rubentium commited on
Commit
6c25804
·
verified ·
1 Parent(s): e839774

EPFL final checkpoints and W&B history 951-1000 of 1463

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +7 -0
  2. epfl_final/SparseAtt/wandb/run-20260822_133517-81uee6bn/run-81uee6bn.wandb +3 -0
  3. epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/files/output.log +0 -0
  4. epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/files/requirements.txt +142 -0
  5. epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/files/wandb-metadata.json +65 -0
  6. epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/logs/debug-internal.log +0 -0
  7. epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/logs/debug.log +19 -0
  8. epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/run-9nppzrru.wandb +3 -0
  9. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/config.yaml +190 -0
  10. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/output.log +0 -0
  11. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/requirements.txt +142 -0
  12. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/wandb-metadata.json +51 -0
  13. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/wandb-summary.json +1 -0
  14. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/logs/debug-internal.log +0 -0
  15. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/logs/debug.log +25 -0
  16. epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/run-8xh3oikj.wandb +3 -0
  17. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/config.yaml +200 -0
  18. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/output.log +0 -0
  19. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/requirements.txt +142 -0
  20. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/wandb-metadata.json +65 -0
  21. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/wandb-summary.json +1 -0
  22. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/logs/debug-internal.log +0 -0
  23. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/logs/debug.log +25 -0
  24. epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/run-9nppzrru.wandb +3 -0
  25. epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/files/output.log +350 -0
  26. epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/files/requirements.txt +142 -0
  27. epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/files/wandb-metadata.json +51 -0
  28. epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/logs/debug-internal.log +0 -0
  29. epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/logs/debug.log +19 -0
  30. epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/run-i8ewyy95.wandb +3 -0
  31. epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/files/output.log +2 -0
  32. epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/files/requirements.txt +142 -0
  33. epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/files/wandb-metadata.json +65 -0
  34. epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/logs/debug-internal.log +13 -0
  35. epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/logs/debug.log +19 -0
  36. epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/run-y8clx6eu.wandb +0 -0
  37. epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/files/output.log +0 -0
  38. epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/files/requirements.txt +142 -0
  39. epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/files/wandb-metadata.json +79 -0
  40. epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/logs/debug-internal.log +0 -0
  41. epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/logs/debug.log +20 -0
  42. epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/run-i8ewyy95.wandb +3 -0
  43. epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/files/output.log +526 -0
  44. epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/files/requirements.txt +142 -0
  45. epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/files/wandb-metadata.json +72 -0
  46. epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/logs/debug-internal.log +0 -0
  47. epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/logs/debug.log +19 -0
  48. epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/run-rgmkfpnl.wandb +3 -0
  49. epfl_final/SparseAtt/wandb/run-20260823_140341-qctui00g/files/wandb-metadata.json +79 -0
  50. epfl_final/SparseAtt/wandb/run-20260823_140341-qctui00g/logs/debug-internal.log +0 -0
.gitattributes CHANGED
@@ -124,3 +124,10 @@ epfl_final/SparseAtt/wandb/run-20260822_043611-xvizh3ih/run-xvizh3ih.wandb filte
124
  epfl_final/SparseAtt/wandb/run-20260822_043614-pq6ov2y3/run-pq6ov2y3.wandb filter=lfs diff=lfs merge=lfs -text
125
  epfl_final/SparseAtt/wandb/run-20260822_043615-8naq20vs/run-8naq20vs.wandb filter=lfs diff=lfs merge=lfs -text
126
  epfl_final/SparseAtt/wandb/run-20260822_115228-8xh3oikj/run-8xh3oikj.wandb filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
124
  epfl_final/SparseAtt/wandb/run-20260822_043614-pq6ov2y3/run-pq6ov2y3.wandb filter=lfs diff=lfs merge=lfs -text
125
  epfl_final/SparseAtt/wandb/run-20260822_043615-8naq20vs/run-8naq20vs.wandb filter=lfs diff=lfs merge=lfs -text
126
  epfl_final/SparseAtt/wandb/run-20260822_115228-8xh3oikj/run-8xh3oikj.wandb filter=lfs diff=lfs merge=lfs -text
127
+ epfl_final/SparseAtt/wandb/run-20260822_133517-81uee6bn/run-81uee6bn.wandb filter=lfs diff=lfs merge=lfs -text
128
+ epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/run-9nppzrru.wandb filter=lfs diff=lfs merge=lfs -text
129
+ epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/run-8xh3oikj.wandb filter=lfs diff=lfs merge=lfs -text
130
+ epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/run-9nppzrru.wandb filter=lfs diff=lfs merge=lfs -text
131
+ epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/run-i8ewyy95.wandb filter=lfs diff=lfs merge=lfs -text
132
+ epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/run-i8ewyy95.wandb filter=lfs diff=lfs merge=lfs -text
133
+ epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/run-rgmkfpnl.wandb filter=lfs diff=lfs merge=lfs -text
epfl_final/SparseAtt/wandb/run-20260822_133517-81uee6bn/run-81uee6bn.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f886d9fee718c3a5206f5e63e80163d3692fc8380c825938e3ce3b290836524
3
+ size 10023668
epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/files/output.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/files/wandb-metadata.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-22T14:34:43.097867Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p1_scheduled_h1004_20260822/l8k365m_resampled_nrb35_seed45.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "h100-workspace-4-gpu-3-0-8",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 96,
21
+ "cpu_count_logical": 192,
22
+ "gpu": "NVIDIA H100 80GB HBM3",
23
+ "gpu_count": 4,
24
+ "disk": {
25
+ "/": {
26
+ "total": "9371709865984",
27
+ "used": "6533931311104"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1623005732864"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA H100 80GB HBM3",
36
+ "memoryTotal": "85520809984",
37
+ "cudaCores": 16896,
38
+ "architecture": "Hopper",
39
+ "uuid": "GPU-37a1c6c2-d3df-3064-44df-33fa82304bd0"
40
+ },
41
+ {
42
+ "name": "NVIDIA H100 80GB HBM3",
43
+ "memoryTotal": "85520809984",
44
+ "cudaCores": 16896,
45
+ "architecture": "Hopper",
46
+ "uuid": "GPU-9a903f43-9348-6fa6-bcf7-e12bd01803a7"
47
+ },
48
+ {
49
+ "name": "NVIDIA H100 80GB HBM3",
50
+ "memoryTotal": "85520809984",
51
+ "cudaCores": 16896,
52
+ "architecture": "Hopper",
53
+ "uuid": "GPU-24b8d952-dc08-33b9-d5df-54ce871fd288"
54
+ },
55
+ {
56
+ "name": "NVIDIA H100 80GB HBM3",
57
+ "memoryTotal": "85520809984",
58
+ "cudaCores": 16896,
59
+ "architecture": "Hopper",
60
+ "uuid": "GPU-b87e8111-63b6-b364-28b9-c4ce747aec08"
61
+ }
62
+ ],
63
+ "cudaVersion": "13.0",
64
+ "writerId": "cbaqtxrs3at22jmk9fd4mv41hetb6rl6"
65
+ }
epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-22 14:34:43,154 INFO MainThread:11913 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-22 14:34:43,154 INFO MainThread:11913 [wandb_setup.py:_flush():81] Configure stats pid to 11913
3
+ 2026-08-22 14:34:43,155 INFO MainThread:11913 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-22 14:34:43,155 INFO MainThread:11913 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_143443-9nppzrru/logs/debug.log
5
+ 2026-08-22 14:34:43,155 INFO MainThread:11913 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_143443-9nppzrru/logs/debug-internal.log
6
+ 2026-08-22 14:34:43,155 INFO MainThread:11913 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-22 14:34:43,155 INFO MainThread:11913 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed45/run_20260822_143256', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 1, 'gradient_accumulation_steps': 5, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 45, 'verbose': True, 'checkpoint_activations': False, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': 'P1 DDP4 365m resampled 8k nrb35 seed45', 'wandb_tags': ['p1', 'structured-control', 'qwen3', '365m', '8k', 'resampled', 'nrb35', 'seed45', 'ddp4', 'h1004', 'scheduled'], 'hidden_size': 768, 'intermediate_size': 3072, 'num_attention_heads': 6, 'num_key_value_heads': 2, 'num_hidden_layers': 28, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 35, 'sparse_mask_pool_size': 10000, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': '/mloscratch/homes/navasard/sparse/mask_cache_8k_resampled_nrb35_s0', 'sparse_mask_seed': 0, 'sparse_mixed_nrb_set': None, 'dense_mode': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-22 14:34:43,155 INFO MainThread:11913 [wandb_init.py:init():898] starting backend
10
+ 2026-08-22 14:34:43,466 INFO MainThread:11913 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-22 14:34:43,899 INFO MainThread:11913 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-22 14:34:43,900 INFO MainThread:11913 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-22 14:34:43,919 INFO MainThread:11913 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-22 14:34:44,299 INFO MainThread:11913 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-08-22 14:34:44,575 INFO MainThread:11913 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-08-22 14:34:44,575 INFO MainThread:11913 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-08-22 14:34:44,575 INFO MainThread:11913 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-08-22 14:34:44,575 INFO MainThread:11913 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-08-22 14:34:44,581 INFO MainThread:11913 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/run-20260822_143443-9nppzrru/run-9nppzrru.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b2f3ac6042b7f5e466166826d75c402798be8c6bb467d4cd8fe6280b2d6c6f24
3
+ size 3047424
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/config.yaml ADDED
@@ -0,0 +1,190 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.26.1
4
+ e:
5
+ 0vqc90e8r076jfrxgonbbfpa2krs4vxw:
6
+ args:
7
+ - --config
8
+ - configs/p1_scheduled_h1002_20260822/l8k365m_resampled_nrb35_seed46.json
9
+ codePath: train.py
10
+ codePathLocal: train.py
11
+ cpu_count: 96
12
+ cpu_count_logical: 192
13
+ cudaVersion: "13.0"
14
+ disk:
15
+ /:
16
+ total: "9374931091456"
17
+ used: "6412527095808"
18
+ email: ruben.navasardyan@mail.utoronto.ca
19
+ executable: /mloscratch/homes/navasard/conda/bin/python
20
+ git:
21
+ commit: 5637886f0cc760009989b14e68c6ae06d4827418
22
+ remote: https://github.com/rubentium/TokenPrune.git
23
+ gpu: NVIDIA H100 80GB HBM3
24
+ gpu_count: 2
25
+ gpu_nvidia:
26
+ - architecture: Hopper
27
+ cudaCores: 16896
28
+ memoryTotal: "85520809984"
29
+ name: NVIDIA H100 80GB HBM3
30
+ uuid: GPU-af79466d-4300-4171-b5bd-7697eca27264
31
+ - architecture: Hopper
32
+ cudaCores: 16896
33
+ memoryTotal: "85520809984"
34
+ name: NVIDIA H100 80GB HBM3
35
+ uuid: GPU-0460dc7d-d83e-fd0f-7548-a3079803b79e
36
+ host: h100-workspace-2-gpu-0-9
37
+ memory:
38
+ total: "1623005732864"
39
+ os: Linux-6.8.0-124-generic-x86_64-with-glibc2.35
40
+ program: /mloscratch/homes/navasard/sparse/SparseAtt/train.py
41
+ python: CPython 3.13.13
42
+ root: /mloscratch/homes/navasard/sparse/SparseAtt
43
+ startedAt: "2026-08-22T16:18:45.521663Z"
44
+ writerId: 0vqc90e8r076jfrxgonbbfpa2krs4vxw
45
+ m: []
46
+ python_version: 3.13.13
47
+ t:
48
+ "1":
49
+ - 1
50
+ - 5
51
+ - 11
52
+ - 49
53
+ - 51
54
+ - 53
55
+ - 71
56
+ "2":
57
+ - 1
58
+ - 5
59
+ - 11
60
+ - 49
61
+ - 51
62
+ - 53
63
+ - 71
64
+ "3":
65
+ - 2
66
+ - 5
67
+ - 13
68
+ - 14
69
+ - 15
70
+ - 16
71
+ - 61
72
+ - 62
73
+ "4": 3.13.13
74
+ "5": 0.26.1
75
+ "6": 5.7.0
76
+ "12": 0.26.1
77
+ "13": linux-x86_64
78
+ attention_dropout:
79
+ value: null
80
+ attention_type:
81
+ value: sparse_stochastic
82
+ checkpoint_activations:
83
+ value: false
84
+ dense_mode:
85
+ value: false
86
+ dense_step_intervals:
87
+ value: null
88
+ eval_batches:
89
+ value: 50
90
+ eval_steps:
91
+ value: 250
92
+ force_new_run:
93
+ value: false
94
+ gradient_accumulation_steps:
95
+ value: 5
96
+ head_dim:
97
+ value: null
98
+ hidden_size:
99
+ value: 768
100
+ init_from_scratch:
101
+ value: true
102
+ intermediate_size:
103
+ value: 3072
104
+ learning_rate:
105
+ value: 0.0005
106
+ logging_steps:
107
+ value: 10
108
+ lr_scheduler:
109
+ value: cosine
110
+ max_grad_norm:
111
+ value: 1
112
+ max_seq_length:
113
+ value: 8192
114
+ max_steps:
115
+ value: 44000
116
+ mixed_precision:
117
+ value: bf16
118
+ model_dtype:
119
+ value: bfloat16
120
+ model_path:
121
+ value: /mloscratch/homes/navasard/qwen3_06b_base
122
+ num_attention_heads:
123
+ value: 6
124
+ num_epochs:
125
+ value: 1
126
+ num_hidden_layers:
127
+ value: 28
128
+ num_key_value_heads:
129
+ value: 2
130
+ num_workers:
131
+ value: 8
132
+ output_dir:
133
+ value: /mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed46/run_20260822_115035
134
+ per_device_batch_size:
135
+ value: 2
136
+ resume_from_checkpoint:
137
+ value: /mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed46/run_20260822_115035/checkpoint-21000
138
+ rope_train_position_scale_set:
139
+ value: null
140
+ save_steps:
141
+ value: 1000
142
+ seed:
143
+ value: 46
144
+ sparse_block_size:
145
+ value: 128
146
+ sparse_layer_indices:
147
+ value: null
148
+ sparse_mask_cache_dir:
149
+ value: /mloscratch/homes/navasard/sparse/mask_cache_8k_resampled_nrb35_s0
150
+ sparse_mask_pool_size:
151
+ value: 10000
152
+ sparse_mask_seed:
153
+ value: 0
154
+ sparse_mixed_nrb_set:
155
+ value: null
156
+ sparse_num_random_blocks:
157
+ value: 35
158
+ sparse_val_dense_compare:
159
+ value: false
160
+ strategy:
161
+ value: ddp
162
+ train_path:
163
+ value: /mloscratch/homes/navasard/data/fineweb_large/tokenized
164
+ val_path:
165
+ value: null
166
+ verbose:
167
+ value: true
168
+ wandb_enabled:
169
+ value: true
170
+ wandb_name:
171
+ value: P1 DDP2 365m resampled 8k nrb35 seed46
172
+ wandb_project:
173
+ value: stochastic_sparsity
174
+ wandb_tags:
175
+ value:
176
+ - p1
177
+ - structured-control
178
+ - qwen3
179
+ - 365m
180
+ - 8k
181
+ - resampled
182
+ - nrb35
183
+ - seed46
184
+ - ddp2
185
+ - h1002
186
+ - scheduled
187
+ warmup_steps:
188
+ value: 750
189
+ weight_decay:
190
+ value: 0.01
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/output.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/wandb-metadata.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-22T16:18:45.521663Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p1_scheduled_h1002_20260822/l8k365m_resampled_nrb35_seed46.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "h100-workspace-2-gpu-0-9",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 96,
21
+ "cpu_count_logical": 192,
22
+ "gpu": "NVIDIA H100 80GB HBM3",
23
+ "gpu_count": 2,
24
+ "disk": {
25
+ "/": {
26
+ "total": "9374931091456",
27
+ "used": "6412527095808"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1623005732864"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA H100 80GB HBM3",
36
+ "memoryTotal": "85520809984",
37
+ "cudaCores": 16896,
38
+ "architecture": "Hopper",
39
+ "uuid": "GPU-af79466d-4300-4171-b5bd-7697eca27264"
40
+ },
41
+ {
42
+ "name": "NVIDIA H100 80GB HBM3",
43
+ "memoryTotal": "85520809984",
44
+ "cudaCores": 16896,
45
+ "architecture": "Hopper",
46
+ "uuid": "GPU-0460dc7d-d83e-fd0f-7548-a3079803b79e"
47
+ }
48
+ ],
49
+ "cudaVersion": "13.0",
50
+ "writerId": "0vqc90e8r076jfrxgonbbfpa2krs4vxw"
51
+ }
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"_wandb":{"runtime":32380},"_step":44000,"eval/uses_dense_attention":0,"speed/ms_per_step":691.0484075546265,"speed/steps_per_sec":1.4470766288842816,"eval/dense_interval_end":-1,"_wandb.runtime":15524,"train/learning_rate":0,"eval/loss":2.9298389625549315,"eval/perplexity":18.72461488972488,"train/epoch":0.2906439083679023,"eval/dense_interval_start":-1,"_runtime":32380.649742912,"eval/uses_sparse_attention":1,"eval/dense_interval_active":0,"train/global_step":44000,"train/loss":2.859005928039551,"_timestamp":1.787432377748786e+09,"speed/samples_per_sec":28.94153257768563}
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/logs/debug.log ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-22 16:18:45,554 INFO MainThread:207 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-22 16:18:45,554 INFO MainThread:207 [wandb_setup.py:_flush():81] Configure stats pid to 207
3
+ 2026-08-22 16:18:45,554 INFO MainThread:207 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-22 16:18:45,555 INFO MainThread:207 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_161845-8xh3oikj/logs/debug.log
5
+ 2026-08-22 16:18:45,558 INFO MainThread:207 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_161845-8xh3oikj/logs/debug-internal.log
6
+ 2026-08-22 16:18:45,558 INFO MainThread:207 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-22 16:18:45,558 INFO MainThread:207 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed46/run_20260822_115035', 'resume_from_checkpoint': '/mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed46/run_20260822_115035/checkpoint-21000', 'force_new_run': False, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, 'gradient_accumulation_steps': 5, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 46, 'verbose': True, 'checkpoint_activations': False, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': 'P1 DDP2 365m resampled 8k nrb35 seed46', 'wandb_tags': ['p1', 'structured-control', 'qwen3', '365m', '8k', 'resampled', 'nrb35', 'seed46', 'ddp2', 'h1002', 'scheduled'], 'hidden_size': 768, 'intermediate_size': 3072, 'num_attention_heads': 6, 'num_key_value_heads': 2, 'num_hidden_layers': 28, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 35, 'sparse_mask_pool_size': 10000, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': '/mloscratch/homes/navasard/sparse/mask_cache_8k_resampled_nrb35_s0', 'sparse_mask_seed': 0, 'sparse_mixed_nrb_set': None, 'dense_mode': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-22 16:18:45,558 INFO MainThread:207 [wandb_init.py:init():898] starting backend
10
+ 2026-08-22 16:18:46,488 INFO MainThread:207 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-22 16:18:46,923 INFO MainThread:207 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-22 16:18:46,926 INFO MainThread:207 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-22 16:18:46,954 INFO MainThread:207 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-22 16:18:47,483 INFO MainThread:207 [wandb_init.py:init():1051] run resumed
15
+ 2026-08-22 16:18:47,485 INFO MainThread:207 [wandb_init.py:init():1056] starting run threads in backend
16
+ 2026-08-22 16:18:47,694 INFO MainThread:207 [wandb_run.py:_console_start():2554] atexit reg
17
+ 2026-08-22 16:18:47,694 INFO MainThread:207 [wandb_run.py:_redirect():2403] redirect: wrap_raw
18
+ 2026-08-22 16:18:47,694 INFO MainThread:207 [wandb_run.py:_redirect():2472] Wrapping output streams.
19
+ 2026-08-22 16:18:47,694 INFO MainThread:207 [wandb_run.py:_redirect():2495] Redirects installed.
20
+ 2026-08-22 16:18:47,701 INFO MainThread:207 [wandb_init.py:init():1094] run started, returning control to user process
21
+ 2026-08-22 20:59:44,133 INFO MainThread:207 [wandb_run.py:_finish():2321] finishing run ruben-navasardyan4869-epfl/stochastic_sparsity/8xh3oikj
22
+ 2026-08-22 20:59:44,134 INFO MainThread:207 [wandb_run.py:_atexit_cleanup():2520] got exitcode: 0
23
+ 2026-08-22 20:59:44,135 INFO MainThread:207 [wandb_run.py:_restore():2502] restore
24
+ 2026-08-22 20:59:44,135 INFO MainThread:207 [wandb_run.py:_restore():2508] restore done
25
+ 2026-08-22 20:59:45,701 INFO MainThread:207 [wandb_run.py:_footer_sync_info():3889] logging synced files
epfl_final/SparseAtt/wandb/run-20260822_161845-8xh3oikj/run-8xh3oikj.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69fc801958838060823721d09222e69e48d51c01d41f93846895634013ab924c
3
+ size 5481444
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/config.yaml ADDED
@@ -0,0 +1,200 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.26.1
4
+ e:
5
+ 2julent36u73idzfye4y5gn025rll495:
6
+ args:
7
+ - --config
8
+ - configs/p1_scheduled_h1004_20260822/l8k365m_resampled_nrb35_seed45.json
9
+ codePath: train.py
10
+ codePathLocal: train.py
11
+ cpu_count: 96
12
+ cpu_count_logical: 192
13
+ cudaVersion: "13.0"
14
+ disk:
15
+ /:
16
+ total: "9371709865984"
17
+ used: "6533732798464"
18
+ email: ruben.navasardyan@mail.utoronto.ca
19
+ executable: /mloscratch/homes/navasard/conda/bin/python
20
+ git:
21
+ commit: 5637886f0cc760009989b14e68c6ae06d4827418
22
+ remote: https://github.com/rubentium/TokenPrune.git
23
+ gpu: NVIDIA H100 80GB HBM3
24
+ gpu_count: 4
25
+ gpu_nvidia:
26
+ - architecture: Hopper
27
+ cudaCores: 16896
28
+ memoryTotal: "85520809984"
29
+ name: NVIDIA H100 80GB HBM3
30
+ uuid: GPU-9a903f43-9348-6fa6-bcf7-e12bd01803a7
31
+ - architecture: Hopper
32
+ cudaCores: 16896
33
+ memoryTotal: "85520809984"
34
+ name: NVIDIA H100 80GB HBM3
35
+ uuid: GPU-d0294574-a180-e984-6ca3-0573e9724f2e
36
+ - architecture: Hopper
37
+ cudaCores: 16896
38
+ memoryTotal: "85520809984"
39
+ name: NVIDIA H100 80GB HBM3
40
+ uuid: GPU-24b8d952-dc08-33b9-d5df-54ce871fd288
41
+ - architecture: Hopper
42
+ cudaCores: 16896
43
+ memoryTotal: "85520809984"
44
+ name: NVIDIA H100 80GB HBM3
45
+ uuid: GPU-b87e8111-63b6-b364-28b9-c4ce747aec08
46
+ host: h100-workspace-4-gpu-3-0-9
47
+ memory:
48
+ total: "1623005732864"
49
+ os: Linux-6.8.0-124-generic-x86_64-with-glibc2.35
50
+ program: /mloscratch/homes/navasard/sparse/SparseAtt/train.py
51
+ python: CPython 3.13.13
52
+ root: /mloscratch/homes/navasard/sparse/SparseAtt
53
+ startedAt: "2026-08-22T16:19:07.033644Z"
54
+ writerId: 2julent36u73idzfye4y5gn025rll495
55
+ m: []
56
+ python_version: 3.13.13
57
+ t:
58
+ "1":
59
+ - 1
60
+ - 5
61
+ - 11
62
+ - 49
63
+ - 51
64
+ - 53
65
+ - 71
66
+ "2":
67
+ - 1
68
+ - 5
69
+ - 11
70
+ - 49
71
+ - 51
72
+ - 53
73
+ - 71
74
+ "3":
75
+ - 2
76
+ - 5
77
+ - 13
78
+ - 14
79
+ - 15
80
+ - 16
81
+ - 61
82
+ - 62
83
+ "4": 3.13.13
84
+ "5": 0.26.1
85
+ "6": 5.7.0
86
+ "12": 0.26.1
87
+ "13": linux-x86_64
88
+ attention_dropout:
89
+ value: null
90
+ attention_type:
91
+ value: sparse_stochastic
92
+ checkpoint_activations:
93
+ value: false
94
+ dense_mode:
95
+ value: false
96
+ dense_step_intervals:
97
+ value: null
98
+ eval_batches:
99
+ value: 50
100
+ eval_steps:
101
+ value: 250
102
+ force_new_run:
103
+ value: false
104
+ gradient_accumulation_steps:
105
+ value: 5
106
+ head_dim:
107
+ value: null
108
+ hidden_size:
109
+ value: 768
110
+ init_from_scratch:
111
+ value: true
112
+ intermediate_size:
113
+ value: 3072
114
+ learning_rate:
115
+ value: 0.0005
116
+ logging_steps:
117
+ value: 10
118
+ lr_scheduler:
119
+ value: cosine
120
+ max_grad_norm:
121
+ value: 1
122
+ max_seq_length:
123
+ value: 8192
124
+ max_steps:
125
+ value: 44000
126
+ mixed_precision:
127
+ value: bf16
128
+ model_dtype:
129
+ value: bfloat16
130
+ model_path:
131
+ value: /mloscratch/homes/navasard/qwen3_06b_base
132
+ num_attention_heads:
133
+ value: 6
134
+ num_epochs:
135
+ value: 1
136
+ num_hidden_layers:
137
+ value: 28
138
+ num_key_value_heads:
139
+ value: 2
140
+ num_workers:
141
+ value: 8
142
+ output_dir:
143
+ value: /mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed45/run_20260822_143256
144
+ per_device_batch_size:
145
+ value: 1
146
+ resume_from_checkpoint:
147
+ value: /mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed45/run_20260822_143256/checkpoint-13000
148
+ rope_train_position_scale_set:
149
+ value: null
150
+ save_steps:
151
+ value: 1000
152
+ seed:
153
+ value: 45
154
+ sparse_block_size:
155
+ value: 128
156
+ sparse_layer_indices:
157
+ value: null
158
+ sparse_mask_cache_dir:
159
+ value: /mloscratch/homes/navasard/sparse/mask_cache_8k_resampled_nrb35_s0
160
+ sparse_mask_pool_size:
161
+ value: 10000
162
+ sparse_mask_seed:
163
+ value: 0
164
+ sparse_mixed_nrb_set:
165
+ value: null
166
+ sparse_num_random_blocks:
167
+ value: 35
168
+ sparse_val_dense_compare:
169
+ value: false
170
+ strategy:
171
+ value: ddp
172
+ train_path:
173
+ value: /mloscratch/homes/navasard/data/fineweb_large/tokenized
174
+ val_path:
175
+ value: null
176
+ verbose:
177
+ value: true
178
+ wandb_enabled:
179
+ value: true
180
+ wandb_name:
181
+ value: P1 DDP4 365m resampled 8k nrb35 seed45
182
+ wandb_project:
183
+ value: stochastic_sparsity
184
+ wandb_tags:
185
+ value:
186
+ - p1
187
+ - structured-control
188
+ - qwen3
189
+ - 365m
190
+ - 8k
191
+ - resampled
192
+ - nrb35
193
+ - seed45
194
+ - ddp4
195
+ - h1004
196
+ - scheduled
197
+ warmup_steps:
198
+ value: 750
199
+ weight_decay:
200
+ value: 0.01
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/output.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/wandb-metadata.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-22T16:19:07.033644Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p1_scheduled_h1004_20260822/l8k365m_resampled_nrb35_seed45.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "h100-workspace-4-gpu-3-0-9",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 96,
21
+ "cpu_count_logical": 192,
22
+ "gpu": "NVIDIA H100 80GB HBM3",
23
+ "gpu_count": 4,
24
+ "disk": {
25
+ "/": {
26
+ "total": "9371709865984",
27
+ "used": "6533732798464"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1623005732864"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA H100 80GB HBM3",
36
+ "memoryTotal": "85520809984",
37
+ "cudaCores": 16896,
38
+ "architecture": "Hopper",
39
+ "uuid": "GPU-9a903f43-9348-6fa6-bcf7-e12bd01803a7"
40
+ },
41
+ {
42
+ "name": "NVIDIA H100 80GB HBM3",
43
+ "memoryTotal": "85520809984",
44
+ "cudaCores": 16896,
45
+ "architecture": "Hopper",
46
+ "uuid": "GPU-d0294574-a180-e984-6ca3-0573e9724f2e"
47
+ },
48
+ {
49
+ "name": "NVIDIA H100 80GB HBM3",
50
+ "memoryTotal": "85520809984",
51
+ "cudaCores": 16896,
52
+ "architecture": "Hopper",
53
+ "uuid": "GPU-24b8d952-dc08-33b9-d5df-54ce871fd288"
54
+ },
55
+ {
56
+ "name": "NVIDIA H100 80GB HBM3",
57
+ "memoryTotal": "85520809984",
58
+ "cudaCores": 16896,
59
+ "architecture": "Hopper",
60
+ "uuid": "GPU-b87e8111-63b6-b364-28b9-c4ce747aec08"
61
+ }
62
+ ],
63
+ "cudaVersion": "13.0",
64
+ "writerId": "2julent36u73idzfye4y5gn025rll495"
65
+ }
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"_wandb":{"runtime":18651},"eval/uses_dense_attention":0,"eval/perplexity":18.510780300655572,"_wandb.runtime":5588,"train/epoch":0.2906439083679023,"speed/steps_per_sec":2.4873652604306247,"_runtime":18651.342134185,"eval/loss":2.918353281323872,"_timestamp":1.7874286057213173e+09,"train/global_step":44000,"train/learning_rate":0,"eval/uses_sparse_attention":1,"speed/samples_per_sec":49.747305208612495,"eval/dense_interval_start":-1,"eval/dense_interval_end":-1,"speed/ms_per_step":402.0318269729614,"eval/dense_interval_active":0,"_step":44000,"train/loss":2.8272599625587462}
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/logs/debug.log ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-22 16:19:07,062 INFO MainThread:236 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_setup.py:_flush():81] Configure stats pid to 236
3
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_161907-9nppzrru/logs/debug.log
5
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_161907-9nppzrru/logs/debug-internal.log
6
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed45/run_20260822_143256', 'resume_from_checkpoint': '/mloscratch/homes/navasard/sparse/checkpoints/l8k365m_resampled_nrb35_seed45/run_20260822_143256/checkpoint-13000', 'force_new_run': False, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 1, 'gradient_accumulation_steps': 5, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 45, 'verbose': True, 'checkpoint_activations': False, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': 'P1 DDP4 365m resampled 8k nrb35 seed45', 'wandb_tags': ['p1', 'structured-control', 'qwen3', '365m', '8k', 'resampled', 'nrb35', 'seed45', 'ddp4', 'h1004', 'scheduled'], 'hidden_size': 768, 'intermediate_size': 3072, 'num_attention_heads': 6, 'num_key_value_heads': 2, 'num_hidden_layers': 28, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 35, 'sparse_mask_pool_size': 10000, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': '/mloscratch/homes/navasard/sparse/mask_cache_8k_resampled_nrb35_s0', 'sparse_mask_seed': 0, 'sparse_mixed_nrb_set': None, 'dense_mode': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-22 16:19:07,063 INFO MainThread:236 [wandb_init.py:init():898] starting backend
10
+ 2026-08-22 16:19:07,879 INFO MainThread:236 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-22 16:19:08,330 INFO MainThread:236 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-22 16:19:08,332 INFO MainThread:236 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-22 16:19:08,361 INFO MainThread:236 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-22 16:19:08,831 INFO MainThread:236 [wandb_init.py:init():1051] run resumed
15
+ 2026-08-22 16:19:08,834 INFO MainThread:236 [wandb_init.py:init():1056] starting run threads in backend
16
+ 2026-08-22 16:19:09,131 INFO MainThread:236 [wandb_run.py:_console_start():2554] atexit reg
17
+ 2026-08-22 16:19:09,131 INFO MainThread:236 [wandb_run.py:_redirect():2403] redirect: wrap_raw
18
+ 2026-08-22 16:19:09,131 INFO MainThread:236 [wandb_run.py:_redirect():2472] Wrapping output streams.
19
+ 2026-08-22 16:19:09,131 INFO MainThread:236 [wandb_run.py:_redirect():2495] Redirects installed.
20
+ 2026-08-22 16:19:09,140 INFO MainThread:236 [wandb_init.py:init():1094] run started, returning control to user process
21
+ 2026-08-22 19:56:52,174 INFO MainThread:236 [wandb_run.py:_finish():2321] finishing run ruben-navasardyan4869-epfl/stochastic_sparsity/9nppzrru
22
+ 2026-08-22 19:56:52,175 INFO MainThread:236 [wandb_run.py:_atexit_cleanup():2520] got exitcode: 0
23
+ 2026-08-22 19:56:52,176 INFO MainThread:236 [wandb_run.py:_restore():2502] restore
24
+ 2026-08-22 19:56:52,176 INFO MainThread:236 [wandb_run.py:_restore():2508] restore done
25
+ 2026-08-22 19:56:53,905 INFO MainThread:236 [wandb_run.py:_footer_sync_info():3889] logging synced files
epfl_final/SparseAtt/wandb/run-20260822_161907-9nppzrru/run-9nppzrru.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57dd155d40f0ecc09d02e1674bbd90cd2df21a4161f2a85a2af7b06ce6b04e47
3
+ size 7137612
epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/files/output.log ADDED
@@ -0,0 +1,350 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Starting training...
2
+ Switching attention mode for step 1: sparse_stochastic (configured sparse attention)
3
+ Step 10/50462 | Loss: 12.2285 | LR: 6.67e-06 | 13558.2ms/step | 4.4 samples/s
4
+ Step 20/50462 | Loss: 11.9217 | LR: 1.33e-05 | 7347.1ms/step | 8.2 samples/s
5
+ Step 30/50462 | Loss: 11.0220 | LR: 2.00e-05 | 7347.0ms/step | 8.2 samples/s
6
+ Step 40/50462 | Loss: 10.1514 | LR: 2.67e-05 | 7315.7ms/step | 8.2 samples/s
7
+ Step 50/50462 | Loss: 9.7299 | LR: 3.33e-05 | 7338.4ms/step | 8.2 samples/s
8
+ Step 60/50462 | Loss: 9.2932 | LR: 4.00e-05 | 7295.1ms/step | 8.2 samples/s
9
+ Step 70/50462 | Loss: 8.7279 | LR: 4.67e-05 | 7344.8ms/step | 8.2 samples/s
10
+ Step 80/50462 | Loss: 8.2714 | LR: 5.33e-05 | 7343.8ms/step | 8.2 samples/s
11
+ Step 90/50462 | Loss: 7.8736 | LR: 6.00e-05 | 7312.5ms/step | 8.2 samples/s
12
+ Step 100/50462 | Loss: 7.4990 | LR: 6.67e-05 | 7328.6ms/step | 8.2 samples/s
13
+ Step 110/50462 | Loss: 7.2479 | LR: 7.33e-05 | 7341.2ms/step | 8.2 samples/s
14
+ Step 120/50462 | Loss: 7.0941 | LR: 8.00e-05 | 7335.0ms/step | 8.2 samples/s
15
+ Step 130/50462 | Loss: 6.9836 | LR: 8.67e-05 | 7345.7ms/step | 8.2 samples/s
16
+ Step 140/50462 | Loss: 6.8738 | LR: 9.33e-05 | 7361.3ms/step | 8.2 samples/s
17
+ Step 150/50462 | Loss: 6.7698 | LR: 1.00e-04 | 7359.0ms/step | 8.2 samples/s
18
+ Step 160/50462 | Loss: 6.6554 | LR: 1.07e-04 | 7384.1ms/step | 8.1 samples/s
19
+ Step 170/50462 | Loss: 6.5238 | LR: 1.13e-04 | 7386.8ms/step | 8.1 samples/s
20
+ Step 180/50462 | Loss: 6.4588 | LR: 1.20e-04 | 7359.1ms/step | 8.2 samples/s
21
+ Step 190/50462 | Loss: 6.3743 | LR: 1.27e-04 | 7343.1ms/step | 8.2 samples/s
22
+ Step 200/50462 | Loss: 6.2806 | LR: 1.33e-04 | 7351.1ms/step | 8.2 samples/s
23
+ Step 210/50462 | Loss: 6.2474 | LR: 1.40e-04 | 7363.2ms/step | 8.1 samples/s
24
+ Step 220/50462 | Loss: 6.2108 | LR: 1.47e-04 | 7324.9ms/step | 8.2 samples/s
25
+ Step 230/50462 | Loss: 6.1300 | LR: 1.53e-04 | 7372.7ms/step | 8.1 samples/s
26
+ Step 240/50462 | Loss: 6.0595 | LR: 1.60e-04 | 7399.2ms/step | 8.1 samples/s
27
+ Step 250/50462 | Loss: 6.0046 | LR: 1.67e-04 | 7362.8ms/step | 8.1 samples/s
28
+ Validation at step 250: Loss = 5.9596, Perplexity = 387.45
29
+ Step 260/50462 | Loss: 5.9345 | LR: 1.73e-04 | 15108.8ms/step | 4.0 samples/s
30
+ Step 270/50462 | Loss: 5.8871 | LR: 1.80e-04 | 7355.1ms/step | 8.2 samples/s
31
+ Step 280/50462 | Loss: 5.8307 | LR: 1.87e-04 | 7401.4ms/step | 8.1 samples/s
32
+ Step 290/50462 | Loss: 5.7848 | LR: 1.93e-04 | 7368.6ms/step | 8.1 samples/s
33
+ Step 300/50462 | Loss: 5.7138 | LR: 2.00e-04 | 7369.7ms/step | 8.1 samples/s
34
+ Step 310/50462 | Loss: 5.6581 | LR: 2.07e-04 | 7373.0ms/step | 8.1 samples/s
35
+ Step 320/50462 | Loss: 5.6094 | LR: 2.13e-04 | 7385.3ms/step | 8.1 samples/s
36
+ Step 330/50462 | Loss: 5.5933 | LR: 2.20e-04 | 7341.1ms/step | 8.2 samples/s
37
+ Step 340/50462 | Loss: 5.5433 | LR: 2.27e-04 | 7334.3ms/step | 8.2 samples/s
38
+ Step 350/50462 | Loss: 5.4962 | LR: 2.33e-04 | 7377.7ms/step | 8.1 samples/s
39
+ Step 360/50462 | Loss: 5.4527 | LR: 2.40e-04 | 7381.3ms/step | 8.1 samples/s
40
+ Step 370/50462 | Loss: 5.3898 | LR: 2.47e-04 | 7373.7ms/step | 8.1 samples/s
41
+ Step 380/50462 | Loss: 5.3137 | LR: 2.53e-04 | 7353.7ms/step | 8.2 samples/s
42
+ Step 390/50462 | Loss: 5.2732 | LR: 2.60e-04 | 7360.9ms/step | 8.2 samples/s
43
+ Step 400/50462 | Loss: 5.2140 | LR: 2.67e-04 | 7363.3ms/step | 8.1 samples/s
44
+ Step 410/50462 | Loss: 5.2023 | LR: 2.73e-04 | 7377.4ms/step | 8.1 samples/s
45
+ Step 420/50462 | Loss: 5.1795 | LR: 2.80e-04 | 7380.2ms/step | 8.1 samples/s
46
+ Step 430/50462 | Loss: 5.1245 | LR: 2.87e-04 | 7372.2ms/step | 8.1 samples/s
47
+ Step 440/50462 | Loss: 5.0656 | LR: 2.93e-04 | 7356.0ms/step | 8.2 samples/s
48
+ Step 450/50462 | Loss: 5.0408 | LR: 3.00e-04 | 7371.9ms/step | 8.1 samples/s
49
+ Step 460/50462 | Loss: 5.0109 | LR: 3.07e-04 | 7383.0ms/step | 8.1 samples/s
50
+ Step 470/50462 | Loss: 5.0111 | LR: 3.13e-04 | 7364.5ms/step | 8.1 samples/s
51
+ Step 480/50462 | Loss: 4.9881 | LR: 3.20e-04 | 7383.8ms/step | 8.1 samples/s
52
+ Step 490/50462 | Loss: 4.8550 | LR: 3.27e-04 | 7361.0ms/step | 8.2 samples/s
53
+ Step 500/50462 | Loss: 4.8979 | LR: 3.33e-04 | 7376.8ms/step | 8.1 samples/s
54
+ Validation at step 500: Loss = 4.9163, Perplexity = 136.50
55
+ Step 510/50462 | Loss: 4.9215 | LR: 3.40e-04 | 11221.6ms/step | 5.3 samples/s
56
+ Step 520/50462 | Loss: 4.8483 | LR: 3.47e-04 | 7382.2ms/step | 8.1 samples/s
57
+ Step 530/50462 | Loss: 4.7789 | LR: 3.53e-04 | 7335.3ms/step | 8.2 samples/s
58
+ Step 540/50462 | Loss: 4.7686 | LR: 3.60e-04 | 7365.3ms/step | 8.1 samples/s
59
+ Step 550/50462 | Loss: 4.8117 | LR: 3.67e-04 | 7364.1ms/step | 8.1 samples/s
60
+ Step 560/50462 | Loss: 4.7064 | LR: 3.73e-04 | 7379.0ms/step | 8.1 samples/s
61
+ Step 570/50462 | Loss: 4.6854 | LR: 3.80e-04 | 7355.6ms/step | 8.2 samples/s
62
+ Step 580/50462 | Loss: 4.6726 | LR: 3.87e-04 | 7331.4ms/step | 8.2 samples/s
63
+ Step 590/50462 | Loss: 4.5973 | LR: 3.93e-04 | 7381.6ms/step | 8.1 samples/s
64
+ Step 600/50462 | Loss: 4.5529 | LR: 4.00e-04 | 7349.9ms/step | 8.2 samples/s
65
+ Step 610/50462 | Loss: 4.6087 | LR: 4.07e-04 | 7334.5ms/step | 8.2 samples/s
66
+ Step 620/50462 | Loss: 4.6407 | LR: 4.13e-04 | 7371.9ms/step | 8.1 samples/s
67
+ Step 630/50462 | Loss: 4.5022 | LR: 4.20e-04 | 7390.2ms/step | 8.1 samples/s
68
+ Step 640/50462 | Loss: 4.4227 | LR: 4.27e-04 | 7391.0ms/step | 8.1 samples/s
69
+ Step 650/50462 | Loss: 4.4289 | LR: 4.33e-04 | 7363.4ms/step | 8.1 samples/s
70
+ Step 660/50462 | Loss: 4.4917 | LR: 4.40e-04 | 7351.9ms/step | 8.2 samples/s
71
+ Step 670/50462 | Loss: 4.4139 | LR: 4.47e-04 | 7390.2ms/step | 8.1 samples/s
72
+ Step 680/50462 | Loss: 4.3439 | LR: 4.53e-04 | 7400.2ms/step | 8.1 samples/s
73
+ Step 690/50462 | Loss: 4.2979 | LR: 4.60e-04 | 7343.6ms/step | 8.2 samples/s
74
+ Step 700/50462 | Loss: 4.3337 | LR: 4.67e-04 | 7367.1ms/step | 8.1 samples/s
75
+ Step 710/50462 | Loss: 4.3510 | LR: 4.73e-04 | 7343.7ms/step | 8.2 samples/s
76
+ Step 720/50462 | Loss: 4.3167 | LR: 4.80e-04 | 7353.9ms/step | 8.2 samples/s
77
+ Step 730/50462 | Loss: 4.2202 | LR: 4.87e-04 | 7370.9ms/step | 8.1 samples/s
78
+ Step 740/50462 | Loss: 4.1858 | LR: 4.93e-04 | 7374.4ms/step | 8.1 samples/s
79
+ Step 750/50462 | Loss: 4.2042 | LR: 5.00e-04 | 7381.0ms/step | 8.1 samples/s
80
+ Validation at step 750: Loss = 4.1897, Perplexity = 66.01
81
+ Step 760/50462 | Loss: 4.1547 | LR: 5.00e-04 | 11212.7ms/step | 5.4 samples/s
82
+ Step 770/50462 | Loss: 4.1554 | LR: 5.00e-04 | 7374.2ms/step | 8.1 samples/s
83
+ Step 780/50462 | Loss: 4.1212 | LR: 5.00e-04 | 7378.7ms/step | 8.1 samples/s
84
+ Step 790/50462 | Loss: 4.0891 | LR: 5.00e-04 | 7378.7ms/step | 8.1 samples/s
85
+ Step 800/50462 | Loss: 4.0808 | LR: 5.00e-04 | 7383.1ms/step | 8.1 samples/s
86
+ Step 810/50462 | Loss: 4.0226 | LR: 5.00e-04 | 7397.9ms/step | 8.1 samples/s
87
+ Step 820/50462 | Loss: 4.0400 | LR: 5.00e-04 | 7372.5ms/step | 8.1 samples/s
88
+ Step 830/50462 | Loss: 4.0047 | LR: 5.00e-04 | 7357.1ms/step | 8.2 samples/s
89
+ Step 840/50462 | Loss: 4.0164 | LR: 5.00e-04 | 7380.1ms/step | 8.1 samples/s
90
+ Step 850/50462 | Loss: 3.9565 | LR: 5.00e-04 | 7381.0ms/step | 8.1 samples/s
91
+ Step 860/50462 | Loss: 3.9267 | LR: 5.00e-04 | 7365.3ms/step | 8.1 samples/s
92
+ Step 870/50462 | Loss: 3.9078 | LR: 5.00e-04 | 7375.7ms/step | 8.1 samples/s
93
+ Step 880/50462 | Loss: 3.9065 | LR: 5.00e-04 | 7360.9ms/step | 8.2 samples/s
94
+ Step 890/50462 | Loss: 3.9025 | LR: 5.00e-04 | 7364.2ms/step | 8.1 samples/s
95
+ Step 900/50462 | Loss: 3.8937 | LR: 5.00e-04 | 7341.4ms/step | 8.2 samples/s
96
+ Step 910/50462 | Loss: 3.8811 | LR: 5.00e-04 | 7388.4ms/step | 8.1 samples/s
97
+ Step 920/50462 | Loss: 3.8701 | LR: 5.00e-04 | 7363.6ms/step | 8.1 samples/s
98
+ Step 930/50462 | Loss: 3.8368 | LR: 5.00e-04 | 7377.5ms/step | 8.1 samples/s
99
+ Step 940/50462 | Loss: 3.8551 | LR: 5.00e-04 | 7390.4ms/step | 8.1 samples/s
100
+ Step 950/50462 | Loss: 3.8354 | LR: 5.00e-04 | 7345.6ms/step | 8.2 samples/s
101
+ Step 960/50462 | Loss: 3.7922 | LR: 5.00e-04 | 7367.2ms/step | 8.1 samples/s
102
+ Step 970/50462 | Loss: 3.8038 | LR: 5.00e-04 | 7372.8ms/step | 8.1 samples/s
103
+ Step 980/50462 | Loss: 3.7930 | LR: 5.00e-04 | 7374.7ms/step | 8.1 samples/s
104
+ Step 990/50462 | Loss: 3.7858 | LR: 5.00e-04 | 7361.0ms/step | 8.2 samples/s
105
+ Step 1000/50462 | Loss: 3.7586 | LR: 5.00e-04 | 7394.0ms/step | 8.1 samples/s
106
+ Validation at step 1000: Loss = 3.7727, Perplexity = 43.50
107
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_mixed_seed70/run_20260822_210107/checkpoint-1000
108
+ Step 1010/50462 | Loss: 3.7469 | LR: 5.00e-04 | 12122.1ms/step | 4.9 samples/s
109
+ Step 1020/50462 | Loss: 3.7524 | LR: 5.00e-04 | 7364.7ms/step | 8.1 samples/s
110
+ Step 1030/50462 | Loss: 3.7150 | LR: 5.00e-04 | 7356.9ms/step | 8.2 samples/s
111
+ Step 1040/50462 | Loss: 3.7204 | LR: 5.00e-04 | 7359.6ms/step | 8.2 samples/s
112
+ Step 1050/50462 | Loss: 3.7048 | LR: 5.00e-04 | 7369.4ms/step | 8.1 samples/s
113
+ Step 1060/50462 | Loss: 3.7026 | LR: 5.00e-04 | 7397.3ms/step | 8.1 samples/s
114
+ Step 1070/50462 | Loss: 3.6855 | LR: 5.00e-04 | 7386.1ms/step | 8.1 samples/s
115
+ Step 1080/50462 | Loss: 3.6904 | LR: 5.00e-04 | 7365.9ms/step | 8.1 samples/s
116
+ Step 1090/50462 | Loss: 3.6655 | LR: 5.00e-04 | 7377.9ms/step | 8.1 samples/s
117
+ Step 1100/50462 | Loss: 3.6564 | LR: 5.00e-04 | 7375.4ms/step | 8.1 samples/s
118
+ Step 1110/50462 | Loss: 3.6768 | LR: 5.00e-04 | 7365.5ms/step | 8.1 samples/s
119
+ Step 1120/50462 | Loss: 3.6380 | LR: 5.00e-04 | 7389.6ms/step | 8.1 samples/s
120
+ Step 1130/50462 | Loss: 3.6196 | LR: 5.00e-04 | 7389.6ms/step | 8.1 samples/s
121
+ Step 1140/50462 | Loss: 3.6203 | LR: 5.00e-04 | 7380.5ms/step | 8.1 samples/s
122
+ Step 1150/50462 | Loss: 3.6380 | LR: 5.00e-04 | 7374.1ms/step | 8.1 samples/s
123
+ Step 1160/50462 | Loss: 3.6174 | LR: 5.00e-04 | 7363.0ms/step | 8.1 samples/s
124
+ Step 1170/50462 | Loss: 3.6118 | LR: 5.00e-04 | 7369.1ms/step | 8.1 samples/s
125
+ Step 1180/50462 | Loss: 3.5977 | LR: 5.00e-04 | 7338.4ms/step | 8.2 samples/s
126
+ Step 1190/50462 | Loss: 3.6211 | LR: 5.00e-04 | 7363.7ms/step | 8.1 samples/s
127
+ Step 1200/50462 | Loss: 3.5694 | LR: 5.00e-04 | 7369.7ms/step | 8.1 samples/s
128
+ Step 1210/50462 | Loss: 3.5754 | LR: 5.00e-04 | 7364.3ms/step | 8.1 samples/s
129
+ Step 1220/50462 | Loss: 3.6185 | LR: 5.00e-04 | 7361.4ms/step | 8.2 samples/s
130
+ Step 1230/50462 | Loss: 3.5867 | LR: 5.00e-04 | 7368.8ms/step | 8.1 samples/s
131
+ Step 1240/50462 | Loss: 3.5624 | LR: 5.00e-04 | 7343.2ms/step | 8.2 samples/s
132
+ Step 1250/50462 | Loss: 3.5676 | LR: 5.00e-04 | 7344.9ms/step | 8.2 samples/s
133
+ Validation at step 1250: Loss = 3.5575, Perplexity = 35.08
134
+ Step 1260/50462 | Loss: 3.5516 | LR: 5.00e-04 | 11209.6ms/step | 5.4 samples/s
135
+ Step 1270/50462 | Loss: 3.5104 | LR: 5.00e-04 | 7370.2ms/step | 8.1 samples/s
136
+ Step 1280/50462 | Loss: 3.5620 | LR: 5.00e-04 | 7376.4ms/step | 8.1 samples/s
137
+ Step 1290/50462 | Loss: 3.5158 | LR: 5.00e-04 | 7349.9ms/step | 8.2 samples/s
138
+ Step 1300/50462 | Loss: 3.5342 | LR: 5.00e-04 | 7373.2ms/step | 8.1 samples/s
139
+ Step 1310/50462 | Loss: 3.5230 | LR: 5.00e-04 | 7391.4ms/step | 8.1 samples/s
140
+ Step 1320/50462 | Loss: 3.5181 | LR: 5.00e-04 | 7381.5ms/step | 8.1 samples/s
141
+ Step 1330/50462 | Loss: 3.4818 | LR: 5.00e-04 | 7386.5ms/step | 8.1 samples/s
142
+ Step 1340/50462 | Loss: 3.4871 | LR: 5.00e-04 | 7382.4ms/step | 8.1 samples/s
143
+ Step 1350/50462 | Loss: 3.4923 | LR: 5.00e-04 | 7370.9ms/step | 8.1 samples/s
144
+ Step 1360/50462 | Loss: 3.4964 | LR: 5.00e-04 | 7355.6ms/step | 8.2 samples/s
145
+ Step 1370/50462 | Loss: 3.5066 | LR: 5.00e-04 | 7372.6ms/step | 8.1 samples/s
146
+ Step 1380/50462 | Loss: 3.4892 | LR: 5.00e-04 | 7384.9ms/step | 8.1 samples/s
147
+ Step 1390/50462 | Loss: 3.4640 | LR: 5.00e-04 | 7387.6ms/step | 8.1 samples/s
148
+ Step 1400/50462 | Loss: 3.4667 | LR: 5.00e-04 | 7355.3ms/step | 8.2 samples/s
149
+ Step 1410/50462 | Loss: 3.4724 | LR: 5.00e-04 | 7371.0ms/step | 8.1 samples/s
150
+ Step 1420/50462 | Loss: 3.4577 | LR: 5.00e-04 | 7366.0ms/step | 8.1 samples/s
151
+ Step 1430/50462 | Loss: 3.4689 | LR: 5.00e-04 | 7380.9ms/step | 8.1 samples/s
152
+ Step 1440/50462 | Loss: 3.4861 | LR: 5.00e-04 | 7367.2ms/step | 8.1 samples/s
153
+ Step 1450/50462 | Loss: 3.4154 | LR: 5.00e-04 | 7378.6ms/step | 8.1 samples/s
154
+ Step 1460/50462 | Loss: 3.4327 | LR: 5.00e-04 | 7386.3ms/step | 8.1 samples/s
155
+ Step 1470/50462 | Loss: 3.4261 | LR: 5.00e-04 | 7363.5ms/step | 8.1 samples/s
156
+ Step 1480/50462 | Loss: 3.4263 | LR: 5.00e-04 | 7359.7ms/step | 8.2 samples/s
157
+ Step 1490/50462 | Loss: 3.4337 | LR: 5.00e-04 | 7356.8ms/step | 8.2 samples/s
158
+ Step 1500/50462 | Loss: 3.4186 | LR: 5.00e-04 | 7358.8ms/step | 8.2 samples/s
159
+ Validation at step 1500: Loss = 3.4334, Perplexity = 30.98
160
+ Step 1510/50462 | Loss: 3.3993 | LR: 5.00e-04 | 11241.4ms/step | 5.3 samples/s
161
+ Step 1520/50462 | Loss: 3.3938 | LR: 5.00e-04 | 7390.4ms/step | 8.1 samples/s
162
+ Step 1530/50462 | Loss: 3.4041 | LR: 5.00e-04 | 7363.7ms/step | 8.1 samples/s
163
+ Step 1540/50462 | Loss: 3.4045 | LR: 5.00e-04 | 7373.7ms/step | 8.1 samples/s
164
+ Step 1550/50462 | Loss: 3.4000 | LR: 5.00e-04 | 7353.9ms/step | 8.2 samples/s
165
+ Step 1560/50462 | Loss: 3.4077 | LR: 5.00e-04 | 7359.4ms/step | 8.2 samples/s
166
+ Step 1570/50462 | Loss: 3.4002 | LR: 5.00e-04 | 7372.3ms/step | 8.1 samples/s
167
+ Step 1580/50462 | Loss: 3.3963 | LR: 5.00e-04 | 7343.2ms/step | 8.2 samples/s
168
+ Step 1590/50462 | Loss: 3.3845 | LR: 5.00e-04 | 7337.8ms/step | 8.2 samples/s
169
+ Step 1600/50462 | Loss: 3.4028 | LR: 5.00e-04 | 7377.2ms/step | 8.1 samples/s
170
+ Step 1610/50462 | Loss: 3.3927 | LR: 5.00e-04 | 7372.9ms/step | 8.1 samples/s
171
+ Step 1620/50462 | Loss: 3.3785 | LR: 5.00e-04 | 7391.5ms/step | 8.1 samples/s
172
+ Step 1630/50462 | Loss: 3.3889 | LR: 5.00e-04 | 7370.4ms/step | 8.1 samples/s
173
+ Step 1640/50462 | Loss: 3.3532 | LR: 5.00e-04 | 7387.5ms/step | 8.1 samples/s
174
+ Step 1650/50462 | Loss: 3.3588 | LR: 5.00e-04 | 7345.7ms/step | 8.2 samples/s
175
+ Step 1660/50462 | Loss: 3.3696 | LR: 5.00e-04 | 7386.8ms/step | 8.1 samples/s
176
+ Step 1670/50462 | Loss: 3.3443 | LR: 5.00e-04 | 7360.3ms/step | 8.2 samples/s
177
+ Step 1680/50462 | Loss: 3.3622 | LR: 5.00e-04 | 7370.5ms/step | 8.1 samples/s
178
+ Step 1690/50462 | Loss: 3.3558 | LR: 5.00e-04 | 7360.3ms/step | 8.2 samples/s
179
+ Step 1700/50462 | Loss: 3.3337 | LR: 5.00e-04 | 7360.3ms/step | 8.2 samples/s
180
+ Step 1710/50462 | Loss: 3.3476 | LR: 5.00e-04 | 7362.4ms/step | 8.1 samples/s
181
+ Step 1720/50462 | Loss: 3.3274 | LR: 5.00e-04 | 7381.6ms/step | 8.1 samples/s
182
+ Step 1730/50462 | Loss: 3.3361 | LR: 5.00e-04 | 7371.6ms/step | 8.1 samples/s
183
+ Step 1740/50462 | Loss: 3.3073 | LR: 5.00e-04 | 7363.3ms/step | 8.1 samples/s
184
+ Step 1750/50462 | Loss: 3.3367 | LR: 5.00e-04 | 7366.8ms/step | 8.1 samples/s
185
+ Validation at step 1750: Loss = 3.3428, Perplexity = 28.30
186
+ Step 1760/50462 | Loss: 3.3488 | LR: 4.99e-04 | 11240.3ms/step | 5.3 samples/s
187
+ Step 1770/50462 | Loss: 3.3295 | LR: 4.99e-04 | 7368.6ms/step | 8.1 samples/s
188
+ Step 1780/50462 | Loss: 3.3044 | LR: 4.99e-04 | 7381.3ms/step | 8.1 samples/s
189
+ Step 1790/50462 | Loss: 3.3265 | LR: 4.99e-04 | 7351.5ms/step | 8.2 samples/s
190
+ Step 1800/50462 | Loss: 3.3426 | LR: 4.99e-04 | 7354.6ms/step | 8.2 samples/s
191
+ Step 1810/50462 | Loss: 3.3453 | LR: 4.99e-04 | 7364.2ms/step | 8.1 samples/s
192
+ Step 1820/50462 | Loss: 3.3235 | LR: 4.99e-04 | 7377.3ms/step | 8.1 samples/s
193
+ Step 1830/50462 | Loss: 3.3117 | LR: 4.99e-04 | 7371.0ms/step | 8.1 samples/s
194
+ Step 1840/50462 | Loss: 3.2834 | LR: 4.99e-04 | 7374.8ms/step | 8.1 samples/s
195
+ Step 1850/50462 | Loss: 3.3039 | LR: 4.99e-04 | 7383.4ms/step | 8.1 samples/s
196
+ Step 1860/50462 | Loss: 3.2982 | LR: 4.99e-04 | 7386.3ms/step | 8.1 samples/s
197
+ Step 1870/50462 | Loss: 3.3197 | LR: 4.99e-04 | 7362.8ms/step | 8.1 samples/s
198
+ Step 1880/50462 | Loss: 3.3018 | LR: 4.99e-04 | 7345.8ms/step | 8.2 samples/s
199
+ Step 1890/50462 | Loss: 3.2601 | LR: 4.99e-04 | 7371.6ms/step | 8.1 samples/s
200
+ Step 1900/50462 | Loss: 3.2972 | LR: 4.99e-04 | 7377.8ms/step | 8.1 samples/s
201
+ Step 1910/50462 | Loss: 3.2853 | LR: 4.99e-04 | 7369.0ms/step | 8.1 samples/s
202
+ Step 1920/50462 | Loss: 3.2891 | LR: 4.99e-04 | 7334.5ms/step | 8.2 samples/s
203
+ Step 1930/50462 | Loss: 3.2646 | LR: 4.99e-04 | 7325.2ms/step | 8.2 samples/s
204
+ Step 1940/50462 | Loss: 3.2892 | LR: 4.99e-04 | 7362.3ms/step | 8.1 samples/s
205
+ Step 1950/50462 | Loss: 3.2870 | LR: 4.99e-04 | 7350.6ms/step | 8.2 samples/s
206
+ Step 1960/50462 | Loss: 3.2640 | LR: 4.99e-04 | 7379.5ms/step | 8.1 samples/s
207
+ Step 1970/50462 | Loss: 3.2668 | LR: 4.99e-04 | 7381.9ms/step | 8.1 samples/s
208
+ Step 1980/50462 | Loss: 3.2723 | LR: 4.99e-04 | 7371.1ms/step | 8.1 samples/s
209
+ Step 1990/50462 | Loss: 3.2744 | LR: 4.99e-04 | 7357.0ms/step | 8.2 samples/s
210
+ Step 2000/50462 | Loss: 3.2649 | LR: 4.99e-04 | 7394.0ms/step | 8.1 samples/s
211
+ Validation at step 2000: Loss = 3.2750, Perplexity = 26.44
212
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_mixed_seed70/run_20260822_210107/checkpoint-2000
213
+ Step 2010/50462 | Loss: 3.2365 | LR: 4.99e-04 | 12182.3ms/step | 4.9 samples/s
214
+ Step 2020/50462 | Loss: 3.2684 | LR: 4.99e-04 | 7350.3ms/step | 8.2 samples/s
215
+ Step 2030/50462 | Loss: 3.2474 | LR: 4.99e-04 | 7358.6ms/step | 8.2 samples/s
216
+ Step 2040/50462 | Loss: 3.2492 | LR: 4.99e-04 | 7387.8ms/step | 8.1 samples/s
217
+ Step 2050/50462 | Loss: 3.2074 | LR: 4.99e-04 | 7398.8ms/step | 8.1 samples/s
218
+ Step 2060/50462 | Loss: 3.2730 | LR: 4.99e-04 | 7362.4ms/step | 8.1 samples/s
219
+ Step 2070/50462 | Loss: 3.2711 | LR: 4.99e-04 | 7352.6ms/step | 8.2 samples/s
220
+ Step 2080/50462 | Loss: 3.2437 | LR: 4.99e-04 | 7358.6ms/step | 8.2 samples/s
221
+ Step 2090/50462 | Loss: 3.2618 | LR: 4.99e-04 | 7356.2ms/step | 8.2 samples/s
222
+ Step 2100/50462 | Loss: 3.2509 | LR: 4.99e-04 | 7385.5ms/step | 8.1 samples/s
223
+ Step 2110/50462 | Loss: 3.2308 | LR: 4.99e-04 | 7381.3ms/step | 8.1 samples/s
224
+ Step 2120/50462 | Loss: 3.2274 | LR: 4.99e-04 | 7387.9ms/step | 8.1 samples/s
225
+ Step 2130/50462 | Loss: 3.2482 | LR: 4.99e-04 | 7376.4ms/step | 8.1 samples/s
226
+ Step 2140/50462 | Loss: 3.2243 | LR: 4.99e-04 | 7374.2ms/step | 8.1 samples/s
227
+ Step 2150/50462 | Loss: 3.2249 | LR: 4.99e-04 | 7367.1ms/step | 8.1 samples/s
228
+ Step 2160/50462 | Loss: 3.2319 | LR: 4.99e-04 | 7336.0ms/step | 8.2 samples/s
229
+ Step 2170/50462 | Loss: 3.2290 | LR: 4.99e-04 | 7345.3ms/step | 8.2 samples/s
230
+ Step 2180/50462 | Loss: 3.2386 | LR: 4.99e-04 | 7376.9ms/step | 8.1 samples/s
231
+ Step 2190/50462 | Loss: 3.2052 | LR: 4.99e-04 | 7368.0ms/step | 8.1 samples/s
232
+ Step 2200/50462 | Loss: 3.2152 | LR: 4.99e-04 | 7360.8ms/step | 8.2 samples/s
233
+ Step 2210/50462 | Loss: 3.2154 | LR: 4.99e-04 | 7399.2ms/step | 8.1 samples/s
234
+ Step 2220/50462 | Loss: 3.2344 | LR: 4.99e-04 | 7379.5ms/step | 8.1 samples/s
235
+ Step 2230/50462 | Loss: 3.2164 | LR: 4.99e-04 | 7384.7ms/step | 8.1 samples/s
236
+ Step 2240/50462 | Loss: 3.2290 | LR: 4.99e-04 | 7360.1ms/step | 8.2 samples/s
237
+ Step 2250/50462 | Loss: 3.2164 | LR: 4.99e-04 | 7361.5ms/step | 8.2 samples/s
238
+ Validation at step 2250: Loss = 3.2215, Perplexity = 25.07
239
+ Step 2260/50462 | Loss: 3.2159 | LR: 4.99e-04 | 11245.4ms/step | 5.3 samples/s
240
+ Step 2270/50462 | Loss: 3.2168 | LR: 4.99e-04 | 7374.4ms/step | 8.1 samples/s
241
+ Step 2280/50462 | Loss: 3.2215 | LR: 4.99e-04 | 7384.3ms/step | 8.1 samples/s
242
+ Step 2290/50462 | Loss: 3.2073 | LR: 4.99e-04 | 7379.9ms/step | 8.1 samples/s
243
+ Step 2300/50462 | Loss: 3.2177 | LR: 4.99e-04 | 7398.5ms/step | 8.1 samples/s
244
+ Step 2310/50462 | Loss: 3.2115 | LR: 4.99e-04 | 7360.8ms/step | 8.2 samples/s
245
+ Step 2320/50462 | Loss: 3.2178 | LR: 4.99e-04 | 7378.7ms/step | 8.1 samples/s
246
+ Step 2330/50462 | Loss: 3.1953 | LR: 4.99e-04 | 7378.5ms/step | 8.1 samples/s
247
+ Step 2340/50462 | Loss: 3.2209 | LR: 4.99e-04 | 7343.1ms/step | 8.2 samples/s
248
+ Step 2350/50462 | Loss: 3.1798 | LR: 4.99e-04 | 7357.7ms/step | 8.2 samples/s
249
+ Step 2360/50462 | Loss: 3.1912 | LR: 4.99e-04 | 7357.5ms/step | 8.2 samples/s
250
+ Step 2370/50462 | Loss: 3.2037 | LR: 4.99e-04 | 7364.4ms/step | 8.1 samples/s
251
+ Step 2380/50462 | Loss: 3.1862 | LR: 4.99e-04 | 7359.4ms/step | 8.2 samples/s
252
+ Step 2390/50462 | Loss: 3.2056 | LR: 4.99e-04 | 7380.6ms/step | 8.1 samples/s
253
+ Step 2400/50462 | Loss: 3.2027 | LR: 4.99e-04 | 7377.0ms/step | 8.1 samples/s
254
+ Step 2410/50462 | Loss: 3.1688 | LR: 4.99e-04 | 7356.0ms/step | 8.2 samples/s
255
+ Step 2420/50462 | Loss: 3.1642 | LR: 4.99e-04 | 7364.2ms/step | 8.1 samples/s
256
+ Step 2430/50462 | Loss: 3.1793 | LR: 4.99e-04 | 7400.6ms/step | 8.1 samples/s
257
+ Step 2440/50462 | Loss: 3.1719 | LR: 4.99e-04 | 7396.7ms/step | 8.1 samples/s
258
+ Step 2450/50462 | Loss: 3.1869 | LR: 4.99e-04 | 7373.8ms/step | 8.1 samples/s
259
+ Step 2460/50462 | Loss: 3.1767 | LR: 4.99e-04 | 7387.4ms/step | 8.1 samples/s
260
+ Step 2470/50462 | Loss: 3.1712 | LR: 4.99e-04 | 7355.1ms/step | 8.2 samples/s
261
+ Step 2480/50462 | Loss: 3.1618 | LR: 4.99e-04 | 7382.6ms/step | 8.1 samples/s
262
+ Step 2490/50462 | Loss: 3.1676 | LR: 4.98e-04 | 7389.2ms/step | 8.1 samples/s
263
+ Step 2500/50462 | Loss: 3.1875 | LR: 4.98e-04 | 7374.6ms/step | 8.1 samples/s
264
+ Validation at step 2500: Loss = 3.1794, Perplexity = 24.03
265
+ Step 2510/50462 | Loss: 3.1607 | LR: 4.98e-04 | 11242.0ms/step | 5.3 samples/s
266
+ Step 2520/50462 | Loss: 3.1564 | LR: 4.98e-04 | 7383.3ms/step | 8.1 samples/s
267
+ Step 2530/50462 | Loss: 3.1756 | LR: 4.98e-04 | 7379.6ms/step | 8.1 samples/s
268
+ Step 2540/50462 | Loss: 3.1665 | LR: 4.98e-04 | 7360.2ms/step | 8.2 samples/s
269
+ Step 2550/50462 | Loss: 3.1647 | LR: 4.98e-04 | 7362.0ms/step | 8.2 samples/s
270
+ Step 2560/50462 | Loss: 3.1507 | LR: 4.98e-04 | 7372.0ms/step | 8.1 samples/s
271
+ Step 2570/50462 | Loss: 3.1585 | LR: 4.98e-04 | 7358.7ms/step | 8.2 samples/s
272
+ Step 2580/50462 | Loss: 3.1548 | LR: 4.98e-04 | 7358.4ms/step | 8.2 samples/s
273
+ Step 2590/50462 | Loss: 3.1384 | LR: 4.98e-04 | 7380.4ms/step | 8.1 samples/s
274
+ Step 2600/50462 | Loss: 3.1661 | LR: 4.98e-04 | 7371.2ms/step | 8.1 samples/s
275
+ Step 2610/50462 | Loss: 3.1530 | LR: 4.98e-04 | 7399.3ms/step | 8.1 samples/s
276
+ Step 2620/50462 | Loss: 3.1696 | LR: 4.98e-04 | 7373.0ms/step | 8.1 samples/s
277
+ Step 2630/50462 | Loss: 3.1277 | LR: 4.98e-04 | 7367.1ms/step | 8.1 samples/s
278
+ Step 2640/50462 | Loss: 3.1604 | LR: 4.98e-04 | 7375.0ms/step | 8.1 samples/s
279
+ Step 2650/50462 | Loss: 3.1427 | LR: 4.98e-04 | 7388.8ms/step | 8.1 samples/s
280
+ Step 2660/50462 | Loss: 3.1738 | LR: 4.98e-04 | 7347.5ms/step | 8.2 samples/s
281
+ Step 2670/50462 | Loss: 3.1324 | LR: 4.98e-04 | 7392.1ms/step | 8.1 samples/s
282
+ Step 2680/50462 | Loss: 3.1455 | LR: 4.98e-04 | 7374.6ms/step | 8.1 samples/s
283
+ Step 2690/50462 | Loss: 3.1413 | LR: 4.98e-04 | 7373.6ms/step | 8.1 samples/s
284
+ Step 2700/50462 | Loss: 3.1465 | LR: 4.98e-04 | 7378.4ms/step | 8.1 samples/s
285
+ Step 2710/50462 | Loss: 3.1472 | LR: 4.98e-04 | 7352.7ms/step | 8.2 samples/s
286
+ Step 2720/50462 | Loss: 3.1333 | LR: 4.98e-04 | 7373.7ms/step | 8.1 samples/s
287
+ Step 2730/50462 | Loss: 3.1362 | LR: 4.98e-04 | 7356.3ms/step | 8.2 samples/s
288
+ Step 2740/50462 | Loss: 3.1527 | LR: 4.98e-04 | 7366.7ms/step | 8.1 samples/s
289
+ Step 2750/50462 | Loss: 3.1580 | LR: 4.98e-04 | 7362.4ms/step | 8.1 samples/s
290
+ Validation at step 2750: Loss = 3.1435, Perplexity = 23.18
291
+ Step 2760/50462 | Loss: 3.1579 | LR: 4.98e-04 | 11246.1ms/step | 5.3 samples/s
292
+ Step 2770/50462 | Loss: 3.1259 | LR: 4.98e-04 | 7389.8ms/step | 8.1 samples/s
293
+ Step 2780/50462 | Loss: 3.1441 | LR: 4.98e-04 | 7386.7ms/step | 8.1 samples/s
294
+ Step 2790/50462 | Loss: 3.1499 | LR: 4.98e-04 | 7361.5ms/step | 8.2 samples/s
295
+ Step 2800/50462 | Loss: 3.1231 | LR: 4.98e-04 | 7362.2ms/step | 8.1 samples/s
296
+ Step 2810/50462 | Loss: 3.1289 | LR: 4.98e-04 | 7354.5ms/step | 8.2 samples/s
297
+ Step 2820/50462 | Loss: 3.1178 | LR: 4.98e-04 | 7373.0ms/step | 8.1 samples/s
298
+ Step 2830/50462 | Loss: 3.0993 | LR: 4.98e-04 | 7355.6ms/step | 8.2 samples/s
299
+ Step 2840/50462 | Loss: 3.1170 | LR: 4.98e-04 | 7344.4ms/step | 8.2 samples/s
300
+ Step 2850/50462 | Loss: 3.1312 | LR: 4.98e-04 | 7373.1ms/step | 8.1 samples/s
301
+ Step 2860/50462 | Loss: 3.1184 | LR: 4.98e-04 | 7376.8ms/step | 8.1 samples/s
302
+ Step 2870/50462 | Loss: 3.1153 | LR: 4.98e-04 | 7374.4ms/step | 8.1 samples/s
303
+ Step 2880/50462 | Loss: 3.1072 | LR: 4.98e-04 | 7364.4ms/step | 8.1 samples/s
304
+ Step 2890/50462 | Loss: 3.1181 | LR: 4.98e-04 | 7380.4ms/step | 8.1 samples/s
305
+ Step 2900/50462 | Loss: 3.1301 | LR: 4.98e-04 | 7360.8ms/step | 8.2 samples/s
306
+ Step 2910/50462 | Loss: 3.0896 | LR: 4.98e-04 | 7358.5ms/step | 8.2 samples/s
307
+ Step 2920/50462 | Loss: 3.1073 | LR: 4.98e-04 | 7381.8ms/step | 8.1 samples/s
308
+ Step 2930/50462 | Loss: 3.1197 | LR: 4.98e-04 | 7370.4ms/step | 8.1 samples/s
309
+ Step 2940/50462 | Loss: 3.0829 | LR: 4.98e-04 | 7377.3ms/step | 8.1 samples/s
310
+ Step 2950/50462 | Loss: 3.1117 | LR: 4.98e-04 | 7384.8ms/step | 8.1 samples/s
311
+ Step 2960/50462 | Loss: 3.1098 | LR: 4.98e-04 | 7358.4ms/step | 8.2 samples/s
312
+ Step 2970/50462 | Loss: 3.1180 | LR: 4.98e-04 | 7372.1ms/step | 8.1 samples/s
313
+ Step 2980/50462 | Loss: 3.1047 | LR: 4.98e-04 | 7366.9ms/step | 8.1 samples/s
314
+ Step 2990/50462 | Loss: 3.1069 | LR: 4.97e-04 | 7360.7ms/step | 8.2 samples/s
315
+ Step 3000/50462 | Loss: 3.0927 | LR: 4.97e-04 | 7360.1ms/step | 8.2 samples/s
316
+ Validation at step 3000: Loss = 3.1122, Perplexity = 22.47
317
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_mixed_seed70/run_20260822_210107/checkpoint-3000
318
+ Step 3010/50462 | Loss: 3.0770 | LR: 4.97e-04 | 12146.7ms/step | 4.9 samples/s
319
+ Step 3020/50462 | Loss: 3.0881 | LR: 4.97e-04 | 7369.0ms/step | 8.1 samples/s
320
+ Step 3030/50462 | Loss: 3.1111 | LR: 4.97e-04 | 7362.5ms/step | 8.1 samples/s
321
+ Step 3040/50462 | Loss: 3.1004 | LR: 4.97e-04 | 7355.5ms/step | 8.2 samples/s
322
+ Step 3050/50462 | Loss: 3.0841 | LR: 4.97e-04 | 7369.9ms/step | 8.1 samples/s
323
+ Step 3060/50462 | Loss: 3.1183 | LR: 4.97e-04 | 7351.9ms/step | 8.2 samples/s
324
+ Step 3070/50462 | Loss: 3.1183 | LR: 4.97e-04 | 7372.4ms/step | 8.1 samples/s
325
+ Step 3080/50462 | Loss: 3.0967 | LR: 4.97e-04 | 7378.5ms/step | 8.1 samples/s
326
+ Step 3090/50462 | Loss: 3.1079 | LR: 4.97e-04 | 7395.7ms/step | 8.1 samples/s
327
+ Step 3100/50462 | Loss: 3.1020 | LR: 4.97e-04 | 7380.6ms/step | 8.1 samples/s
328
+ Step 3110/50462 | Loss: 3.0934 | LR: 4.97e-04 | 7358.4ms/step | 8.2 samples/s
329
+ Step 3120/50462 | Loss: 3.0919 | LR: 4.97e-04 | 7363.8ms/step | 8.1 samples/s
330
+ Step 3130/50462 | Loss: 3.0860 | LR: 4.97e-04 | 7368.5ms/step | 8.1 samples/s
331
+ Step 3140/50462 | Loss: 3.1037 | LR: 4.97e-04 | 7366.7ms/step | 8.1 samples/s
332
+ Step 3150/50462 | Loss: 3.0955 | LR: 4.97e-04 | 7369.2ms/step | 8.1 samples/s
333
+ Step 3160/50462 | Loss: 3.0900 | LR: 4.97e-04 | 7368.1ms/step | 8.1 samples/s
334
+ Step 3170/50462 | Loss: 3.0800 | LR: 4.97e-04 | 7368.7ms/step | 8.1 samples/s
335
+ Step 3180/50462 | Loss: 3.0779 | LR: 4.97e-04 | 7364.3ms/step | 8.1 samples/s
336
+ Step 3190/50462 | Loss: 3.0759 | LR: 4.97e-04 | 7358.5ms/step | 8.2 samples/s
337
+ Step 3200/50462 | Loss: 3.0620 | LR: 4.97e-04 | 7349.9ms/step | 8.2 samples/s
338
+ Step 3210/50462 | Loss: 3.0839 | LR: 4.97e-04 | 7366.5ms/step | 8.1 samples/s
339
+ Step 3220/50462 | Loss: 3.0646 | LR: 4.97e-04 | 7389.4ms/step | 8.1 samples/s
340
+ Step 3230/50462 | Loss: 3.0610 | LR: 4.97e-04 | 7372.2ms/step | 8.1 samples/s
341
+ Step 3240/50462 | Loss: 3.0832 | LR: 4.97e-04 | 7349.6ms/step | 8.2 samples/s
342
+ Step 3250/50462 | Loss: 3.0886 | LR: 4.97e-04 | 7361.0ms/step | 8.2 samples/s
343
+ Validation at step 3250: Loss = 3.0823, Perplexity = 21.81
344
+ Step 3260/50462 | Loss: 3.0558 | LR: 4.97e-04 | 11233.7ms/step | 5.3 samples/s
345
+ Step 3270/50462 | Loss: 3.0566 | LR: 4.97e-04 | 7398.0ms/step | 8.1 samples/s
346
+ Step 3280/50462 | Loss: 3.0632 | LR: 4.97e-04 | 7358.3ms/step | 8.2 samples/s
347
+ Step 3290/50462 | Loss: 3.0667 | LR: 4.97e-04 | 7353.2ms/step | 8.2 samples/s
348
+ Step 3300/50462 | Loss: 3.0615 | LR: 4.97e-04 | 7352.4ms/step | 8.2 samples/s
349
+ Step 3310/50462 | Loss: 3.0545 | LR: 4.97e-04 | 7351.0ms/step | 8.2 samples/s
350
+ Step 3320/50462 | Loss: 3.0483 | LR: 4.97e-04 | 7390.7ms/step | 8.1 samples/s
epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/files/wandb-metadata.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-22T21:03:48.952755Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p2_1p2b_mixed_seed70_2gpu.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "h100-workspace-2-gpu-0-9",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 96,
21
+ "cpu_count_logical": 192,
22
+ "gpu": "NVIDIA H100 80GB HBM3",
23
+ "gpu_count": 2,
24
+ "disk": {
25
+ "/": {
26
+ "total": "9374931091456",
27
+ "used": "6420774977536"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1623005732864"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA H100 80GB HBM3",
36
+ "memoryTotal": "85520809984",
37
+ "cudaCores": 16896,
38
+ "architecture": "Hopper",
39
+ "uuid": "GPU-af79466d-4300-4171-b5bd-7697eca27264"
40
+ },
41
+ {
42
+ "name": "NVIDIA H100 80GB HBM3",
43
+ "memoryTotal": "85520809984",
44
+ "cudaCores": 16896,
45
+ "architecture": "Hopper",
46
+ "uuid": "GPU-0460dc7d-d83e-fd0f-7548-a3079803b79e"
47
+ }
48
+ ],
49
+ "cudaVersion": "13.0",
50
+ "writerId": "clb2ll0v1m1hv818bzb391b0hn7ensmv"
51
+ }
epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_setup.py:_flush():81] Configure stats pid to 4690
3
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_210348-i8ewyy95/logs/debug.log
5
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260822_210348-i8ewyy95/logs/debug-internal.log
6
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_mixed_seed70/run_20260822_210107', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 50462, 'per_device_batch_size': 2, 'gradient_accumulation_steps': 15, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 70, 'verbose': True, 'checkpoint_activations': True, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '1.2b mixed 8k seed70', 'wandb_tags': ['qwen3', '1.2b', '8k', 'p2', 'mixed', 'seed70'], 'hidden_size': 1536, 'intermediate_size': 6144, 'num_attention_heads': 12, 'num_key_value_heads': 4, 'num_hidden_layers': None, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 35, 'sparse_mask_pool_size': 10000, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': '/mloscratch/homes/navasard/sparse/mask_cache_8k_mixed', 'sparse_mask_seed': 70, 'sparse_mixed_nrb_set': [63, 47, 31, 21, 11], 'dense_mode': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-22 21:03:49,027 INFO MainThread:4690 [wandb_init.py:init():898] starting backend
10
+ 2026-08-22 21:03:49,306 INFO MainThread:4690 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-22 21:03:49,615 INFO MainThread:4690 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-22 21:03:49,617 INFO MainThread:4690 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-22 21:03:49,638 INFO MainThread:4690 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-22 21:03:50,169 INFO MainThread:4690 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-08-22 21:03:50,405 INFO MainThread:4690 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-08-22 21:03:50,405 INFO MainThread:4690 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-08-22 21:03:50,405 INFO MainThread:4690 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-08-22 21:03:50,406 INFO MainThread:4690 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-08-22 21:03:50,411 INFO MainThread:4690 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/run-20260822_210348-i8ewyy95/run-i8ewyy95.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5faf62872164f62ff7093cf1309689deb3ffb1a7a7bcc372fce2efca315ad318
3
+ size 5079040
epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/files/output.log ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ Starting training...
2
+ Switching attention mode for step 1: dense (dense step interval)
epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/files/wandb-metadata.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-23T04:01:01.681514Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p2_1p2b_dense_seed70_4gpu.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "h100-workspace-4-gpu-3-0-9",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 96,
21
+ "cpu_count_logical": 192,
22
+ "gpu": "NVIDIA H100 80GB HBM3",
23
+ "gpu_count": 4,
24
+ "disk": {
25
+ "/": {
26
+ "total": "9371709865984",
27
+ "used": "6561452351488"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1623005732864"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA H100 80GB HBM3",
36
+ "memoryTotal": "85520809984",
37
+ "cudaCores": 16896,
38
+ "architecture": "Hopper",
39
+ "uuid": "GPU-9a903f43-9348-6fa6-bcf7-e12bd01803a7"
40
+ },
41
+ {
42
+ "name": "NVIDIA H100 80GB HBM3",
43
+ "memoryTotal": "85520809984",
44
+ "cudaCores": 16896,
45
+ "architecture": "Hopper",
46
+ "uuid": "GPU-d0294574-a180-e984-6ca3-0573e9724f2e"
47
+ },
48
+ {
49
+ "name": "NVIDIA H100 80GB HBM3",
50
+ "memoryTotal": "85520809984",
51
+ "cudaCores": 16896,
52
+ "architecture": "Hopper",
53
+ "uuid": "GPU-24b8d952-dc08-33b9-d5df-54ce871fd288"
54
+ },
55
+ {
56
+ "name": "NVIDIA H100 80GB HBM3",
57
+ "memoryTotal": "85520809984",
58
+ "cudaCores": 16896,
59
+ "architecture": "Hopper",
60
+ "uuid": "GPU-b87e8111-63b6-b364-28b9-c4ce747aec08"
61
+ }
62
+ ],
63
+ "cudaVersion": "13.0",
64
+ "writerId": "5xn9247xpp7bkytaub0n7rw61p5n7iwz"
65
+ }
epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/logs/debug-internal.log ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-08-23T04:01:02.004539028Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-08-23T04:01:02.005119234Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-08-23T04:01:02.334253575Z","level":"INFO","msg":"stream: created new stream","id":"y8clx6eu"}
4
+ {"time":"2026-08-23T04:01:02.334373184Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-08-23T04:01:02.336722983Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-08-23T04:01:02.336795701Z","level":"INFO","msg":"writer: started","stream_id":"y8clx6eu"}
7
+ {"time":"2026-08-23T04:01:02.336846818Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-08-23T04:01:02.892387047Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-08-23T04:01:03.150807079Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-08-23T04:01:17.892692579Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":1,"console_offset":1,"console_lines":1,"uploaded_len":2}
11
+ {"time":"2026-08-23T04:01:18.129623503Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
12
+ {"time":"2026-08-23T04:01:32.89303166Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":1,"events_lines":2}
13
+ {"time":"2026-08-23T04:01:33.117742527Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-23 04:01:01,711 INFO MainThread:4320 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-23 04:01:01,711 INFO MainThread:4320 [wandb_setup.py:_flush():81] Configure stats pid to 4320
3
+ 2026-08-23 04:01:01,712 INFO MainThread:4320 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-23 04:01:01,712 INFO MainThread:4320 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260823_040101-y8clx6eu/logs/debug.log
5
+ 2026-08-23 04:01:01,712 INFO MainThread:4320 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260823_040101-y8clx6eu/logs/debug-internal.log
6
+ 2026-08-23 04:01:01,712 INFO MainThread:4320 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-23 04:01:01,712 INFO MainThread:4320 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_040030', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 50462, 'per_device_batch_size': 1, 'gradient_accumulation_steps': 15, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 70, 'verbose': True, 'checkpoint_activations': False, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '1.2b dense 8k seed70', 'wandb_tags': ['qwen3', '1.2b', '8k', 'p2', 'dense', 'seed70'], 'hidden_size': 1536, 'intermediate_size': 6144, 'num_attention_heads': 12, 'num_key_value_heads': 4, 'num_hidden_layers': 28, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 31, 'sparse_mask_pool_size': 1, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': None, 'sparse_mask_seed': 0, 'sparse_mixed_nrb_set': None, 'dense_mode': True, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-23 04:01:01,712 INFO MainThread:4320 [wandb_init.py:init():898] starting backend
10
+ 2026-08-23 04:01:01,989 INFO MainThread:4320 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-23 04:01:02,337 INFO MainThread:4320 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-23 04:01:02,342 INFO MainThread:4320 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-23 04:01:02,359 INFO MainThread:4320 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-23 04:01:02,695 INFO MainThread:4320 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-08-23 04:01:02,879 INFO MainThread:4320 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-08-23 04:01:02,879 INFO MainThread:4320 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-08-23 04:01:02,879 INFO MainThread:4320 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-08-23 04:01:02,879 INFO MainThread:4320 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-08-23 04:01:02,883 INFO MainThread:4320 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/run-20260823_040101-y8clx6eu/run-y8clx6eu.wandb ADDED
Binary file (7 Bytes). View file
 
epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/files/output.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/files/wandb-metadata.json ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-23T04:11:48.465653Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p2_1p2b_mixed_seed70_6gpu.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "h100-train-6-gpu-0-31",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 96,
21
+ "cpu_count_logical": 192,
22
+ "gpu": "NVIDIA H100 80GB HBM3",
23
+ "gpu_count": 6,
24
+ "disk": {
25
+ "/": {
26
+ "total": "9371709865984",
27
+ "used": "6560433696768"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1623005732864"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA H100 80GB HBM3",
36
+ "memoryTotal": "85520809984",
37
+ "cudaCores": 16896,
38
+ "architecture": "Hopper",
39
+ "uuid": "GPU-98029aa4-c589-b65b-3ab7-6495256f31ed"
40
+ },
41
+ {
42
+ "name": "NVIDIA H100 80GB HBM3",
43
+ "memoryTotal": "85520809984",
44
+ "cudaCores": 16896,
45
+ "architecture": "Hopper",
46
+ "uuid": "GPU-37a1c6c2-d3df-3064-44df-33fa82304bd0"
47
+ },
48
+ {
49
+ "name": "NVIDIA H100 80GB HBM3",
50
+ "memoryTotal": "85520809984",
51
+ "cudaCores": 16896,
52
+ "architecture": "Hopper",
53
+ "uuid": "GPU-9a903f43-9348-6fa6-bcf7-e12bd01803a7"
54
+ },
55
+ {
56
+ "name": "NVIDIA H100 80GB HBM3",
57
+ "memoryTotal": "85520809984",
58
+ "cudaCores": 16896,
59
+ "architecture": "Hopper",
60
+ "uuid": "GPU-d0294574-a180-e984-6ca3-0573e9724f2e"
61
+ },
62
+ {
63
+ "name": "NVIDIA H100 80GB HBM3",
64
+ "memoryTotal": "85520809984",
65
+ "cudaCores": 16896,
66
+ "architecture": "Hopper",
67
+ "uuid": "GPU-24b8d952-dc08-33b9-d5df-54ce871fd288"
68
+ },
69
+ {
70
+ "name": "NVIDIA H100 80GB HBM3",
71
+ "memoryTotal": "85520809984",
72
+ "cudaCores": 16896,
73
+ "architecture": "Hopper",
74
+ "uuid": "GPU-b87e8111-63b6-b364-28b9-c4ce747aec08"
75
+ }
76
+ ],
77
+ "cudaVersion": "13.0",
78
+ "writerId": "xh8q4c30086d52gf48fc191wcdie0cdj"
79
+ }
epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/logs/debug.log ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_setup.py:_flush():81] Configure stats pid to 217
3
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260823_041148-i8ewyy95/logs/debug.log
5
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260823_041148-i8ewyy95/logs/debug-internal.log
6
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_mixed_seed70/run_20260822_210107', 'resume_from_checkpoint': '/mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_mixed_seed70/run_20260822_210107/checkpoint-3000', 'force_new_run': False, 'num_epochs': 1, 'max_steps': 50462, 'per_device_batch_size': 1, 'gradient_accumulation_steps': 10, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 250, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 70, 'verbose': True, 'checkpoint_activations': True, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '1.2b mixed 8k seed70 6gpu resume', 'wandb_tags': ['qwen3', '1.2b', '8k', 'p2', 'mixed', 'seed70'], 'hidden_size': 1536, 'intermediate_size': 6144, 'num_attention_heads': 12, 'num_key_value_heads': 4, 'num_hidden_layers': None, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 35, 'sparse_mask_pool_size': 10000, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': '/mloscratch/homes/navasard/sparse/mask_cache_8k_mixed', 'sparse_mask_seed': 70, 'sparse_mixed_nrb_set': [63, 47, 31, 21, 11], 'dense_mode': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-23 04:11:48,490 INFO MainThread:217 [wandb_init.py:init():898] starting backend
10
+ 2026-08-23 04:11:49,135 INFO MainThread:217 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-23 04:11:49,695 INFO MainThread:217 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-23 04:11:49,700 INFO MainThread:217 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-23 04:11:49,737 INFO MainThread:217 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-23 04:11:50,209 INFO MainThread:217 [wandb_init.py:init():1051] run resumed
15
+ 2026-08-23 04:11:50,212 INFO MainThread:217 [wandb_init.py:init():1056] starting run threads in backend
16
+ 2026-08-23 04:11:50,469 INFO MainThread:217 [wandb_run.py:_console_start():2554] atexit reg
17
+ 2026-08-23 04:11:50,469 INFO MainThread:217 [wandb_run.py:_redirect():2403] redirect: wrap_raw
18
+ 2026-08-23 04:11:50,469 INFO MainThread:217 [wandb_run.py:_redirect():2472] Wrapping output streams.
19
+ 2026-08-23 04:11:50,469 INFO MainThread:217 [wandb_run.py:_redirect():2495] Redirects installed.
20
+ 2026-08-23 04:11:50,477 INFO MainThread:217 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/run-20260823_041148-i8ewyy95/run-i8ewyy95.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c560dc13b2b59868404a464767df09e0e8d758e3eaf372b8cc66bf4bf70352bd
3
+ size 17661952
epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/files/output.log ADDED
@@ -0,0 +1,526 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Starting training...
2
+ Switching attention mode for step 1: dense (dense step interval)
3
+ Step 10/50462 | Loss: 12.2299 | LR: 6.67e-06 | 13397.9ms/step | 4.5 samples/s
4
+ Step 20/50462 | Loss: 11.9156 | LR: 1.33e-05 | 6965.1ms/step | 8.6 samples/s
5
+ Step 30/50462 | Loss: 11.0080 | LR: 2.00e-05 | 6966.3ms/step | 8.6 samples/s
6
+ Step 40/50462 | Loss: 10.1412 | LR: 2.67e-05 | 6953.0ms/step | 8.6 samples/s
7
+ Step 50/50462 | Loss: 9.7365 | LR: 3.33e-05 | 6952.0ms/step | 8.6 samples/s
8
+ Step 60/50462 | Loss: 9.2596 | LR: 4.00e-05 | 6956.4ms/step | 8.6 samples/s
9
+ Step 70/50462 | Loss: 8.7069 | LR: 4.67e-05 | 6960.8ms/step | 8.6 samples/s
10
+ Step 80/50462 | Loss: 8.2501 | LR: 5.33e-05 | 6957.5ms/step | 8.6 samples/s
11
+ Step 90/50462 | Loss: 7.8702 | LR: 6.00e-05 | 6961.8ms/step | 8.6 samples/s
12
+ Step 100/50462 | Loss: 7.4724 | LR: 6.67e-05 | 6967.9ms/step | 8.6 samples/s
13
+ Step 110/50462 | Loss: 7.2534 | LR: 7.33e-05 | 6966.4ms/step | 8.6 samples/s
14
+ Step 120/50462 | Loss: 7.0820 | LR: 8.00e-05 | 6971.5ms/step | 8.6 samples/s
15
+ Step 130/50462 | Loss: 6.9967 | LR: 8.67e-05 | 6974.3ms/step | 8.6 samples/s
16
+ Step 140/50462 | Loss: 6.8992 | LR: 9.33e-05 | 6976.3ms/step | 8.6 samples/s
17
+ Step 150/50462 | Loss: 6.7704 | LR: 1.00e-04 | 6973.5ms/step | 8.6 samples/s
18
+ Step 160/50462 | Loss: 6.6579 | LR: 1.07e-04 | 6975.6ms/step | 8.6 samples/s
19
+ Step 170/50462 | Loss: 6.5643 | LR: 1.13e-04 | 6977.5ms/step | 8.6 samples/s
20
+ Step 180/50462 | Loss: 6.4492 | LR: 1.20e-04 | 6976.6ms/step | 8.6 samples/s
21
+ Step 190/50462 | Loss: 6.3763 | LR: 1.27e-04 | 6977.4ms/step | 8.6 samples/s
22
+ Step 200/50462 | Loss: 6.3030 | LR: 1.33e-04 | 6977.5ms/step | 8.6 samples/s
23
+ Step 210/50462 | Loss: 6.2139 | LR: 1.40e-04 | 6977.2ms/step | 8.6 samples/s
24
+ Step 220/50462 | Loss: 6.2008 | LR: 1.47e-04 | 6976.3ms/step | 8.6 samples/s
25
+ Step 230/50462 | Loss: 6.1205 | LR: 1.53e-04 | 6977.5ms/step | 8.6 samples/s
26
+ Step 240/50462 | Loss: 6.0458 | LR: 1.60e-04 | 6976.5ms/step | 8.6 samples/s
27
+ Step 250/50462 | Loss: 6.0234 | LR: 1.67e-04 | 6978.1ms/step | 8.6 samples/s
28
+ Validation at step 250: Loss = 5.9605, Perplexity = 387.82
29
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-250
30
+ Step 260/50462 | Loss: 5.9211 | LR: 1.73e-04 | 14290.4ms/step | 4.2 samples/s
31
+ Step 270/50462 | Loss: 5.8880 | LR: 1.80e-04 | 6979.7ms/step | 8.6 samples/s
32
+ Step 280/50462 | Loss: 5.8289 | LR: 1.87e-04 | 6978.4ms/step | 8.6 samples/s
33
+ Step 290/50462 | Loss: 5.7598 | LR: 1.93e-04 | 6983.4ms/step | 8.6 samples/s
34
+ Step 300/50462 | Loss: 5.7123 | LR: 2.00e-04 | 6978.2ms/step | 8.6 samples/s
35
+ Step 310/50462 | Loss: 5.6371 | LR: 2.07e-04 | 6974.8ms/step | 8.6 samples/s
36
+ Step 320/50462 | Loss: 5.6010 | LR: 2.13e-04 | 6974.8ms/step | 8.6 samples/s
37
+ Step 330/50462 | Loss: 5.5710 | LR: 2.20e-04 | 6978.6ms/step | 8.6 samples/s
38
+ Step 340/50462 | Loss: 5.5359 | LR: 2.27e-04 | 6974.4ms/step | 8.6 samples/s
39
+ Step 350/50462 | Loss: 5.4892 | LR: 2.33e-04 | 6973.2ms/step | 8.6 samples/s
40
+ Step 360/50462 | Loss: 5.4533 | LR: 2.40e-04 | 6972.2ms/step | 8.6 samples/s
41
+ Step 370/50462 | Loss: 5.3670 | LR: 2.47e-04 | 6974.5ms/step | 8.6 samples/s
42
+ Step 380/50462 | Loss: 5.3213 | LR: 2.53e-04 | 6973.7ms/step | 8.6 samples/s
43
+ Step 390/50462 | Loss: 5.2728 | LR: 2.60e-04 | 6976.0ms/step | 8.6 samples/s
44
+ Step 400/50462 | Loss: 5.2346 | LR: 2.67e-04 | 6992.1ms/step | 8.6 samples/s
45
+ Step 410/50462 | Loss: 5.2008 | LR: 2.73e-04 | 7002.5ms/step | 8.6 samples/s
46
+ Step 420/50462 | Loss: 5.1574 | LR: 2.80e-04 | 7002.8ms/step | 8.6 samples/s
47
+ Step 430/50462 | Loss: 5.1082 | LR: 2.87e-04 | 7004.1ms/step | 8.6 samples/s
48
+ Step 440/50462 | Loss: 5.0639 | LR: 2.93e-04 | 7001.7ms/step | 8.6 samples/s
49
+ Step 450/50462 | Loss: 5.0605 | LR: 3.00e-04 | 6992.6ms/step | 8.6 samples/s
50
+ Step 460/50462 | Loss: 5.0267 | LR: 3.07e-04 | 7001.6ms/step | 8.6 samples/s
51
+ Step 470/50462 | Loss: 4.9638 | LR: 3.13e-04 | 6999.0ms/step | 8.6 samples/s
52
+ Step 480/50462 | Loss: 4.9458 | LR: 3.20e-04 | 6985.4ms/step | 8.6 samples/s
53
+ Step 490/50462 | Loss: 4.8491 | LR: 3.27e-04 | 6980.0ms/step | 8.6 samples/s
54
+ Step 500/50462 | Loss: 4.8597 | LR: 3.33e-04 | 6981.3ms/step | 8.6 samples/s
55
+ Validation at step 500: Loss = 4.8557, Perplexity = 128.48
56
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-500
57
+ Step 510/50462 | Loss: 4.8445 | LR: 3.40e-04 | 10743.5ms/step | 5.6 samples/s
58
+ Step 520/50462 | Loss: 4.7613 | LR: 3.47e-04 | 6994.9ms/step | 8.6 samples/s
59
+ Step 530/50462 | Loss: 4.7293 | LR: 3.53e-04 | 6997.9ms/step | 8.6 samples/s
60
+ Step 540/50462 | Loss: 4.7942 | LR: 3.60e-04 | 6991.0ms/step | 8.6 samples/s
61
+ Step 550/50462 | Loss: 4.7975 | LR: 3.67e-04 | 6995.8ms/step | 8.6 samples/s
62
+ Step 560/50462 | Loss: 4.7114 | LR: 3.73e-04 | 6991.5ms/step | 8.6 samples/s
63
+ Step 570/50462 | Loss: 4.6380 | LR: 3.80e-04 | 6992.5ms/step | 8.6 samples/s
64
+ Step 580/50462 | Loss: 4.6088 | LR: 3.87e-04 | 6995.6ms/step | 8.6 samples/s
65
+ Step 590/50462 | Loss: 4.6519 | LR: 3.93e-04 | 6993.0ms/step | 8.6 samples/s
66
+ Step 600/50462 | Loss: 4.5644 | LR: 4.00e-04 | 6992.5ms/step | 8.6 samples/s
67
+ Step 610/50462 | Loss: 4.5056 | LR: 4.07e-04 | 6989.8ms/step | 8.6 samples/s
68
+ Step 620/50462 | Loss: 4.5163 | LR: 4.13e-04 | 6991.3ms/step | 8.6 samples/s
69
+ Step 630/50462 | Loss: 4.5185 | LR: 4.20e-04 | 6993.7ms/step | 8.6 samples/s
70
+ Step 640/50462 | Loss: 4.4487 | LR: 4.27e-04 | 6991.2ms/step | 8.6 samples/s
71
+ Step 650/50462 | Loss: 4.3603 | LR: 4.33e-04 | 6993.4ms/step | 8.6 samples/s
72
+ Step 660/50462 | Loss: 4.3409 | LR: 4.40e-04 | 6994.4ms/step | 8.6 samples/s
73
+ Step 670/50462 | Loss: 4.4596 | LR: 4.47e-04 | 6992.8ms/step | 8.6 samples/s
74
+ Step 680/50462 | Loss: 4.3716 | LR: 4.53e-04 | 6994.4ms/step | 8.6 samples/s
75
+ Step 690/50462 | Loss: 4.2619 | LR: 4.60e-04 | 6997.6ms/step | 8.6 samples/s
76
+ Step 700/50462 | Loss: 4.2409 | LR: 4.67e-04 | 6991.7ms/step | 8.6 samples/s
77
+ Step 710/50462 | Loss: 4.2600 | LR: 4.73e-04 | 6990.2ms/step | 8.6 samples/s
78
+ Step 720/50462 | Loss: 4.2935 | LR: 4.80e-04 | 6996.9ms/step | 8.6 samples/s
79
+ Step 730/50462 | Loss: 4.2164 | LR: 4.87e-04 | 6997.2ms/step | 8.6 samples/s
80
+ Step 740/50462 | Loss: 4.1447 | LR: 4.93e-04 | 6999.5ms/step | 8.6 samples/s
81
+ Step 750/50462 | Loss: 4.1294 | LR: 5.00e-04 | 6998.5ms/step | 8.6 samples/s
82
+ Validation at step 750: Loss = 4.1136, Perplexity = 61.17
83
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-750
84
+ Step 760/50462 | Loss: 4.0937 | LR: 5.00e-04 | 10758.6ms/step | 5.6 samples/s
85
+ Step 770/50462 | Loss: 4.1263 | LR: 5.00e-04 | 6986.5ms/step | 8.6 samples/s
86
+ Step 780/50462 | Loss: 4.0728 | LR: 5.00e-04 | 6989.5ms/step | 8.6 samples/s
87
+ Step 790/50462 | Loss: 4.0538 | LR: 5.00e-04 | 6994.4ms/step | 8.6 samples/s
88
+ Step 800/50462 | Loss: 4.0512 | LR: 5.00e-04 | 6995.7ms/step | 8.6 samples/s
89
+ Step 810/50462 | Loss: 3.9448 | LR: 5.00e-04 | 6991.6ms/step | 8.6 samples/s
90
+ Step 820/50462 | Loss: 4.0090 | LR: 5.00e-04 | 6996.2ms/step | 8.6 samples/s
91
+ Step 830/50462 | Loss: 3.9644 | LR: 5.00e-04 | 6984.1ms/step | 8.6 samples/s
92
+ Step 840/50462 | Loss: 3.9260 | LR: 5.00e-04 | 6986.2ms/step | 8.6 samples/s
93
+ Step 850/50462 | Loss: 3.9136 | LR: 5.00e-04 | 6989.1ms/step | 8.6 samples/s
94
+ Step 860/50462 | Loss: 3.8627 | LR: 5.00e-04 | 6981.9ms/step | 8.6 samples/s
95
+ Step 870/50462 | Loss: 3.8631 | LR: 5.00e-04 | 6974.7ms/step | 8.6 samples/s
96
+ Step 880/50462 | Loss: 3.8533 | LR: 5.00e-04 | 6977.1ms/step | 8.6 samples/s
97
+ Step 890/50462 | Loss: 3.8499 | LR: 5.00e-04 | 6977.7ms/step | 8.6 samples/s
98
+ Step 900/50462 | Loss: 3.8223 | LR: 5.00e-04 | 6974.0ms/step | 8.6 samples/s
99
+ Step 910/50462 | Loss: 3.8381 | LR: 5.00e-04 | 6973.0ms/step | 8.6 samples/s
100
+ Step 920/50462 | Loss: 3.8187 | LR: 5.00e-04 | 6972.2ms/step | 8.6 samples/s
101
+ Step 930/50462 | Loss: 3.7817 | LR: 5.00e-04 | 6973.8ms/step | 8.6 samples/s
102
+ Step 940/50462 | Loss: 3.8183 | LR: 5.00e-04 | 6976.2ms/step | 8.6 samples/s
103
+ Step 950/50462 | Loss: 3.7734 | LR: 5.00e-04 | 6972.4ms/step | 8.6 samples/s
104
+ Step 960/50462 | Loss: 3.7546 | LR: 5.00e-04 | 6980.5ms/step | 8.6 samples/s
105
+ Step 970/50462 | Loss: 3.7664 | LR: 5.00e-04 | 6997.3ms/step | 8.6 samples/s
106
+ Step 980/50462 | Loss: 3.7394 | LR: 5.00e-04 | 7001.2ms/step | 8.6 samples/s
107
+ Step 990/50462 | Loss: 3.7606 | LR: 5.00e-04 | 7003.0ms/step | 8.6 samples/s
108
+ Step 1000/50462 | Loss: 3.7216 | LR: 5.00e-04 | 7003.4ms/step | 8.6 samples/s
109
+ Validation at step 1000: Loss = 3.6926, Perplexity = 40.15
110
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-1000
111
+ Step 1010/50462 | Loss: 3.6989 | LR: 5.00e-04 | 10784.3ms/step | 5.6 samples/s
112
+ Step 1020/50462 | Loss: 3.6964 | LR: 5.00e-04 | 7000.7ms/step | 8.6 samples/s
113
+ Step 1030/50462 | Loss: 3.6747 | LR: 5.00e-04 | 6995.7ms/step | 8.6 samples/s
114
+ Step 1040/50462 | Loss: 3.6733 | LR: 5.00e-04 | 6981.7ms/step | 8.6 samples/s
115
+ Step 1050/50462 | Loss: 3.6548 | LR: 5.00e-04 | 6976.3ms/step | 8.6 samples/s
116
+ Step 1060/50462 | Loss: 3.6781 | LR: 5.00e-04 | 6978.6ms/step | 8.6 samples/s
117
+ Step 1070/50462 | Loss: 3.6458 | LR: 5.00e-04 | 6984.2ms/step | 8.6 samples/s
118
+ Step 1080/50462 | Loss: 3.6464 | LR: 5.00e-04 | 6995.5ms/step | 8.6 samples/s
119
+ Step 1090/50462 | Loss: 3.6192 | LR: 5.00e-04 | 6999.7ms/step | 8.6 samples/s
120
+ Step 1100/50462 | Loss: 3.6117 | LR: 5.00e-04 | 6995.7ms/step | 8.6 samples/s
121
+ Step 1110/50462 | Loss: 3.6503 | LR: 5.00e-04 | 6990.1ms/step | 8.6 samples/s
122
+ Step 1120/50462 | Loss: 3.5846 | LR: 5.00e-04 | 6988.8ms/step | 8.6 samples/s
123
+ Step 1130/50462 | Loss: 3.5687 | LR: 5.00e-04 | 6988.1ms/step | 8.6 samples/s
124
+ Step 1140/50462 | Loss: 3.5888 | LR: 5.00e-04 | 6989.8ms/step | 8.6 samples/s
125
+ Step 1150/50462 | Loss: 3.5869 | LR: 5.00e-04 | 6985.9ms/step | 8.6 samples/s
126
+ Step 1160/50462 | Loss: 3.5741 | LR: 5.00e-04 | 6986.5ms/step | 8.6 samples/s
127
+ Step 1170/50462 | Loss: 3.5614 | LR: 5.00e-04 | 6986.4ms/step | 8.6 samples/s
128
+ Step 1180/50462 | Loss: 3.5469 | LR: 5.00e-04 | 6987.6ms/step | 8.6 samples/s
129
+ Step 1190/50462 | Loss: 3.5895 | LR: 5.00e-04 | 6986.5ms/step | 8.6 samples/s
130
+ Step 1200/50462 | Loss: 3.4897 | LR: 5.00e-04 | 6987.6ms/step | 8.6 samples/s
131
+ Step 1210/50462 | Loss: 3.5457 | LR: 5.00e-04 | 6985.7ms/step | 8.6 samples/s
132
+ Step 1220/50462 | Loss: 3.5586 | LR: 5.00e-04 | 6985.5ms/step | 8.6 samples/s
133
+ Step 1230/50462 | Loss: 3.5576 | LR: 5.00e-04 | 6987.6ms/step | 8.6 samples/s
134
+ Step 1240/50462 | Loss: 3.5320 | LR: 5.00e-04 | 6987.8ms/step | 8.6 samples/s
135
+ Step 1250/50462 | Loss: 3.5228 | LR: 5.00e-04 | 6986.3ms/step | 8.6 samples/s
136
+ Validation at step 1250: Loss = 3.4897, Perplexity = 32.78
137
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-1250
138
+ Step 1260/50462 | Loss: 3.4950 | LR: 5.00e-04 | 10757.8ms/step | 5.6 samples/s
139
+ Step 1270/50462 | Loss: 3.4530 | LR: 5.00e-04 | 6984.5ms/step | 8.6 samples/s
140
+ Step 1280/50462 | Loss: 3.5241 | LR: 5.00e-04 | 6988.9ms/step | 8.6 samples/s
141
+ Step 1290/50462 | Loss: 3.4532 | LR: 5.00e-04 | 6986.1ms/step | 8.6 samples/s
142
+ Step 1300/50462 | Loss: 3.4954 | LR: 5.00e-04 | 6988.0ms/step | 8.6 samples/s
143
+ Step 1310/50462 | Loss: 3.5101 | LR: 5.00e-04 | 6986.1ms/step | 8.6 samples/s
144
+ Step 1320/50462 | Loss: 3.4786 | LR: 5.00e-04 | 6989.8ms/step | 8.6 samples/s
145
+ Step 1330/50462 | Loss: 3.4448 | LR: 5.00e-04 | 6984.2ms/step | 8.6 samples/s
146
+ Step 1340/50462 | Loss: 3.4296 | LR: 5.00e-04 | 6989.5ms/step | 8.6 samples/s
147
+ Step 1350/50462 | Loss: 3.4514 | LR: 5.00e-04 | 6988.0ms/step | 8.6 samples/s
148
+ Step 1360/50462 | Loss: 3.4206 | LR: 5.00e-04 | 6985.8ms/step | 8.6 samples/s
149
+ Step 1370/50462 | Loss: 3.4638 | LR: 5.00e-04 | 6990.2ms/step | 8.6 samples/s
150
+ Step 1380/50462 | Loss: 3.4280 | LR: 5.00e-04 | 6986.4ms/step | 8.6 samples/s
151
+ Step 1390/50462 | Loss: 3.3971 | LR: 5.00e-04 | 6987.5ms/step | 8.6 samples/s
152
+ Step 1400/50462 | Loss: 3.4146 | LR: 5.00e-04 | 6987.4ms/step | 8.6 samples/s
153
+ Step 1410/50462 | Loss: 3.4468 | LR: 5.00e-04 | 6980.2ms/step | 8.6 samples/s
154
+ Step 1420/50462 | Loss: 3.4424 | LR: 5.00e-04 | 6979.3ms/step | 8.6 samples/s
155
+ Step 1430/50462 | Loss: 3.4211 | LR: 5.00e-04 | 6980.1ms/step | 8.6 samples/s
156
+ Step 1440/50462 | Loss: 3.4276 | LR: 5.00e-04 | 6978.7ms/step | 8.6 samples/s
157
+ Step 1450/50462 | Loss: 3.3767 | LR: 5.00e-04 | 6978.1ms/step | 8.6 samples/s
158
+ Step 1460/50462 | Loss: 3.3895 | LR: 5.00e-04 | 6975.8ms/step | 8.6 samples/s
159
+ Step 1470/50462 | Loss: 3.4021 | LR: 5.00e-04 | 6977.1ms/step | 8.6 samples/s
160
+ Step 1480/50462 | Loss: 3.3726 | LR: 5.00e-04 | 6975.9ms/step | 8.6 samples/s
161
+ Step 1490/50462 | Loss: 3.4028 | LR: 5.00e-04 | 6975.7ms/step | 8.6 samples/s
162
+ Step 1500/50462 | Loss: 3.3659 | LR: 5.00e-04 | 6976.2ms/step | 8.6 samples/s
163
+ Validation at step 1500: Loss = 3.3674, Perplexity = 29.00
164
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-1500
165
+ Step 1510/50462 | Loss: 3.3810 | LR: 5.00e-04 | 10759.7ms/step | 5.6 samples/s
166
+ Step 1520/50462 | Loss: 3.3442 | LR: 5.00e-04 | 7000.4ms/step | 8.6 samples/s
167
+ Step 1530/50462 | Loss: 3.3526 | LR: 5.00e-04 | 7016.7ms/step | 8.6 samples/s
168
+ Step 1540/50462 | Loss: 3.3672 | LR: 5.00e-04 | 7020.0ms/step | 8.5 samples/s
169
+ Step 1550/50462 | Loss: 3.3788 | LR: 5.00e-04 | 7018.8ms/step | 8.5 samples/s
170
+ Step 1560/50462 | Loss: 3.3520 | LR: 5.00e-04 | 7002.4ms/step | 8.6 samples/s
171
+ Step 1570/50462 | Loss: 3.3618 | LR: 5.00e-04 | 7009.3ms/step | 8.6 samples/s
172
+ Step 1580/50462 | Loss: 3.3222 | LR: 5.00e-04 | 7015.6ms/step | 8.6 samples/s
173
+ Step 1590/50462 | Loss: 3.3415 | LR: 5.00e-04 | 6992.0ms/step | 8.6 samples/s
174
+ Step 1600/50462 | Loss: 3.3620 | LR: 5.00e-04 | 6991.1ms/step | 8.6 samples/s
175
+ Step 1610/50462 | Loss: 3.3402 | LR: 5.00e-04 | 6984.7ms/step | 8.6 samples/s
176
+ Step 1620/50462 | Loss: 3.3494 | LR: 5.00e-04 | 6991.9ms/step | 8.6 samples/s
177
+ Step 1630/50462 | Loss: 3.3329 | LR: 5.00e-04 | 7008.2ms/step | 8.6 samples/s
178
+ Step 1640/50462 | Loss: 3.3089 | LR: 5.00e-04 | 7019.5ms/step | 8.5 samples/s
179
+ Step 1650/50462 | Loss: 3.3211 | LR: 5.00e-04 | 7009.9ms/step | 8.6 samples/s
180
+ Step 1660/50462 | Loss: 3.3309 | LR: 5.00e-04 | 7007.8ms/step | 8.6 samples/s
181
+ Step 1670/50462 | Loss: 3.2594 | LR: 5.00e-04 | 7007.3ms/step | 8.6 samples/s
182
+ Step 1680/50462 | Loss: 3.3242 | LR: 5.00e-04 | 7004.2ms/step | 8.6 samples/s
183
+ Step 1690/50462 | Loss: 3.3271 | LR: 5.00e-04 | 6994.7ms/step | 8.6 samples/s
184
+ Step 1700/50462 | Loss: 3.2683 | LR: 5.00e-04 | 6996.2ms/step | 8.6 samples/s
185
+ Step 1710/50462 | Loss: 3.2889 | LR: 5.00e-04 | 6994.8ms/step | 8.6 samples/s
186
+ Step 1720/50462 | Loss: 3.2799 | LR: 5.00e-04 | 6991.3ms/step | 8.6 samples/s
187
+ Step 1730/50462 | Loss: 3.2937 | LR: 5.00e-04 | 6992.7ms/step | 8.6 samples/s
188
+ Step 1740/50462 | Loss: 3.2602 | LR: 5.00e-04 | 6992.2ms/step | 8.6 samples/s
189
+ Step 1750/50462 | Loss: 3.2789 | LR: 5.00e-04 | 6999.4ms/step | 8.6 samples/s
190
+ Validation at step 1750: Loss = 3.2792, Perplexity = 26.55
191
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-1750
192
+ Step 1760/50462 | Loss: 3.2848 | LR: 4.99e-04 | 10795.4ms/step | 5.6 samples/s
193
+ Step 1770/50462 | Loss: 3.2934 | LR: 4.99e-04 | 6996.0ms/step | 8.6 samples/s
194
+ Step 1780/50462 | Loss: 3.2656 | LR: 4.99e-04 | 7001.8ms/step | 8.6 samples/s
195
+ Step 1790/50462 | Loss: 3.2855 | LR: 4.99e-04 | 7008.4ms/step | 8.6 samples/s
196
+ Step 1800/50462 | Loss: 3.2852 | LR: 4.99e-04 | 7007.4ms/step | 8.6 samples/s
197
+ Step 1810/50462 | Loss: 3.3031 | LR: 4.99e-04 | 7002.4ms/step | 8.6 samples/s
198
+ Step 1820/50462 | Loss: 3.2804 | LR: 4.99e-04 | 7004.9ms/step | 8.6 samples/s
199
+ Step 1830/50462 | Loss: 3.2724 | LR: 4.99e-04 | 7004.9ms/step | 8.6 samples/s
200
+ Step 1840/50462 | Loss: 3.2435 | LR: 4.99e-04 | 6999.7ms/step | 8.6 samples/s
201
+ Step 1850/50462 | Loss: 3.2467 | LR: 4.99e-04 | 6999.1ms/step | 8.6 samples/s
202
+ Step 1860/50462 | Loss: 3.2667 | LR: 4.99e-04 | 7000.4ms/step | 8.6 samples/s
203
+ Step 1870/50462 | Loss: 3.2725 | LR: 4.99e-04 | 7002.4ms/step | 8.6 samples/s
204
+ Step 1880/50462 | Loss: 3.2516 | LR: 4.99e-04 | 7005.4ms/step | 8.6 samples/s
205
+ Step 1890/50462 | Loss: 3.2262 | LR: 4.99e-04 | 7006.3ms/step | 8.6 samples/s
206
+ Step 1900/50462 | Loss: 3.2665 | LR: 4.99e-04 | 7008.0ms/step | 8.6 samples/s
207
+ Step 1910/50462 | Loss: 3.2557 | LR: 4.99e-04 | 6999.6ms/step | 8.6 samples/s
208
+ Step 1920/50462 | Loss: 3.2344 | LR: 4.99e-04 | 7007.4ms/step | 8.6 samples/s
209
+ Step 1930/50462 | Loss: 3.2331 | LR: 4.99e-04 | 7006.3ms/step | 8.6 samples/s
210
+ Step 1940/50462 | Loss: 3.2540 | LR: 4.99e-04 | 7005.4ms/step | 8.6 samples/s
211
+ Step 1950/50462 | Loss: 3.2642 | LR: 4.99e-04 | 7003.3ms/step | 8.6 samples/s
212
+ Step 1960/50462 | Loss: 3.2328 | LR: 4.99e-04 | 7001.2ms/step | 8.6 samples/s
213
+ Step 1970/50462 | Loss: 3.2330 | LR: 4.99e-04 | 6995.3ms/step | 8.6 samples/s
214
+ Step 1980/50462 | Loss: 3.2341 | LR: 4.99e-04 | 6988.7ms/step | 8.6 samples/s
215
+ Step 1990/50462 | Loss: 3.2301 | LR: 4.99e-04 | 6986.6ms/step | 8.6 samples/s
216
+ Step 2000/50462 | Loss: 3.2461 | LR: 4.99e-04 | 6986.6ms/step | 8.6 samples/s
217
+ Validation at step 2000: Loss = 3.2130, Perplexity = 24.85
218
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-2000
219
+ Step 2010/50462 | Loss: 3.1880 | LR: 4.99e-04 | 10783.7ms/step | 5.6 samples/s
220
+ Step 2020/50462 | Loss: 3.2372 | LR: 4.99e-04 | 6986.3ms/step | 8.6 samples/s
221
+ Step 2030/50462 | Loss: 3.2083 | LR: 4.99e-04 | 6990.2ms/step | 8.6 samples/s
222
+ Step 2040/50462 | Loss: 3.2101 | LR: 4.99e-04 | 6988.6ms/step | 8.6 samples/s
223
+ Step 2050/50462 | Loss: 3.1618 | LR: 4.99e-04 | 6982.2ms/step | 8.6 samples/s
224
+ Step 2060/50462 | Loss: 3.2344 | LR: 4.99e-04 | 6984.4ms/step | 8.6 samples/s
225
+ Step 2070/50462 | Loss: 3.2079 | LR: 4.99e-04 | 7000.0ms/step | 8.6 samples/s
226
+ Step 2080/50462 | Loss: 3.2033 | LR: 4.99e-04 | 7013.4ms/step | 8.6 samples/s
227
+ Step 2090/50462 | Loss: 3.2283 | LR: 4.99e-04 | 7008.2ms/step | 8.6 samples/s
228
+ Step 2100/50462 | Loss: 3.2070 | LR: 4.99e-04 | 7010.9ms/step | 8.6 samples/s
229
+ Step 2110/50462 | Loss: 3.2140 | LR: 4.99e-04 | 7016.3ms/step | 8.6 samples/s
230
+ Step 2120/50462 | Loss: 3.2027 | LR: 4.99e-04 | 7006.6ms/step | 8.6 samples/s
231
+ Step 2130/50462 | Loss: 3.2023 | LR: 4.99e-04 | 7015.1ms/step | 8.6 samples/s
232
+ Step 2140/50462 | Loss: 3.2078 | LR: 4.99e-04 | 7003.2ms/step | 8.6 samples/s
233
+ Step 2150/50462 | Loss: 3.1806 | LR: 4.99e-04 | 6993.1ms/step | 8.6 samples/s
234
+ Step 2160/50462 | Loss: 3.2147 | LR: 4.99e-04 | 6990.6ms/step | 8.6 samples/s
235
+ Step 2170/50462 | Loss: 3.1855 | LR: 4.99e-04 | 6987.0ms/step | 8.6 samples/s
236
+ Step 2180/50462 | Loss: 3.2051 | LR: 4.99e-04 | 6994.4ms/step | 8.6 samples/s
237
+ Step 2190/50462 | Loss: 3.1860 | LR: 4.99e-04 | 7002.5ms/step | 8.6 samples/s
238
+ Step 2200/50462 | Loss: 3.1900 | LR: 4.99e-04 | 7005.0ms/step | 8.6 samples/s
239
+ Step 2210/50462 | Loss: 3.1696 | LR: 4.99e-04 | 6998.2ms/step | 8.6 samples/s
240
+ Step 2220/50462 | Loss: 3.2034 | LR: 4.99e-04 | 6996.9ms/step | 8.6 samples/s
241
+ Step 2230/50462 | Loss: 3.1856 | LR: 4.99e-04 | 6994.2ms/step | 8.6 samples/s
242
+ Step 2240/50462 | Loss: 3.1857 | LR: 4.99e-04 | 6994.2ms/step | 8.6 samples/s
243
+ Step 2250/50462 | Loss: 3.1901 | LR: 4.99e-04 | 6991.9ms/step | 8.6 samples/s
244
+ Validation at step 2250: Loss = 3.1624, Perplexity = 23.63
245
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-2250
246
+ Step 2260/50462 | Loss: 3.1717 | LR: 4.99e-04 | 10763.1ms/step | 5.6 samples/s
247
+ Step 2270/50462 | Loss: 3.1833 | LR: 4.99e-04 | 6996.3ms/step | 8.6 samples/s
248
+ Step 2280/50462 | Loss: 3.1842 | LR: 4.99e-04 | 6993.1ms/step | 8.6 samples/s
249
+ Step 2290/50462 | Loss: 3.1703 | LR: 4.99e-04 | 6994.1ms/step | 8.6 samples/s
250
+ Step 2300/50462 | Loss: 3.1720 | LR: 4.99e-04 | 6991.5ms/step | 8.6 samples/s
251
+ Step 2310/50462 | Loss: 3.1915 | LR: 4.99e-04 | 6993.6ms/step | 8.6 samples/s
252
+ Step 2320/50462 | Loss: 3.1855 | LR: 4.99e-04 | 6995.1ms/step | 8.6 samples/s
253
+ Step 2330/50462 | Loss: 3.1540 | LR: 4.99e-04 | 7005.4ms/step | 8.6 samples/s
254
+ Step 2340/50462 | Loss: 3.1747 | LR: 4.99e-04 | 7005.2ms/step | 8.6 samples/s
255
+ Step 2350/50462 | Loss: 3.1328 | LR: 4.99e-04 | 7004.4ms/step | 8.6 samples/s
256
+ Step 2360/50462 | Loss: 3.1299 | LR: 4.99e-04 | 7000.5ms/step | 8.6 samples/s
257
+ Step 2370/50462 | Loss: 3.1529 | LR: 4.99e-04 | 6998.4ms/step | 8.6 samples/s
258
+ Step 2380/50462 | Loss: 3.1293 | LR: 4.99e-04 | 6999.3ms/step | 8.6 samples/s
259
+ Step 2390/50462 | Loss: 3.1752 | LR: 4.99e-04 | 6999.6ms/step | 8.6 samples/s
260
+ Step 2400/50462 | Loss: 3.1625 | LR: 4.99e-04 | 6996.9ms/step | 8.6 samples/s
261
+ Step 2410/50462 | Loss: 3.1279 | LR: 4.99e-04 | 7000.3ms/step | 8.6 samples/s
262
+ Step 2420/50462 | Loss: 3.1246 | LR: 4.99e-04 | 7003.3ms/step | 8.6 samples/s
263
+ Step 2430/50462 | Loss: 3.1375 | LR: 4.99e-04 | 7000.2ms/step | 8.6 samples/s
264
+ Step 2440/50462 | Loss: 3.1167 | LR: 4.99e-04 | 7009.0ms/step | 8.6 samples/s
265
+ Step 2450/50462 | Loss: 3.1839 | LR: 4.99e-04 | 7001.4ms/step | 8.6 samples/s
266
+ Step 2460/50462 | Loss: 3.1271 | LR: 4.99e-04 | 6998.0ms/step | 8.6 samples/s
267
+ Step 2470/50462 | Loss: 3.1414 | LR: 4.99e-04 | 7006.0ms/step | 8.6 samples/s
268
+ Step 2480/50462 | Loss: 3.1352 | LR: 4.99e-04 | 7005.4ms/step | 8.6 samples/s
269
+ Step 2490/50462 | Loss: 3.1253 | LR: 4.98e-04 | 7004.1ms/step | 8.6 samples/s
270
+ Step 2500/50462 | Loss: 3.1194 | LR: 4.98e-04 | 7005.5ms/step | 8.6 samples/s
271
+ Validation at step 2500: Loss = 3.1202, Perplexity = 22.65
272
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-2500
273
+ Step 2510/50462 | Loss: 3.1033 | LR: 4.98e-04 | 10794.6ms/step | 5.6 samples/s
274
+ Step 2520/50462 | Loss: 3.1173 | LR: 4.98e-04 | 6997.5ms/step | 8.6 samples/s
275
+ Step 2530/50462 | Loss: 3.1184 | LR: 4.98e-04 | 6997.1ms/step | 8.6 samples/s
276
+ Step 2540/50462 | Loss: 3.1084 | LR: 4.98e-04 | 6996.1ms/step | 8.6 samples/s
277
+ Step 2550/50462 | Loss: 3.1290 | LR: 4.98e-04 | 6995.3ms/step | 8.6 samples/s
278
+ Step 2560/50462 | Loss: 3.1058 | LR: 4.98e-04 | 6989.3ms/step | 8.6 samples/s
279
+ Step 2570/50462 | Loss: 3.1115 | LR: 4.98e-04 | 6985.1ms/step | 8.6 samples/s
280
+ Step 2580/50462 | Loss: 3.0768 | LR: 4.98e-04 | 6987.8ms/step | 8.6 samples/s
281
+ Step 2590/50462 | Loss: 3.0940 | LR: 4.98e-04 | 6995.2ms/step | 8.6 samples/s
282
+ Step 2600/50462 | Loss: 3.1402 | LR: 4.98e-04 | 6995.5ms/step | 8.6 samples/s
283
+ Step 2610/50462 | Loss: 3.1374 | LR: 4.98e-04 | 6993.3ms/step | 8.6 samples/s
284
+ Step 2620/50462 | Loss: 3.1452 | LR: 4.98e-04 | 6999.2ms/step | 8.6 samples/s
285
+ Step 2630/50462 | Loss: 3.0915 | LR: 4.98e-04 | 6996.9ms/step | 8.6 samples/s
286
+ Step 2640/50462 | Loss: 3.1067 | LR: 4.98e-04 | 6995.4ms/step | 8.6 samples/s
287
+ Step 2650/50462 | Loss: 3.1091 | LR: 4.98e-04 | 7000.9ms/step | 8.6 samples/s
288
+ Step 2660/50462 | Loss: 3.1264 | LR: 4.98e-04 | 7014.9ms/step | 8.6 samples/s
289
+ Step 2670/50462 | Loss: 3.1062 | LR: 4.98e-04 | 7016.7ms/step | 8.6 samples/s
290
+ Step 2680/50462 | Loss: 3.1093 | LR: 4.98e-04 | 7029.1ms/step | 8.5 samples/s
291
+ Step 2690/50462 | Loss: 3.0895 | LR: 4.98e-04 | 7018.9ms/step | 8.5 samples/s
292
+ Step 2700/50462 | Loss: 3.1212 | LR: 4.98e-04 | 7019.5ms/step | 8.5 samples/s
293
+ Step 2710/50462 | Loss: 3.1282 | LR: 4.98e-04 | 7016.4ms/step | 8.6 samples/s
294
+ Step 2720/50462 | Loss: 3.1085 | LR: 4.98e-04 | 7016.2ms/step | 8.6 samples/s
295
+ Step 2730/50462 | Loss: 3.1022 | LR: 4.98e-04 | 7005.8ms/step | 8.6 samples/s
296
+ Step 2740/50462 | Loss: 3.1090 | LR: 4.98e-04 | 6988.8ms/step | 8.6 samples/s
297
+ Step 2750/50462 | Loss: 3.1129 | LR: 4.98e-04 | 6995.2ms/step | 8.6 samples/s
298
+ Validation at step 2750: Loss = 3.0843, Perplexity = 21.85
299
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-2750
300
+ Step 2760/50462 | Loss: 3.1199 | LR: 4.98e-04 | 10776.6ms/step | 5.6 samples/s
301
+ Step 2770/50462 | Loss: 3.1243 | LR: 4.98e-04 | 7010.9ms/step | 8.6 samples/s
302
+ Step 2780/50462 | Loss: 3.1189 | LR: 4.98e-04 | 7020.1ms/step | 8.5 samples/s
303
+ Step 2790/50462 | Loss: 3.1252 | LR: 4.98e-04 | 7017.2ms/step | 8.6 samples/s
304
+ Step 2800/50462 | Loss: 3.0913 | LR: 4.98e-04 | 7014.1ms/step | 8.6 samples/s
305
+ Step 2810/50462 | Loss: 3.1074 | LR: 4.98e-04 | 7006.9ms/step | 8.6 samples/s
306
+ Step 2820/50462 | Loss: 3.0836 | LR: 4.98e-04 | 7007.0ms/step | 8.6 samples/s
307
+ Step 2830/50462 | Loss: 3.0726 | LR: 4.98e-04 | 7006.4ms/step | 8.6 samples/s
308
+ Step 2840/50462 | Loss: 3.0717 | LR: 4.98e-04 | 7008.0ms/step | 8.6 samples/s
309
+ Step 2850/50462 | Loss: 3.0960 | LR: 4.98e-04 | 7006.4ms/step | 8.6 samples/s
310
+ Step 2860/50462 | Loss: 3.0877 | LR: 4.98e-04 | 7006.9ms/step | 8.6 samples/s
311
+ Step 2870/50462 | Loss: 3.0788 | LR: 4.98e-04 | 7007.4ms/step | 8.6 samples/s
312
+ Step 2880/50462 | Loss: 3.0712 | LR: 4.98e-04 | 7008.6ms/step | 8.6 samples/s
313
+ Step 2890/50462 | Loss: 3.0741 | LR: 4.98e-04 | 7005.6ms/step | 8.6 samples/s
314
+ Step 2900/50462 | Loss: 3.1151 | LR: 4.98e-04 | 7003.6ms/step | 8.6 samples/s
315
+ Step 2910/50462 | Loss: 3.0641 | LR: 4.98e-04 | 7006.8ms/step | 8.6 samples/s
316
+ Step 2920/50462 | Loss: 3.0694 | LR: 4.98e-04 | 7006.8ms/step | 8.6 samples/s
317
+ Step 2930/50462 | Loss: 3.0819 | LR: 4.98e-04 | 7009.3ms/step | 8.6 samples/s
318
+ Step 2940/50462 | Loss: 3.0491 | LR: 4.98e-04 | 7006.2ms/step | 8.6 samples/s
319
+ Step 2950/50462 | Loss: 3.0619 | LR: 4.98e-04 | 7012.4ms/step | 8.6 samples/s
320
+ Step 2960/50462 | Loss: 3.0773 | LR: 4.98e-04 | 7011.6ms/step | 8.6 samples/s
321
+ Step 2970/50462 | Loss: 3.0798 | LR: 4.98e-04 | 7007.0ms/step | 8.6 samples/s
322
+ Step 2980/50462 | Loss: 3.0732 | LR: 4.98e-04 | 7011.1ms/step | 8.6 samples/s
323
+ Step 2990/50462 | Loss: 3.0639 | LR: 4.97e-04 | 7007.3ms/step | 8.6 samples/s
324
+ Step 3000/50462 | Loss: 3.0502 | LR: 4.97e-04 | 7009.6ms/step | 8.6 samples/s
325
+ Validation at step 3000: Loss = 3.0547, Perplexity = 21.22
326
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-3000
327
+ Step 3010/50462 | Loss: 3.0558 | LR: 4.97e-04 | 10794.4ms/step | 5.6 samples/s
328
+ Step 3020/50462 | Loss: 3.0383 | LR: 4.97e-04 | 7004.8ms/step | 8.6 samples/s
329
+ Step 3030/50462 | Loss: 3.0735 | LR: 4.97e-04 | 7006.4ms/step | 8.6 samples/s
330
+ Step 3040/50462 | Loss: 3.0494 | LR: 4.97e-04 | 7010.6ms/step | 8.6 samples/s
331
+ Step 3050/50462 | Loss: 3.0584 | LR: 4.97e-04 | 7009.5ms/step | 8.6 samples/s
332
+ Step 3060/50462 | Loss: 3.0662 | LR: 4.97e-04 | 7012.8ms/step | 8.6 samples/s
333
+ Step 3070/50462 | Loss: 3.0913 | LR: 4.97e-04 | 7011.4ms/step | 8.6 samples/s
334
+ Step 3080/50462 | Loss: 3.0764 | LR: 4.97e-04 | 7006.5ms/step | 8.6 samples/s
335
+ Step 3090/50462 | Loss: 3.0916 | LR: 4.97e-04 | 7009.7ms/step | 8.6 samples/s
336
+ Step 3100/50462 | Loss: 3.0698 | LR: 4.97e-04 | 7011.2ms/step | 8.6 samples/s
337
+ Step 3110/50462 | Loss: 3.0645 | LR: 4.97e-04 | 7010.5ms/step | 8.6 samples/s
338
+ Step 3120/50462 | Loss: 3.0299 | LR: 4.97e-04 | 7008.8ms/step | 8.6 samples/s
339
+ Step 3130/50462 | Loss: 3.0276 | LR: 4.97e-04 | 7005.5ms/step | 8.6 samples/s
340
+ Step 3140/50462 | Loss: 3.0592 | LR: 4.97e-04 | 7006.3ms/step | 8.6 samples/s
341
+ Step 3150/50462 | Loss: 3.0562 | LR: 4.97e-04 | 7005.0ms/step | 8.6 samples/s
342
+ Step 3160/50462 | Loss: 3.0228 | LR: 4.97e-04 | 7003.1ms/step | 8.6 samples/s
343
+ Step 3170/50462 | Loss: 3.0266 | LR: 4.97e-04 | 7005.8ms/step | 8.6 samples/s
344
+ Step 3180/50462 | Loss: 3.0595 | LR: 4.97e-04 | 7001.5ms/step | 8.6 samples/s
345
+ Step 3190/50462 | Loss: 3.0480 | LR: 4.97e-04 | 6990.3ms/step | 8.6 samples/s
346
+ Step 3200/50462 | Loss: 3.0290 | LR: 4.97e-04 | 6992.9ms/step | 8.6 samples/s
347
+ Step 3210/50462 | Loss: 3.0576 | LR: 4.97e-04 | 6991.4ms/step | 8.6 samples/s
348
+ Step 3220/50462 | Loss: 3.0312 | LR: 4.97e-04 | 6986.6ms/step | 8.6 samples/s
349
+ Step 3230/50462 | Loss: 3.0388 | LR: 4.97e-04 | 6985.9ms/step | 8.6 samples/s
350
+ Step 3240/50462 | Loss: 3.0631 | LR: 4.97e-04 | 6986.2ms/step | 8.6 samples/s
351
+ Step 3250/50462 | Loss: 3.0602 | LR: 4.97e-04 | 6995.1ms/step | 8.6 samples/s
352
+ Validation at step 3250: Loss = 3.0245, Perplexity = 20.58
353
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-3250
354
+ Step 3260/50462 | Loss: 3.0145 | LR: 4.97e-04 | 10809.9ms/step | 5.6 samples/s
355
+ Step 3270/50462 | Loss: 3.0456 | LR: 4.97e-04 | 7006.2ms/step | 8.6 samples/s
356
+ Step 3280/50462 | Loss: 3.0094 | LR: 4.97e-04 | 7012.6ms/step | 8.6 samples/s
357
+ Step 3290/50462 | Loss: 3.0381 | LR: 4.97e-04 | 7013.9ms/step | 8.6 samples/s
358
+ Step 3300/50462 | Loss: 3.0133 | LR: 4.97e-04 | 7018.6ms/step | 8.5 samples/s
359
+ Step 3310/50462 | Loss: 3.0206 | LR: 4.97e-04 | 7014.5ms/step | 8.6 samples/s
360
+ Step 3320/50462 | Loss: 3.0257 | LR: 4.97e-04 | 7006.3ms/step | 8.6 samples/s
361
+ Step 3330/50462 | Loss: 3.0168 | LR: 4.97e-04 | 7018.8ms/step | 8.5 samples/s
362
+ Step 3340/50462 | Loss: 3.0034 | LR: 4.97e-04 | 7012.0ms/step | 8.6 samples/s
363
+ Step 3350/50462 | Loss: 3.0155 | LR: 4.97e-04 | 6993.5ms/step | 8.6 samples/s
364
+ Step 3360/50462 | Loss: 3.0274 | LR: 4.97e-04 | 6995.4ms/step | 8.6 samples/s
365
+ Step 3370/50462 | Loss: 3.0142 | LR: 4.97e-04 | 6987.6ms/step | 8.6 samples/s
366
+ Step 3380/50462 | Loss: 3.0444 | LR: 4.97e-04 | 6988.8ms/step | 8.6 samples/s
367
+ Step 3390/50462 | Loss: 3.0144 | LR: 4.97e-04 | 7003.5ms/step | 8.6 samples/s
368
+ Step 3400/50462 | Loss: 3.0029 | LR: 4.97e-04 | 7014.1ms/step | 8.6 samples/s
369
+ Step 3410/50462 | Loss: 3.0591 | LR: 4.96e-04 | 7007.0ms/step | 8.6 samples/s
370
+ Step 3420/50462 | Loss: 3.0262 | LR: 4.96e-04 | 7005.3ms/step | 8.6 samples/s
371
+ Step 3430/50462 | Loss: 3.0395 | LR: 4.96e-04 | 7007.2ms/step | 8.6 samples/s
372
+ Step 3440/50462 | Loss: 3.0176 | LR: 4.96e-04 | 7003.7ms/step | 8.6 samples/s
373
+ Step 3450/50462 | Loss: 2.9979 | LR: 4.96e-04 | 7002.4ms/step | 8.6 samples/s
374
+ Step 3460/50462 | Loss: 3.0453 | LR: 4.96e-04 | 7002.0ms/step | 8.6 samples/s
375
+ Step 3470/50462 | Loss: 2.9988 | LR: 4.96e-04 | 7003.9ms/step | 8.6 samples/s
376
+ Step 3480/50462 | Loss: 3.0340 | LR: 4.96e-04 | 6999.6ms/step | 8.6 samples/s
377
+ Step 3490/50462 | Loss: 3.0024 | LR: 4.96e-04 | 7003.7ms/step | 8.6 samples/s
378
+ Step 3500/50462 | Loss: 3.0125 | LR: 4.96e-04 | 7004.1ms/step | 8.6 samples/s
379
+ Validation at step 3500: Loss = 2.9992, Perplexity = 20.07
380
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-3500
381
+ Step 3510/50462 | Loss: 3.0054 | LR: 4.96e-04 | 10800.0ms/step | 5.6 samples/s
382
+ Step 3520/50462 | Loss: 3.0265 | LR: 4.96e-04 | 6996.9ms/step | 8.6 samples/s
383
+ Step 3530/50462 | Loss: 2.9760 | LR: 4.96e-04 | 6996.8ms/step | 8.6 samples/s
384
+ Step 3540/50462 | Loss: 3.0349 | LR: 4.96e-04 | 6994.7ms/step | 8.6 samples/s
385
+ Step 3550/50462 | Loss: 3.0334 | LR: 4.96e-04 | 6995.0ms/step | 8.6 samples/s
386
+ Step 3560/50462 | Loss: 3.0091 | LR: 4.96e-04 | 6994.6ms/step | 8.6 samples/s
387
+ Step 3570/50462 | Loss: 3.0159 | LR: 4.96e-04 | 6995.8ms/step | 8.6 samples/s
388
+ Step 3580/50462 | Loss: 2.9684 | LR: 4.96e-04 | 6997.7ms/step | 8.6 samples/s
389
+ Step 3590/50462 | Loss: 2.9992 | LR: 4.96e-04 | 6993.4ms/step | 8.6 samples/s
390
+ Step 3600/50462 | Loss: 2.9951 | LR: 4.96e-04 | 6996.2ms/step | 8.6 samples/s
391
+ Step 3610/50462 | Loss: 3.0092 | LR: 4.96e-04 | 6998.2ms/step | 8.6 samples/s
392
+ Step 3620/50462 | Loss: 2.9983 | LR: 4.96e-04 | 6993.0ms/step | 8.6 samples/s
393
+ Step 3630/50462 | Loss: 3.0109 | LR: 4.96e-04 | 6995.3ms/step | 8.6 samples/s
394
+ Step 3640/50462 | Loss: 3.0164 | LR: 4.96e-04 | 6995.2ms/step | 8.6 samples/s
395
+ Step 3650/50462 | Loss: 2.9972 | LR: 4.96e-04 | 6996.5ms/step | 8.6 samples/s
396
+ Step 3660/50462 | Loss: 2.9938 | LR: 4.96e-04 | 6996.9ms/step | 8.6 samples/s
397
+ Step 3670/50462 | Loss: 3.0070 | LR: 4.96e-04 | 6995.0ms/step | 8.6 samples/s
398
+ Step 3680/50462 | Loss: 2.9592 | LR: 4.96e-04 | 6998.3ms/step | 8.6 samples/s
399
+ Step 3690/50462 | Loss: 2.9704 | LR: 4.96e-04 | 6994.9ms/step | 8.6 samples/s
400
+ Step 3700/50462 | Loss: 2.9944 | LR: 4.96e-04 | 6992.0ms/step | 8.6 samples/s
401
+ Step 3710/50462 | Loss: 2.9934 | LR: 4.96e-04 | 6995.7ms/step | 8.6 samples/s
402
+ Step 3720/50462 | Loss: 2.9773 | LR: 4.96e-04 | 6999.1ms/step | 8.6 samples/s
403
+ Step 3730/50462 | Loss: 2.9717 | LR: 4.96e-04 | 6990.4ms/step | 8.6 samples/s
404
+ Step 3740/50462 | Loss: 2.9750 | LR: 4.96e-04 | 6995.3ms/step | 8.6 samples/s
405
+ Step 3750/50462 | Loss: 2.9603 | LR: 4.96e-04 | 6990.2ms/step | 8.6 samples/s
406
+ Validation at step 3750: Loss = 2.9805, Perplexity = 19.70
407
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-3750
408
+ Step 3760/50462 | Loss: 2.9941 | LR: 4.95e-04 | 10820.4ms/step | 5.5 samples/s
409
+ Step 3770/50462 | Loss: 2.9756 | LR: 4.95e-04 | 6985.8ms/step | 8.6 samples/s
410
+ Step 3780/50462 | Loss: 3.0174 | LR: 4.95e-04 | 6988.8ms/step | 8.6 samples/s
411
+ Step 3790/50462 | Loss: 3.0043 | LR: 4.95e-04 | 6985.6ms/step | 8.6 samples/s
412
+ Step 3800/50462 | Loss: 2.9699 | LR: 4.95e-04 | 6985.1ms/step | 8.6 samples/s
413
+ Step 3810/50462 | Loss: 2.9569 | LR: 4.95e-04 | 6990.3ms/step | 8.6 samples/s
414
+ Step 3820/50462 | Loss: 2.9567 | LR: 4.95e-04 | 6987.1ms/step | 8.6 samples/s
415
+ Step 3830/50462 | Loss: 3.0193 | LR: 4.95e-04 | 6989.8ms/step | 8.6 samples/s
416
+ Step 3840/50462 | Loss: 2.9586 | LR: 4.95e-04 | 6985.7ms/step | 8.6 samples/s
417
+ Step 3850/50462 | Loss: 3.0179 | LR: 4.95e-04 | 6983.8ms/step | 8.6 samples/s
418
+ Step 3860/50462 | Loss: 2.9404 | LR: 4.95e-04 | 6985.8ms/step | 8.6 samples/s
419
+ Step 3870/50462 | Loss: 2.9509 | LR: 4.95e-04 | 6988.0ms/step | 8.6 samples/s
420
+ Step 3880/50462 | Loss: 2.9673 | LR: 4.95e-04 | 6990.4ms/step | 8.6 samples/s
421
+ Step 3890/50462 | Loss: 2.9607 | LR: 4.95e-04 | 6985.9ms/step | 8.6 samples/s
422
+ Step 3900/50462 | Loss: 3.0096 | LR: 4.95e-04 | 6987.0ms/step | 8.6 samples/s
423
+ Step 3910/50462 | Loss: 2.9598 | LR: 4.95e-04 | 6984.0ms/step | 8.6 samples/s
424
+ Step 3920/50462 | Loss: 2.9733 | LR: 4.95e-04 | 6986.5ms/step | 8.6 samples/s
425
+ Step 3930/50462 | Loss: 2.9840 | LR: 4.95e-04 | 6988.0ms/step | 8.6 samples/s
426
+ Step 3940/50462 | Loss: 2.9849 | LR: 4.95e-04 | 6991.4ms/step | 8.6 samples/s
427
+ Step 3950/50462 | Loss: 2.9811 | LR: 4.95e-04 | 6990.0ms/step | 8.6 samples/s
428
+ Step 3960/50462 | Loss: 2.9710 | LR: 4.95e-04 | 6990.3ms/step | 8.6 samples/s
429
+ Step 3970/50462 | Loss: 3.0009 | LR: 4.95e-04 | 6989.9ms/step | 8.6 samples/s
430
+ Step 3980/50462 | Loss: 2.9627 | LR: 4.95e-04 | 6993.3ms/step | 8.6 samples/s
431
+ Step 3990/50462 | Loss: 2.9685 | LR: 4.95e-04 | 6991.1ms/step | 8.6 samples/s
432
+ Step 4000/50462 | Loss: 2.9596 | LR: 4.95e-04 | 6990.9ms/step | 8.6 samples/s
433
+ Validation at step 4000: Loss = 2.9601, Perplexity = 19.30
434
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-4000
435
+ Step 4010/50462 | Loss: 2.9981 | LR: 4.95e-04 | 10780.6ms/step | 5.6 samples/s
436
+ Step 4020/50462 | Loss: 2.9723 | LR: 4.95e-04 | 6995.1ms/step | 8.6 samples/s
437
+ Step 4030/50462 | Loss: 2.9815 | LR: 4.95e-04 | 6993.8ms/step | 8.6 samples/s
438
+ Step 4040/50462 | Loss: 2.9473 | LR: 4.95e-04 | 6996.5ms/step | 8.6 samples/s
439
+ Step 4050/50462 | Loss: 2.9688 | LR: 4.95e-04 | 6997.4ms/step | 8.6 samples/s
440
+ Step 4060/50462 | Loss: 2.9749 | LR: 4.95e-04 | 6997.4ms/step | 8.6 samples/s
441
+ Step 4070/50462 | Loss: 2.9614 | LR: 4.95e-04 | 7004.9ms/step | 8.6 samples/s
442
+ Step 4080/50462 | Loss: 2.9888 | LR: 4.94e-04 | 6995.8ms/step | 8.6 samples/s
443
+ Step 4090/50462 | Loss: 2.9840 | LR: 4.94e-04 | 6994.3ms/step | 8.6 samples/s
444
+ Step 4100/50462 | Loss: 2.9587 | LR: 4.94e-04 | 6995.2ms/step | 8.6 samples/s
445
+ Step 4110/50462 | Loss: 2.9552 | LR: 4.94e-04 | 6994.8ms/step | 8.6 samples/s
446
+ Step 4120/50462 | Loss: 2.9543 | LR: 4.94e-04 | 6997.1ms/step | 8.6 samples/s
447
+ Step 4130/50462 | Loss: 2.9524 | LR: 4.94e-04 | 6997.7ms/step | 8.6 samples/s
448
+ Step 4140/50462 | Loss: 2.9478 | LR: 4.94e-04 | 6997.7ms/step | 8.6 samples/s
449
+ Step 4150/50462 | Loss: 2.9574 | LR: 4.94e-04 | 6996.8ms/step | 8.6 samples/s
450
+ Step 4160/50462 | Loss: 2.9604 | LR: 4.94e-04 | 6996.1ms/step | 8.6 samples/s
451
+ Step 4170/50462 | Loss: 2.9167 | LR: 4.94e-04 | 6997.5ms/step | 8.6 samples/s
452
+ Step 4180/50462 | Loss: 2.9211 | LR: 4.94e-04 | 6996.1ms/step | 8.6 samples/s
453
+ Step 4190/50462 | Loss: 2.9405 | LR: 4.94e-04 | 6998.5ms/step | 8.6 samples/s
454
+ Step 4200/50462 | Loss: 2.9799 | LR: 4.94e-04 | 6997.3ms/step | 8.6 samples/s
455
+ Step 4210/50462 | Loss: 2.9519 | LR: 4.94e-04 | 6995.0ms/step | 8.6 samples/s
456
+ Step 4220/50462 | Loss: 2.9385 | LR: 4.94e-04 | 6997.7ms/step | 8.6 samples/s
457
+ Step 4230/50462 | Loss: 2.9527 | LR: 4.94e-04 | 6997.7ms/step | 8.6 samples/s
458
+ Step 4240/50462 | Loss: 2.9633 | LR: 4.94e-04 | 6997.3ms/step | 8.6 samples/s
459
+ Step 4250/50462 | Loss: 2.9604 | LR: 4.94e-04 | 6996.6ms/step | 8.6 samples/s
460
+ Validation at step 4250: Loss = 2.9433, Perplexity = 18.98
461
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-4250
462
+ Step 4260/50462 | Loss: 2.9435 | LR: 4.94e-04 | 10779.6ms/step | 5.6 samples/s
463
+ Step 4270/50462 | Loss: 2.9536 | LR: 4.94e-04 | 6984.5ms/step | 8.6 samples/s
464
+ Step 4280/50462 | Loss: 2.9440 | LR: 4.94e-04 | 6987.6ms/step | 8.6 samples/s
465
+ Step 4290/50462 | Loss: 2.9702 | LR: 4.94e-04 | 6994.7ms/step | 8.6 samples/s
466
+ Step 4300/50462 | Loss: 2.9500 | LR: 4.94e-04 | 6992.7ms/step | 8.6 samples/s
467
+ Step 4310/50462 | Loss: 2.9406 | LR: 4.94e-04 | 6990.2ms/step | 8.6 samples/s
468
+ Step 4320/50462 | Loss: 2.9522 | LR: 4.94e-04 | 6991.9ms/step | 8.6 samples/s
469
+ Step 4330/50462 | Loss: 2.9094 | LR: 4.94e-04 | 6992.1ms/step | 8.6 samples/s
470
+ Step 4340/50462 | Loss: 2.9586 | LR: 4.94e-04 | 6989.2ms/step | 8.6 samples/s
471
+ Step 4350/50462 | Loss: 2.9547 | LR: 4.94e-04 | 6992.2ms/step | 8.6 samples/s
472
+ Step 4360/50462 | Loss: 2.9547 | LR: 4.94e-04 | 6997.4ms/step | 8.6 samples/s
473
+ Step 4370/50462 | Loss: 2.9486 | LR: 4.93e-04 | 7023.0ms/step | 8.5 samples/s
474
+ Step 4380/50462 | Loss: 2.9227 | LR: 4.93e-04 | 7033.4ms/step | 8.5 samples/s
475
+ Step 4390/50462 | Loss: 2.9447 | LR: 4.93e-04 | 7029.2ms/step | 8.5 samples/s
476
+ Step 4400/50462 | Loss: 2.9726 | LR: 4.93e-04 | 7028.9ms/step | 8.5 samples/s
477
+ Step 4410/50462 | Loss: 2.9471 | LR: 4.93e-04 | 7015.8ms/step | 8.6 samples/s
478
+ Step 4420/50462 | Loss: 2.9220 | LR: 4.93e-04 | 7021.1ms/step | 8.5 samples/s
479
+ Step 4430/50462 | Loss: 2.9621 | LR: 4.93e-04 | 7015.4ms/step | 8.6 samples/s
480
+ Step 4440/50462 | Loss: 2.8961 | LR: 4.93e-04 | 6995.7ms/step | 8.6 samples/s
481
+ Step 4450/50462 | Loss: 2.9816 | LR: 4.93e-04 | 6988.4ms/step | 8.6 samples/s
482
+ Step 4460/50462 | Loss: 2.9553 | LR: 4.93e-04 | 6989.8ms/step | 8.6 samples/s
483
+ Step 4470/50462 | Loss: 2.9377 | LR: 4.93e-04 | 6990.9ms/step | 8.6 samples/s
484
+ Step 4480/50462 | Loss: 2.9308 | LR: 4.93e-04 | 7017.4ms/step | 8.6 samples/s
485
+ Step 4490/50462 | Loss: 2.9559 | LR: 4.93e-04 | 7020.0ms/step | 8.5 samples/s
486
+ Step 4500/50462 | Loss: 2.9532 | LR: 4.93e-04 | 7015.9ms/step | 8.6 samples/s
487
+ Validation at step 4500: Loss = 2.9257, Perplexity = 18.65
488
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-4500
489
+ Step 4510/50462 | Loss: 2.9406 | LR: 4.93e-04 | 10835.1ms/step | 5.5 samples/s
490
+ Step 4520/50462 | Loss: 2.9427 | LR: 4.93e-04 | 6984.9ms/step | 8.6 samples/s
491
+ Step 4530/50462 | Loss: 2.9319 | LR: 4.93e-04 | 6986.4ms/step | 8.6 samples/s
492
+ Step 4540/50462 | Loss: 2.9591 | LR: 4.93e-04 | 6988.8ms/step | 8.6 samples/s
493
+ Step 4550/50462 | Loss: 2.8968 | LR: 4.93e-04 | 6994.0ms/step | 8.6 samples/s
494
+ Step 4560/50462 | Loss: 2.9353 | LR: 4.93e-04 | 6996.5ms/step | 8.6 samples/s
495
+ Step 4570/50462 | Loss: 2.9200 | LR: 4.93e-04 | 6998.4ms/step | 8.6 samples/s
496
+ Step 4580/50462 | Loss: 2.9691 | LR: 4.93e-04 | 7001.0ms/step | 8.6 samples/s
497
+ Step 4590/50462 | Loss: 2.9395 | LR: 4.93e-04 | 7001.3ms/step | 8.6 samples/s
498
+ Step 4600/50462 | Loss: 2.9255 | LR: 4.93e-04 | 7000.8ms/step | 8.6 samples/s
499
+ Step 4610/50462 | Loss: 2.9342 | LR: 4.93e-04 | 6999.3ms/step | 8.6 samples/s
500
+ Step 4620/50462 | Loss: 2.9498 | LR: 4.93e-04 | 7002.2ms/step | 8.6 samples/s
501
+ Step 4630/50462 | Loss: 2.9489 | LR: 4.93e-04 | 7002.7ms/step | 8.6 samples/s
502
+ Step 4640/50462 | Loss: 2.8864 | LR: 4.92e-04 | 7004.4ms/step | 8.6 samples/s
503
+ Step 4650/50462 | Loss: 2.9097 | LR: 4.92e-04 | 7003.4ms/step | 8.6 samples/s
504
+ Step 4660/50462 | Loss: 2.9420 | LR: 4.92e-04 | 7003.1ms/step | 8.6 samples/s
505
+ Step 4670/50462 | Loss: 2.9389 | LR: 4.92e-04 | 7002.2ms/step | 8.6 samples/s
506
+ Step 4680/50462 | Loss: 2.9490 | LR: 4.92e-04 | 6998.3ms/step | 8.6 samples/s
507
+ Step 4690/50462 | Loss: 2.9286 | LR: 4.92e-04 | 6999.6ms/step | 8.6 samples/s
508
+ Step 4700/50462 | Loss: 2.9467 | LR: 4.92e-04 | 6999.5ms/step | 8.6 samples/s
509
+ Step 4710/50462 | Loss: 2.9263 | LR: 4.92e-04 | 6996.3ms/step | 8.6 samples/s
510
+ Step 4720/50462 | Loss: 2.9237 | LR: 4.92e-04 | 6999.5ms/step | 8.6 samples/s
511
+ Step 4730/50462 | Loss: 2.9466 | LR: 4.92e-04 | 7001.0ms/step | 8.6 samples/s
512
+ Step 4740/50462 | Loss: 2.9121 | LR: 4.92e-04 | 6999.4ms/step | 8.6 samples/s
513
+ Step 4750/50462 | Loss: 2.9019 | LR: 4.92e-04 | 6992.7ms/step | 8.6 samples/s
514
+ Validation at step 4750: Loss = 2.9103, Perplexity = 18.36
515
+ Saved checkpoint to /mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810/checkpoint-4750
516
+ Step 4760/50462 | Loss: 2.9249 | LR: 4.92e-04 | 10842.3ms/step | 5.5 samples/s
517
+ Step 4770/50462 | Loss: 2.9126 | LR: 4.92e-04 | 7000.6ms/step | 8.6 samples/s
518
+ Step 4780/50462 | Loss: 2.9187 | LR: 4.92e-04 | 7000.4ms/step | 8.6 samples/s
519
+ Step 4790/50462 | Loss: 2.9223 | LR: 4.92e-04 | 6999.1ms/step | 8.6 samples/s
520
+ Step 4800/50462 | Loss: 2.9357 | LR: 4.92e-04 | 7009.3ms/step | 8.6 samples/s
521
+ Step 4810/50462 | Loss: 2.9072 | LR: 4.92e-04 | 6997.4ms/step | 8.6 samples/s
522
+ Step 4820/50462 | Loss: 2.9050 | LR: 4.92e-04 | 6997.2ms/step | 8.6 samples/s
523
+ Step 4830/50462 | Loss: 2.9059 | LR: 4.92e-04 | 6995.3ms/step | 8.6 samples/s
524
+ Step 4840/50462 | Loss: 2.9244 | LR: 4.92e-04 | 6998.3ms/step | 8.6 samples/s
525
+ Step 4850/50462 | Loss: 2.9042 | LR: 4.92e-04 | 6996.2ms/step | 8.6 samples/s
526
+ Step 4860/50462 | Loss: 2.9137 | LR: 4.92e-04 | 6993.9ms/step | 8.6 samples/s
epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/files/requirements.txt ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pip==26.0.1
2
+ nvidia-cusparselt-cu13==0.8.0
3
+ mpmath==1.3.0
4
+ cuda-toolkit==13.0.2
5
+ zstandard==0.25.0
6
+ xxhash==3.7.0
7
+ urllib3==2.6.3
8
+ typing_extensions==4.15.0
9
+ triton==3.6.0
10
+ tqdm==4.67.3
11
+ torchinfo==1.8.0
12
+ sympy==1.14.0
13
+ smmap==5.0.3
14
+ six==1.17.0
15
+ shellingham==1.5.4
16
+ setuptools==81.0.0
17
+ safetensors==0.7.0
18
+ regex==2026.4.4
19
+ PyYAML==6.0.3
20
+ Pygments==2.20.0
21
+ pyarrow==24.0.0
22
+ psutil==7.2.2
23
+ protobuf==7.34.1
24
+ propcache==0.4.1
25
+ platformdirs==4.9.6
26
+ packaging==26.2
27
+ nvidia-nvtx==13.0.85
28
+ nvidia-nvshmem-cu13==3.4.5
29
+ nvidia-nvjitlink==13.0.88
30
+ nvidia-nccl-cu13==2.28.9
31
+ nvidia-curand==10.4.0.35
32
+ nvidia-cufile==1.15.1.6
33
+ nvidia-cuda-runtime==13.0.96
34
+ nvidia-cuda-nvrtc==13.0.88
35
+ nvidia-cuda-cupti==13.0.85
36
+ nvidia-cublas==13.1.0.3
37
+ numpy==2.4.4
38
+ networkx==3.6.1
39
+ multidict==6.7.1
40
+ mdurl==0.1.2
41
+ MarkupSafe==3.0.3
42
+ idna==3.13
43
+ hf-xet==1.4.3
44
+ h11==0.16.0
45
+ fsspec==2026.2.0
46
+ frozenlist==1.8.0
47
+ filelock==3.29.0
48
+ dill==0.4.1
49
+ cuda-pathfinder==1.5.4
50
+ click==8.3.3
51
+ charset-normalizer==3.4.7
52
+ certifi==2026.4.22
53
+ attrs==26.1.0
54
+ annotated-types==0.7.0
55
+ annotated-doc==0.0.4
56
+ aiohappyeyeballs==2.6.1
57
+ yarl==1.23.0
58
+ typing-inspection==0.4.2
59
+ sentry-sdk==2.58.0
60
+ requests==2.33.1
61
+ python-dateutil==2.9.0.post0
62
+ pydantic_core==2.46.3
63
+ nvidia-cusparse==12.6.3.3
64
+ nvidia-cufft==12.0.0.61
65
+ nvidia-cudnn-cu13==9.19.0.56
66
+ multiprocess==0.70.19
67
+ markdown-it-py==4.0.0
68
+ Jinja2==3.1.6
69
+ httpcore==1.0.9
70
+ gitdb==4.0.12
71
+ cuda-bindings==13.2.0
72
+ anyio==4.13.0
73
+ aiosignal==1.4.0
74
+ rich==15.0.0
75
+ pydantic==2.13.3
76
+ pandas==3.0.2
77
+ nvidia-cusolver==12.0.4.66
78
+ httpx==0.28.1
79
+ GitPython==3.1.49
80
+ aiohttp==3.13.5
81
+ wandb==0.26.1
82
+ typer==0.25.1
83
+ torch==2.11.0
84
+ huggingface_hub==1.13.0
85
+ tokenizers==0.22.2
86
+ datasets==4.8.5
87
+ accelerate==1.13.0
88
+ transformers==5.7.0
89
+ word2number==1.1
90
+ sqlitedict==2.1.0
91
+ pytz==2026.2
92
+ threadpoolctl==3.6.0
93
+ tcolorpy==0.1.7
94
+ tabulate==0.10.0
95
+ scipy==1.17.1
96
+ portalocker==3.2.0
97
+ pathvalidate==3.3.1
98
+ narwhals==2.22.1
99
+ more-itertools==11.1.0
100
+ lxml==6.1.1
101
+ joblib==1.5.3
102
+ colorama==0.4.6
103
+ chardet==6.0.0.post1
104
+ absl-py==2.4.0
105
+ scikit-learn==1.9.0
106
+ sacrebleu==2.6.0
107
+ nltk==3.9.4
108
+ mbstrdecoder==1.1.5
109
+ typepy==1.3.5
110
+ rouge_score==0.1.2
111
+ DataProperty==1.1.1
112
+ tabledata==1.3.5
113
+ evaluate==0.4.6
114
+ pytablewriter==1.2.1
115
+ lm_eval==0.4.12
116
+ pluggy==1.6.0
117
+ iniconfig==2.3.0
118
+ pytest==9.1.0
119
+ py-spy==0.4.2
120
+ pillow==12.2.0
121
+ tifffile==2026.6.1
122
+ imagecodecs==2026.6.26
123
+ zipp==4.1.0
124
+ importlib_metadata==9.0.0
125
+ derivative==0.6.3
126
+ pysindy==2.1.0
127
+ pyparsing==3.3.2
128
+ kiwisolver==1.5.0
129
+ fonttools==4.63.0
130
+ cycler==0.12.1
131
+ contourpy==1.3.3
132
+ matplotlib==3.11.0
133
+ sortedcontainers==2.4.0
134
+ wcwidth==0.8.2
135
+ sigfig==1.3.19
136
+ importlib_resources==7.1.0
137
+ enum-tools==0.13.0
138
+ cramjam==2.11.0
139
+ python-snappy==0.7.3
140
+ compact-json==1.8.2
141
+ numbers-parser==4.18.5
142
+ entmax==1.3
epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/files/wandb-metadata.json ADDED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-23T04:38:41.848448Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p2_1p2b_dense_seed70_4gpu.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "a100-train-5-gpu-0-2",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 64,
21
+ "cpu_count_logical": 128,
22
+ "gpu": "NVIDIA A100-SXM4-80GB",
23
+ "gpu_count": 5,
24
+ "disk": {
25
+ "/": {
26
+ "total": "7679362727936",
27
+ "used": "4492280492032"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1081884971008"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA A100-SXM4-80GB",
36
+ "memoryTotal": "85899345920",
37
+ "cudaCores": 6912,
38
+ "architecture": "Ampere",
39
+ "uuid": "GPU-da403873-77fd-0400-8874-ee85122e86ce"
40
+ },
41
+ {
42
+ "name": "NVIDIA A100-SXM4-80GB",
43
+ "memoryTotal": "85899345920",
44
+ "cudaCores": 6912,
45
+ "architecture": "Ampere",
46
+ "uuid": "GPU-ac7c2b06-0e54-6c0e-5c5e-bef297b937e2"
47
+ },
48
+ {
49
+ "name": "NVIDIA A100-SXM4-80GB",
50
+ "memoryTotal": "85899345920",
51
+ "cudaCores": 6912,
52
+ "architecture": "Ampere",
53
+ "uuid": "GPU-2f2cf903-6615-a5f8-353b-076dccd64125"
54
+ },
55
+ {
56
+ "name": "NVIDIA A100-SXM4-80GB",
57
+ "memoryTotal": "85899345920",
58
+ "cudaCores": 6912,
59
+ "architecture": "Ampere",
60
+ "uuid": "GPU-75d03f5a-9fc2-0173-5629-3ef21c2f450c"
61
+ },
62
+ {
63
+ "name": "NVIDIA A100-SXM4-80GB",
64
+ "memoryTotal": "85899345920",
65
+ "cudaCores": 6912,
66
+ "architecture": "Ampere",
67
+ "uuid": "GPU-1d4bb5b9-c680-19af-9659-b21aa13b3c4d"
68
+ }
69
+ ],
70
+ "cudaVersion": "13.0",
71
+ "writerId": "bsb4p1tr8xxk4uc4w45xux1wm5szyk5p"
72
+ }
epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-08-23 04:38:41,885 INFO MainThread:19202 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-08-23 04:38:41,885 INFO MainThread:19202 [wandb_setup.py:_flush():81] Configure stats pid to 19202
3
+ 2026-08-23 04:38:41,885 INFO MainThread:19202 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-08-23 04:38:41,886 INFO MainThread:19202 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/logs/debug.log
5
+ 2026-08-23 04:38:41,886 INFO MainThread:19202 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/logs/debug-internal.log
6
+ 2026-08-23 04:38:41,886 INFO MainThread:19202 [wandb_init.py:init():850] calling init triggers
7
+ 2026-08-23 04:38:41,886 INFO MainThread:19202 [wandb_init.py:init():855] wandb.init called with sweep_config: {}
8
+ config: {'model_path': '/mloscratch/homes/navasard/qwen3_06b_base', 'model_dtype': 'bfloat16', 'init_from_scratch': True, 'train_path': '/mloscratch/homes/navasard/data/fineweb_large/tokenized', 'val_path': None, 'max_seq_length': 8192, 'num_workers': 8, 'output_dir': '/mloscratch/homes/navasard/sparse/checkpoints/l8k1p2b_dense_seed70/run_20260823_043810', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 50462, 'per_device_batch_size': 1, 'gradient_accumulation_steps': 15, 'learning_rate': 0.0005, 'weight_decay': 0.01, 'warmup_steps': 750, 'lr_scheduler': 'cosine', 'max_grad_norm': 1.0, 'save_steps': 250, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 10, 'seed': 70, 'verbose': True, 'checkpoint_activations': False, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '1.2b dense 8k seed70', 'wandb_tags': ['qwen3', '1.2b', '8k', 'p2', 'dense', 'seed70'], 'hidden_size': 1536, 'intermediate_size': 6144, 'num_attention_heads': 12, 'num_key_value_heads': 4, 'num_hidden_layers': 28, 'head_dim': None, 'attention_dropout': None, 'rope_train_position_scale_set': None, 'attention_type': 'sparse_stochastic', 'sparse_block_size': 128, 'sparse_num_random_blocks': 31, 'sparse_mask_pool_size': 1, 'sparse_val_dense_compare': False, 'sparse_layer_indices': None, 'sparse_mask_cache_dir': None, 'sparse_mask_seed': 0, 'sparse_mixed_nrb_set': None, 'dense_mode': True, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-08-23 04:38:41,886 INFO MainThread:19202 [wandb_init.py:init():898] starting backend
10
+ 2026-08-23 04:38:42,416 INFO MainThread:19202 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-08-23 04:38:42,938 INFO MainThread:19202 [wandb_init.py:init():918] backend started and connected
12
+ 2026-08-23 04:38:42,945 INFO MainThread:19202 [wandb_init.py:init():988] updated telemetry
13
+ 2026-08-23 04:38:42,974 INFO MainThread:19202 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-08-23 04:38:43,420 INFO MainThread:19202 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-08-23 04:38:43,596 INFO MainThread:19202 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-08-23 04:38:43,596 INFO MainThread:19202 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-08-23 04:38:43,596 INFO MainThread:19202 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-08-23 04:38:43,596 INFO MainThread:19202 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-08-23 04:38:43,603 INFO MainThread:19202 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/run-20260823_043841-rgmkfpnl/run-rgmkfpnl.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f64a10c8a14fe30c917bab29afc12077d78d48524e30fd9403cff0186c66262e
3
+ size 6651904
epfl_final/SparseAtt/wandb/run-20260823_140341-qctui00g/files/wandb-metadata.json ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.8.0-124-generic-x86_64-with-glibc2.35",
3
+ "python": "CPython 3.13.13",
4
+ "startedAt": "2026-08-23T14:03:41.679954Z",
5
+ "args": [
6
+ "--config",
7
+ "configs/p2_1p2b_dense_seed71_6gpu.json"
8
+ ],
9
+ "program": "/mloscratch/homes/navasard/sparse/SparseAtt/train.py",
10
+ "codePath": "train.py",
11
+ "codePathLocal": "train.py",
12
+ "git": {
13
+ "remote": "https://github.com/rubentium/TokenPrune.git",
14
+ "commit": "5637886f0cc760009989b14e68c6ae06d4827418"
15
+ },
16
+ "email": "ruben.navasardyan@mail.utoronto.ca",
17
+ "root": "/mloscratch/homes/navasard/sparse/SparseAtt",
18
+ "host": "a100-train-6-gpu-0-0",
19
+ "executable": "/mloscratch/homes/navasard/conda/bin/python",
20
+ "cpu_count": 64,
21
+ "cpu_count_logical": 128,
22
+ "gpu": "NVIDIA A100-SXM4-80GB",
23
+ "gpu_count": 6,
24
+ "disk": {
25
+ "/": {
26
+ "total": "7679362727936",
27
+ "used": "5253326848000"
28
+ }
29
+ },
30
+ "memory": {
31
+ "total": "1081884884992"
32
+ },
33
+ "gpu_nvidia": [
34
+ {
35
+ "name": "NVIDIA A100-SXM4-80GB",
36
+ "memoryTotal": "85899345920",
37
+ "cudaCores": 6912,
38
+ "architecture": "Ampere",
39
+ "uuid": "GPU-d89fb348-1c38-9080-1e32-49dcf38bb677"
40
+ },
41
+ {
42
+ "name": "NVIDIA A100-SXM4-80GB",
43
+ "memoryTotal": "85899345920",
44
+ "cudaCores": 6912,
45
+ "architecture": "Ampere",
46
+ "uuid": "GPU-8a962b76-efe8-388f-24a2-7f80426e9ebb"
47
+ },
48
+ {
49
+ "name": "NVIDIA A100-SXM4-80GB",
50
+ "memoryTotal": "85899345920",
51
+ "cudaCores": 6912,
52
+ "architecture": "Ampere",
53
+ "uuid": "GPU-184e67df-fbb6-0afd-f911-a3467fac75a5"
54
+ },
55
+ {
56
+ "name": "NVIDIA A100-SXM4-80GB",
57
+ "memoryTotal": "85899345920",
58
+ "cudaCores": 6912,
59
+ "architecture": "Ampere",
60
+ "uuid": "GPU-df0fae65-9451-78e9-7e2d-965a8a4f3427"
61
+ },
62
+ {
63
+ "name": "NVIDIA A100-SXM4-80GB",
64
+ "memoryTotal": "85899345920",
65
+ "cudaCores": 6912,
66
+ "architecture": "Ampere",
67
+ "uuid": "GPU-da84a93e-cacd-38a5-d0c2-c6f7e46f0285"
68
+ },
69
+ {
70
+ "name": "NVIDIA A100-SXM4-80GB",
71
+ "memoryTotal": "85899345920",
72
+ "cudaCores": 6912,
73
+ "architecture": "Ampere",
74
+ "uuid": "GPU-330b94fa-0daf-9a72-5a17-135faae05e20"
75
+ }
76
+ ],
77
+ "cudaVersion": "13.0",
78
+ "writerId": "lmk1a2m3lyyjdah2zmapfcvju4md0aif"
79
+ }
epfl_final/SparseAtt/wandb/run-20260823_140341-qctui00g/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff