils/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
425
+ "utils/arg_utils.py": "e9eb7971f6273e9630b56fd87a03725b29183a8c80abb014c4c84f8b1bdb6445",
426
+ "utils/benchmark.py": "ccf436ef72a4e626c7f671d8e63a683057bf502a3852e5a7142410d1dc3e5501",
427
+ "utils/data.py": "78cce11e777813755cd171ccaa3247f89ab99ac4079e0157e55c72ce8a5a7fe5",
428
+ "utils/dist.py": "b2849d3ce03a8b1737e3dd61216e76e4a9e82d63ad1ea74cddf72e2285eaca64",
429
+ "utils/experiment_config.py": "616a70fefaac99ab0d856a1b6c36787a28fb9af8a4c55fc6d4765012fbaaeb38",
430
+ "utils/iclr_ablations.py": "be5f26dab47d57a07e22519c64251d4d0da6d19437984d726e85026acecd3a45",
431
+ "utils/iclr_training.py": "b1bc115f037932eea120a652fb609c3d22ca54a1192b527da59540218ac96bf9",
432
+ "utils/logging.py": "901e67515cc65a35f575f05d16a3b9ef4a5b2cb0ea252cd7abbc80ecce9e88e8",
433
+ "utils/lr.py": "4d1ea2c25340afc4bdde8b582dab36f174e0b2965fa7bafa2a7539019dbe2b7e",
434
+ "utils/misc.py": "be4a323116a0956a77960d1baf613f268057d6a36023663d2af9f8e4728cbe77",
435
+ "utils/plot_levels.py": "916b39885ced191960c26c334a3b16f9c99751f6c29c46bd69d14fee414965f3",
436
+ "utils/registry.py": "f7a44c6d18433100873d5df462b9f08a8bdeb11cf3d92226334c7e5ece318318",
437
+ "utils/train.py": "43eef988000f3a5e20856653d725337a0fe0dbe276f0cf80e1b200292abe812d",
438
+ "train_iclr_debug.py": "94844fceb0f0d1e0e48fa27384887ef727f11ce820ebdc4805e58982434e0d35",
439
+ "train_joint.py": "21b7c4f179632e9c7e319178e4d661f01a3702215dcbe1c8aeae90e430ebb4bf",
440
+ "config/iclr-debug/README.md": "1c757d2ff37026d08c22324ec44cfa760b0599af62c8ad9b728ada471a8f2905",
441
+ "config/iclr-debug/benchmark-16sq.yaml": "18c9cef7032f9d3806d8b0c8b20fe0fc7d1ea98ad1d2440e24a67ee873c97f9b",
442
+ "config/iclr-debug/VANILLA.md": "29f86c602b4aaf5caf42e2397346c19c0e99c5f29911be117c59eb134f99114b",
443
+ "config/iclr-debug/MEDIUM.md": "16cdf56a3fb9f4b29ff1c05b49976f941cd630527094d363906e3e55b815d1c9",
444
+ "config/iclr-debug/ABLATIONS.md": "7c9fa0af54fa83d68b1ecec8a167ff1c61c2ca46ae3112489f4fbe89f1d8b314",
445
+ "config/iclr-debug/full-owt-audit.json": "fefa394834f6e9f65944d3b514633c29061071ebe6ca60fce21a6edef9748c1e",
446
+ "config/iclr-debug/training-defaults.yaml": "c9c712942ab33cfdc8787363219424dd7d7f6505d5c6a3795b0ab0295b7a508a",
447
+ "config/iclr-debug/NCM-BUDGET-AUDIT.md": "e734caeab7edb612de7ff508a8c0ec52445530ab7a35895b5b793d6a7b752e4b",
448
+ "config/iclr-debug/TRAINING.md": "29af073e81bf64983148bf760d329999eb736908c816070ea04b69b73a53ab9e",
449
+ "config/iclr-debug/medium.yaml": "1157a2c8ab87ae107ba634f3a11f186dc7f8eef02ff26a0a317be0e1fe6d71c6",
450
+ "config/iclr-debug/BENCHMARK.md": "2781ea506212e12a0640a4b0d2a2c881355ec85756ca4014bccb1b1aedccdf59",
451
+ "config/iclr-debug/HF-MODEL-CARD.md": "de1fe176ebde262eacdf32af7d3f8c6497a764dd98e5a9094dad26de4aade38e",
452
+ "config/iclr-debug/ablations/no-recon.yaml": "c5df2ec5fbfe3736bf33fa26f8f8f03ac2aa4e3e920c64513f8846d913add1ea",
453
+ "config/iclr-debug/ablations/no-ema.yaml": "7916aa8b4c11a2cc23d8400c3c5f641ac4da3ac86ead5116203bca70d9d4b9fa",
454
+ "config/iclr-debug/ablations/no-semantic.yaml": "fea52b438d97fc1a0d9b0f8e488c35600aad68be14087f710eb6d213c2d45478"
455
+ },
456
+ "dataset": {
457
+ "dataset": "hazyresearch/ncm-tokenized-datasets",
458
+ "revision": "951e04517fd53d57521d8e09abd6511a662787d0",
459
+ "dtype": "little-endian uint16",
460
+ "eos_id": 50256,
461
+ "offset_units": "token positions; [0, each EOS position + 1] with final sentinel",
462
+ "sampling": "length-weighted within-segment windows; pad short segments with 50257",
463
+ "caveat": "EOS segments are not independently verified source-document identities",
464
+ "splits": {
465
+ "train": {
466
+ "bytes": 18070847356,
467
+ "tokens": 9035423678,
468
+ "sha256": "b24b78a07f56974fd0c9855dc16f25b836d2cb172be23eb5d3c650cd6d73f56a",
469
+ "min_token_id": 0,
470
+ "max_token_id": 50256,
471
+ "segments": 8009770,
472
+ "trailing_tokens": 0,
473
+ "eos_only_segments": 0,
474
+ "length_percentiles": {
475
+ "min": 132.0,
476
+ "p25": 416.0,
477
+ "p50": 718.0,
478
+ "p75": 1249.0,
479
+ "p99": 7610.0,
480
+ "max": 131288.0
481
+ },
482
+ "segments_shorter_than_256": 673011,
483
+ "expected_padding_fraction_at_256": 1.5955357567300495e-05,
484
+ "offsets_status": "created",
485
+ "offsets_sha256": "7e84564946b8cf8190dbdc1e86bdf2fe8cfd9235a517328c80b4df038ea0028b",
486
+ "offset_entries": 8009771
487
+ },
488
+ "valid": {
489
+ "bytes": 9186834,
490
+ "tokens": 4593417,
491
+ "sha256": "1f2ee17e08327a39d4e2417948d02c7b998d626d2dcb2661197a6e208f58266f",
492
+ "min_token_id": 0,
493
+ "max_token_id": 50256,
494
+ "segments": 3999,
495
+ "trailing_tokens": 0,
496
+ "eos_only_segments": 0,
497
+ "length_percentiles": {
498
+ "min": 143.0,
499
+ "p25": 433.0,
500
+ "p50": 734.0,
501
+ "p75": 1292.5,
502
+ "p99": 7122.719999999997,
503
+ "max": 24868.0
504
+ },
505
+ "segments_shorter_than_256": 363,
506
+ "expected_padding_fraction_at_256": 1.6064002171981108e-05,
507
+ "offsets_status": "created",
508
+ "offsets_sha256": "09a748ce1dcec7de36c77c48e39ccbf5c0293cc073689fa7c5f48bdfb9bb9445",
509
+ "offset_entries": 4000
510
+ }
511
+ },
512
+ "overlap": {
513
+ "exact_train_segment_matches": 0,
514
+ "unique_validation_segments_in_train": 0,
515
+ "unique_validation_segments": 3999,
516
+ "near_duplicates_checked": false
517
+ }
518
+ },
519
+ "hf_repo": null,
520
+ "generation_evaluation": true,
521
+ "versions": {
522
+ "torch": "2.13.0+cu130",
523
+ "transformers": "4.44.2",
524
+ "numpy": "1.26.4",
525
+ "hydra-core": "1.3.6"
526
+ },
527
+ "hf_run_prefix": "ablations/iclr-debug-ablation-no-ema-13p5b-20260919"
528
+ }