rubentium commited on
Commit
f3c0c5a
·
verified ·
1 Parent(s): 22a8305

EPFL final checkpoints and W&B history 1351-1400 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 +13 -0
  2. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/files/requirements.txt +142 -0
  3. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/logs/debug-internal.log +11 -0
  4. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/logs/debug.log +19 -0
  5. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/run-6k91ldtc.wandb +3 -0
  6. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/files/requirements.txt +142 -0
  7. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/logs/debug-internal.log +11 -0
  8. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/logs/debug.log +19 -0
  9. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/run-8km2sc23.wandb +3 -0
  10. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ao5edbbv/run-ao5edbbv.wandb +3 -0
  11. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/files/requirements.txt +142 -0
  12. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/logs/debug-internal.log +11 -0
  13. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/logs/debug.log +19 -0
  14. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/run-hmdx78bp.wandb +3 -0
  15. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/files/requirements.txt +142 -0
  16. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/logs/debug-internal.log +11 -0
  17. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/logs/debug.log +19 -0
  18. epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/run-ut49ok16.wandb +3 -0
  19. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/files/requirements.txt +142 -0
  20. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/logs/debug-internal.log +11 -0
  21. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/logs/debug.log +19 -0
  22. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/run-3la36nzy.wandb +3 -0
  23. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/files/requirements.txt +142 -0
  24. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/logs/debug-internal.log +11 -0
  25. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/logs/debug.log +19 -0
  26. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/run-3uan03m3.wandb +3 -0
  27. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/files/requirements.txt +142 -0
  28. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/logs/debug-internal.log +11 -0
  29. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/logs/debug.log +19 -0
  30. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/run-ifgmkpzt.wandb +3 -0
  31. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/files/requirements.txt +142 -0
  32. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/logs/debug-internal.log +11 -0
  33. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/logs/debug.log +19 -0
  34. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/run-meicy3md.wandb +3 -0
  35. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/files/requirements.txt +142 -0
  36. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/logs/debug-internal.log +11 -0
  37. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/logs/debug.log +19 -0
  38. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/run-n6s56yfr.wandb +3 -0
  39. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/files/requirements.txt +142 -0
  40. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/logs/debug-internal.log +11 -0
  41. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/logs/debug.log +19 -0
  42. epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/run-xc6c08ps.wandb +3 -0
  43. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-ghfbqjah/logs/debug.log +19 -0
  44. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/files/requirements.txt +142 -0
  45. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/logs/debug-internal.log +11 -0
  46. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/logs/debug.log +19 -0
  47. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/run-kbn66izc.wandb +3 -0
  48. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/files/requirements.txt +142 -0
  49. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/logs/debug-internal.log +11 -0
  50. epfl_final/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/logs/debug.log +19 -0
.gitattributes CHANGED
@@ -176,3 +176,16 @@ epfl_final/SparseAtt/wandb/run-20260903_142704-7tfq6nwm/run-7tfq6nwm.wandb filte
176
  epfl_final/SparseAtt/wandb/run-20260903_191410-7tfq6nwm/run-7tfq6nwm.wandb filter=lfs diff=lfs merge=lfs -text
177
  epfl_final/SparseAtt/wandb/run-20260904_042057-7tfq6nwm/run-7tfq6nwm.wandb filter=lfs diff=lfs merge=lfs -text
178
  epfl_final/SparseAtt/wandb/offline-run-20260921_030515-1savcbqu/run-1savcbqu.wandb filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
176
  epfl_final/SparseAtt/wandb/run-20260903_191410-7tfq6nwm/run-7tfq6nwm.wandb filter=lfs diff=lfs merge=lfs -text
177
  epfl_final/SparseAtt/wandb/run-20260904_042057-7tfq6nwm/run-7tfq6nwm.wandb filter=lfs diff=lfs merge=lfs -text
178
  epfl_final/SparseAtt/wandb/offline-run-20260921_030515-1savcbqu/run-1savcbqu.wandb filter=lfs diff=lfs merge=lfs -text
179
+ epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ao5edbbv/run-ao5edbbv.wandb filter=lfs diff=lfs merge=lfs -text
180
+ epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/run-ut49ok16.wandb filter=lfs diff=lfs merge=lfs -text
181
+ epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/run-6k91ldtc.wandb filter=lfs diff=lfs merge=lfs -text
182
+ epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/run-hmdx78bp.wandb filter=lfs diff=lfs merge=lfs -text
183
+ epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/run-8km2sc23.wandb filter=lfs diff=lfs merge=lfs -text
184
+ epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/run-ifgmkpzt.wandb filter=lfs diff=lfs merge=lfs -text
185
+ epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/run-meicy3md.wandb filter=lfs diff=lfs merge=lfs -text
186
+ epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/run-n6s56yfr.wandb filter=lfs diff=lfs merge=lfs -text
187
+ epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/run-3la36nzy.wandb filter=lfs diff=lfs merge=lfs -text
188
+ epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/run-xc6c08ps.wandb filter=lfs diff=lfs merge=lfs -text
189
+ epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/run-3uan03m3.wandb filter=lfs diff=lfs merge=lfs -text
190
+ epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/run-kbn66izc.wandb filter=lfs diff=lfs merge=lfs -text
191
+ epfl_final/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/run-zgrsvkbi.wandb filter=lfs diff=lfs merge=lfs -text
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/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/offline-run-20260921_030515-6k91ldtc/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T03:05:15.627748328Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T03:05:15.627774557Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T03:05:15.977507774Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T03:05:15.977569276Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T03:05:15.977607514Z","level":"INFO","msg":"stream: created new stream","id":"6k91ldtc"}
6
+ {"time":"2026-09-21T03:05:15.977692221Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T03:05:15.986110077Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T03:05:15.986165781Z","level":"INFO","msg":"writer: started","stream_id":"6k91ldtc"}
9
+ {"time":"2026-09-21T03:05:15.986204038Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T03:05:16.013061227Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T03:05:16.0130764Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_setup.py:_flush():81] Configure stats pid to 4776
3
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/logs/debug.log
5
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/logs/debug-internal.log
6
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [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/models/ALIBI_20260921/365m_alibi_s44/run_20260921_030452', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 22000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 44, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s44', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed44'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 03:05:15,374 INFO MainThread:4776 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 03:05:15,616 INFO MainThread:4776 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 03:05:15,987 INFO MainThread:4776 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 03:05:15,988 INFO MainThread:4776 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 03:05:16,009 INFO MainThread:4776 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 03:05:16,014 INFO MainThread:4776 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 03:05:16,229 INFO MainThread:4776 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 03:05:16,229 INFO MainThread:4776 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 03:05:16,229 INFO MainThread:4776 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 03:05:16,230 INFO MainThread:4776 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 03:05:16,235 INFO MainThread:4776 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-6k91ldtc/run-6k91ldtc.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9fa5a8893b0d64471e1060bb2634231d6cfeb06f87b5f0bdf963159327fa592
3
+ size 9502720
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/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/offline-run-20260921_030515-8km2sc23/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T03:05:15.647686351Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T03:05:15.647714233Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T03:05:16.001225716Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T03:05:16.001272596Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T03:05:16.001306587Z","level":"INFO","msg":"stream: created new stream","id":"8km2sc23"}
6
+ {"time":"2026-09-21T03:05:16.001376873Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T03:05:16.007121864Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T03:05:16.007230587Z","level":"INFO","msg":"writer: started","stream_id":"8km2sc23"}
9
+ {"time":"2026-09-21T03:05:16.007241514Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T03:05:16.028513461Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T03:05:16.028534102Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 03:05:15,380 INFO MainThread:4779 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 03:05:15,380 INFO MainThread:4779 [wandb_setup.py:_flush():81] Configure stats pid to 4779
3
+ 2026-09-21 03:05:15,380 INFO MainThread:4779 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 03:05:15,380 INFO MainThread:4779 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/logs/debug.log
5
+ 2026-09-21 03:05:15,380 INFO MainThread:4779 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/logs/debug-internal.log
6
+ 2026-09-21 03:05:15,380 INFO MainThread:4779 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 03:05:15,381 INFO MainThread:4779 [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/models/ALIBI_20260921/365m_alibi_s47/run_20260921_030452', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 22000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 47, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s47', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed47'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 03:05:15,381 INFO MainThread:4779 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 03:05:15,636 INFO MainThread:4779 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 03:05:16,008 INFO MainThread:4779 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 03:05:16,010 INFO MainThread:4779 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 03:05:16,025 INFO MainThread:4779 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 03:05:16,030 INFO MainThread:4779 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 03:05:16,260 INFO MainThread:4779 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 03:05:16,260 INFO MainThread:4779 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 03:05:16,260 INFO MainThread:4779 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 03:05:16,260 INFO MainThread:4779 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 03:05:16,262 INFO MainThread:4779 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-8km2sc23/run-8km2sc23.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f20ea78ff25351a9482d7556532b12dfc04bb9feccf50b508a5d996379de376
3
+ size 9502720
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ao5edbbv/run-ao5edbbv.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a47ccd47018e8925ebe18a6bcc6a903833a70cb17c477052d08acb8df1e90288
3
+ size 9502720
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/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/offline-run-20260921_030515-hmdx78bp/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T03:05:15.644133914Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T03:05:15.644161065Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T03:05:15.995507455Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T03:05:15.995566224Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T03:05:15.995608958Z","level":"INFO","msg":"stream: created new stream","id":"hmdx78bp"}
6
+ {"time":"2026-09-21T03:05:15.995691291Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T03:05:15.997829013Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T03:05:15.997901802Z","level":"INFO","msg":"writer: started","stream_id":"hmdx78bp"}
9
+ {"time":"2026-09-21T03:05:15.997918447Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T03:05:16.02085118Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T03:05:16.020873704Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_setup.py:_flush():81] Configure stats pid to 4777
3
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/logs/debug.log
5
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/logs/debug-internal.log
6
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [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/models/ALIBI_20260921/365m_alibi_s45/run_20260921_030452', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 22000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 45, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s45', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed45'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 03:05:15,378 INFO MainThread:4777 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 03:05:15,632 INFO MainThread:4777 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 03:05:15,998 INFO MainThread:4777 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 03:05:16,001 INFO MainThread:4777 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 03:05:16,018 INFO MainThread:4777 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 03:05:16,023 INFO MainThread:4777 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 03:05:16,249 INFO MainThread:4777 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 03:05:16,249 INFO MainThread:4777 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 03:05:16,249 INFO MainThread:4777 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 03:05:16,249 INFO MainThread:4777 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 03:05:16,251 INFO MainThread:4777 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-hmdx78bp/run-hmdx78bp.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:529c1489903aaab749fa0277af1d0a0149c7354fdf132a44620412bbaa1cc1d3
3
+ size 9502720
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/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/offline-run-20260921_030515-ut49ok16/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T03:05:15.626653773Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T03:05:15.626680303Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T03:05:15.981530157Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T03:05:15.981590878Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T03:05:15.981626161Z","level":"INFO","msg":"stream: created new stream","id":"ut49ok16"}
6
+ {"time":"2026-09-21T03:05:15.981707033Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T03:05:15.99192309Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T03:05:15.991947527Z","level":"INFO","msg":"writer: started","stream_id":"ut49ok16"}
9
+ {"time":"2026-09-21T03:05:15.991976481Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T03:05:16.024492801Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T03:05:16.0245186Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 03:05:15,152 INFO MainThread:4775 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [wandb_setup.py:_flush():81] Configure stats pid to 4775
3
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/logs/debug.log
5
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/logs/debug-internal.log
6
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [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/models/ALIBI_20260921/365m_alibi_s43/run_20260921_030452', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 22000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 43, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s43', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed43'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 03:05:15,153 INFO MainThread:4775 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 03:05:15,613 INFO MainThread:4775 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 03:05:15,992 INFO MainThread:4775 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 03:05:15,998 INFO MainThread:4775 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 03:05:16,020 INFO MainThread:4775 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 03:05:16,028 INFO MainThread:4775 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 03:05:16,251 INFO MainThread:4775 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 03:05:16,251 INFO MainThread:4775 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 03:05:16,251 INFO MainThread:4775 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 03:05:16,251 INFO MainThread:4775 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 03:05:16,253 INFO MainThread:4775 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_030515-ut49ok16/run-ut49ok16.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:652398a1a92d2ce4c8f034a80d6119d7e6e8370b537dc3bfb78449df8b3d97dc
3
+ size 9502720
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/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/offline-run-20260921_131249-3la36nzy/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T13:12:50.483986586Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T13:12:50.484020637Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T13:12:50.838023892Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T13:12:50.838063321Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T13:12:50.838094006Z","level":"INFO","msg":"stream: created new stream","id":"3la36nzy"}
6
+ {"time":"2026-09-21T13:12:50.838213515Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T13:12:50.841655341Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T13:12:50.841683162Z","level":"INFO","msg":"writer: started","stream_id":"3la36nzy"}
9
+ {"time":"2026-09-21T13:12:50.841732275Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T13:12:50.870894371Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T13:12:50.87092038Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 13:12:49,924 INFO MainThread:356 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 13:12:49,924 INFO MainThread:356 [wandb_setup.py:_flush():81] Configure stats pid to 356
3
+ 2026-09-21 13:12:49,924 INFO MainThread:356 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 13:12:49,924 INFO MainThread:356 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/logs/debug.log
5
+ 2026-09-21 13:12:49,925 INFO MainThread:356 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/logs/debug-internal.log
6
+ 2026-09-21 13:12:49,925 INFO MainThread:356 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 13:12:49,925 INFO MainThread:356 [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/models/ALIBI_20260921/365m_alibi_s45/run_20260921_131213', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 45, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s45', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed45'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 13:12:49,925 INFO MainThread:356 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 13:12:50,469 INFO MainThread:356 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 13:12:50,842 INFO MainThread:356 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 13:12:50,847 INFO MainThread:356 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 13:12:50,867 INFO MainThread:356 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 13:12:50,873 INFO MainThread:356 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 13:12:51,121 INFO MainThread:356 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 13:12:51,121 INFO MainThread:356 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 13:12:51,121 INFO MainThread:356 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 13:12:51,121 INFO MainThread:356 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 13:12:51,127 INFO MainThread:356 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3la36nzy/run-3la36nzy.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:50d40438b665e32cf8402c79131dd92f63553951050e3e7266dc0c35d6e0b96c
3
+ size 622592
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/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/offline-run-20260921_131249-3uan03m3/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T13:12:50.483874479Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T13:12:50.483901889Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T13:12:50.841088435Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T13:12:50.841137017Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T13:12:50.841184688Z","level":"INFO","msg":"stream: created new stream","id":"3uan03m3"}
6
+ {"time":"2026-09-21T13:12:50.841265799Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T13:12:50.84313623Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T13:12:50.843167918Z","level":"INFO","msg":"writer: started","stream_id":"3uan03m3"}
9
+ {"time":"2026-09-21T13:12:50.84320998Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T13:12:50.870785509Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T13:12:50.870814782Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 13:12:49,929 INFO MainThread:353 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 13:12:49,929 INFO MainThread:353 [wandb_setup.py:_flush():81] Configure stats pid to 353
3
+ 2026-09-21 13:12:49,929 INFO MainThread:353 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 13:12:49,929 INFO MainThread:353 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/logs/debug.log
5
+ 2026-09-21 13:12:49,930 INFO MainThread:353 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/logs/debug-internal.log
6
+ 2026-09-21 13:12:49,930 INFO MainThread:353 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 13:12:49,930 INFO MainThread:353 [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/models/ALIBI_20260921/365m_alibi_s42/run_20260921_131213', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 42, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s42', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed42'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 13:12:49,930 INFO MainThread:353 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 13:12:50,469 INFO MainThread:353 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 13:12:50,844 INFO MainThread:353 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 13:12:50,847 INFO MainThread:353 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 13:12:50,867 INFO MainThread:353 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 13:12:50,873 INFO MainThread:353 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 13:12:51,108 INFO MainThread:353 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 13:12:51,108 INFO MainThread:353 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 13:12:51,108 INFO MainThread:353 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 13:12:51,108 INFO MainThread:353 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 13:12:51,115 INFO MainThread:353 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-3uan03m3/run-3uan03m3.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8131887d0f9a50fc1d08591b27b346f6ac4586d89dfbf1fd0cf022321587af0f
3
+ size 622592
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/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/offline-run-20260921_131249-ifgmkpzt/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T13:12:50.483216577Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T13:12:50.483239Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T13:12:50.836243114Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T13:12:50.836287921Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T13:12:50.836332528Z","level":"INFO","msg":"stream: created new stream","id":"ifgmkpzt"}
6
+ {"time":"2026-09-21T13:12:50.836402232Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T13:12:50.839011786Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T13:12:50.839030384Z","level":"INFO","msg":"writer: started","stream_id":"ifgmkpzt"}
9
+ {"time":"2026-09-21T13:12:50.839085496Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T13:12:50.870954671Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T13:12:50.870977545Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_setup.py:_flush():81] Configure stats pid to 355
3
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/logs/debug.log
5
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/logs/debug-internal.log
6
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [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/models/ALIBI_20260921/365m_alibi_s44/run_20260921_131213', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 44, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s44', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed44'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 13:12:49,902 INFO MainThread:355 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 13:12:50,469 INFO MainThread:355 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 13:12:50,839 INFO MainThread:355 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 13:12:50,843 INFO MainThread:355 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 13:12:50,867 INFO MainThread:355 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 13:12:50,873 INFO MainThread:355 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 13:12:51,130 INFO MainThread:355 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 13:12:51,130 INFO MainThread:355 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 13:12:51,130 INFO MainThread:355 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 13:12:51,131 INFO MainThread:355 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 13:12:51,133 INFO MainThread:355 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-ifgmkpzt/run-ifgmkpzt.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b50e684727aea5e38ea564a5e13082ab3e25791c61b34c009b307cc1a7eb8bbd
3
+ size 622592
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/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/offline-run-20260921_131249-meicy3md/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T13:12:50.483553329Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T13:12:50.48358054Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T13:12:50.839693533Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T13:12:50.839788525Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T13:12:50.839831339Z","level":"INFO","msg":"stream: created new stream","id":"meicy3md"}
6
+ {"time":"2026-09-21T13:12:50.83995932Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T13:12:50.843120577Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T13:12:50.843209209Z","level":"INFO","msg":"writer: started","stream_id":"meicy3md"}
9
+ {"time":"2026-09-21T13:12:50.843275598Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T13:12:50.87087387Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T13:12:50.870909333Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 13:12:49,918 INFO MainThread:357 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 13:12:49,918 INFO MainThread:357 [wandb_setup.py:_flush():81] Configure stats pid to 357
3
+ 2026-09-21 13:12:49,918 INFO MainThread:357 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 13:12:49,919 INFO MainThread:357 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/logs/debug.log
5
+ 2026-09-21 13:12:49,919 INFO MainThread:357 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/logs/debug-internal.log
6
+ 2026-09-21 13:12:49,919 INFO MainThread:357 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 13:12:49,919 INFO MainThread:357 [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/models/ALIBI_20260921/365m_alibi_s46/run_20260921_131213', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 46, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s46', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed46'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 13:12:49,919 INFO MainThread:357 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 13:12:50,470 INFO MainThread:357 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 13:12:50,844 INFO MainThread:357 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 13:12:50,849 INFO MainThread:357 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 13:12:50,867 INFO MainThread:357 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 13:12:50,873 INFO MainThread:357 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 13:12:51,111 INFO MainThread:357 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 13:12:51,111 INFO MainThread:357 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 13:12:51,111 INFO MainThread:357 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 13:12:51,111 INFO MainThread:357 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 13:12:51,115 INFO MainThread:357 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-meicy3md/run-meicy3md.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78693de0a766c541b0b8c83847e80ef9f3fb9a094f3b14113b5310df63b79b4c
3
+ size 622592
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/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/offline-run-20260921_131249-n6s56yfr/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T13:12:50.484598128Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T13:12:50.484624147Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T13:12:50.831760545Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T13:12:50.831852012Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T13:12:50.831921986Z","level":"INFO","msg":"stream: created new stream","id":"n6s56yfr"}
6
+ {"time":"2026-09-21T13:12:50.832074734Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T13:12:50.834780502Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T13:12:50.83493327Z","level":"INFO","msg":"writer: started","stream_id":"n6s56yfr"}
9
+ {"time":"2026-09-21T13:12:50.834970566Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T13:12:50.870640873Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T13:12:50.870664909Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 13:12:49,926 INFO MainThread:354 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [wandb_setup.py:_flush():81] Configure stats pid to 354
3
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/logs/debug.log
5
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/logs/debug-internal.log
6
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [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/models/ALIBI_20260921/365m_alibi_s43/run_20260921_131213', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 43, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s43', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed43'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 13:12:49,927 INFO MainThread:354 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 13:12:50,469 INFO MainThread:354 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 13:12:50,836 INFO MainThread:354 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 13:12:50,839 INFO MainThread:354 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 13:12:50,867 INFO MainThread:354 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 13:12:50,873 INFO MainThread:354 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 13:12:51,107 INFO MainThread:354 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 13:12:51,107 INFO MainThread:354 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 13:12:51,107 INFO MainThread:354 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 13:12:51,107 INFO MainThread:354 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 13:12:51,115 INFO MainThread:354 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-n6s56yfr/run-n6s56yfr.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c2e5252108c00a233fb04e5c68db422bc976a5a19e24447d297f1d4ab141507
3
+ size 622592
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/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/offline-run-20260921_131249-xc6c08ps/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T13:12:50.484437969Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T13:12:50.484464449Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T13:12:50.838624218Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T13:12:50.838691207Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T13:12:50.838730746Z","level":"INFO","msg":"stream: created new stream","id":"xc6c08ps"}
6
+ {"time":"2026-09-21T13:12:50.838795473Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T13:12:50.840941876Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T13:12:50.840979752Z","level":"INFO","msg":"writer: started","stream_id":"xc6c08ps"}
9
+ {"time":"2026-09-21T13:12:50.841010919Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T13:12:50.870744397Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T13:12:50.870771448Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_setup.py:_flush():81] Configure stats pid to 358
3
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/logs/debug.log
5
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/logs/debug-internal.log
6
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [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/models/ALIBI_20260921/365m_alibi_s47/run_20260921_131213', 'resume_from_checkpoint': None, 'force_new_run': True, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 47, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s47', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed47'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 13:12:49,928 INFO MainThread:358 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 13:12:50,469 INFO MainThread:358 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 13:12:50,841 INFO MainThread:358 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 13:12:50,846 INFO MainThread:358 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 13:12:50,867 INFO MainThread:358 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 13:12:50,873 INFO MainThread:358 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 13:12:51,127 INFO MainThread:358 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 13:12:51,127 INFO MainThread:358 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 13:12:51,127 INFO MainThread:358 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 13:12:51,127 INFO MainThread:358 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 13:12:51,129 INFO MainThread:358 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_131249-xc6c08ps/run-xc6c08ps.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e9483abf7a5154cc7067a0597415e245fd9c190499f29c23aa5f133ed608d76c
3
+ size 622592
epfl_final/SparseAtt/wandb/offline-run-20260921_143024-ghfbqjah/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 14:30:24,657 INFO MainThread:70 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 14:30:24,657 INFO MainThread:70 [wandb_setup.py:_flush():81] Configure stats pid to 70
3
+ 2026-09-21 14:30:24,658 INFO MainThread:70 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 14:30:24,658 INFO MainThread:70 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_143024-ghfbqjah/logs/debug.log
5
+ 2026-09-21 14:30:24,658 INFO MainThread:70 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_143024-ghfbqjah/logs/debug-internal.log
6
+ 2026-09-21 14:30:24,658 INFO MainThread:70 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 14:30:24,658 INFO MainThread:70 [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/models/ALIBI_20260921/365m_alibi_s45/run_20260921_142951', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 45, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s45', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed45'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 14:30:24,658 INFO MainThread:70 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 14:30:25,191 INFO MainThread:70 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 14:30:25,568 INFO MainThread:70 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 14:30:25,570 INFO MainThread:70 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 14:30:25,584 INFO MainThread:70 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 14:30:25,589 INFO MainThread:70 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 14:30:25,756 INFO MainThread:70 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 14:30:25,756 INFO MainThread:70 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 14:30:25,757 INFO MainThread:70 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 14:30:25,757 INFO MainThread:70 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 14:30:25,763 INFO MainThread:70 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/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/offline-run-20260921_143024-kbn66izc/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T14:30:25.206945921Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T14:30:25.206970507Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T14:30:25.559404152Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T14:30:25.559465233Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T14:30:25.559507276Z","level":"INFO","msg":"stream: created new stream","id":"kbn66izc"}
6
+ {"time":"2026-09-21T14:30:25.559588297Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T14:30:25.561924124Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T14:30:25.561956583Z","level":"INFO","msg":"writer: started","stream_id":"kbn66izc"}
9
+ {"time":"2026-09-21T14:30:25.562002852Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T14:30:25.588409287Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T14:30:25.588434474Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_setup.py:_flush():81] Configure stats pid to 69
3
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/logs/debug.log
5
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/logs/debug-internal.log
6
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [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/models/ALIBI_20260921/365m_alibi_s44/run_20260921_142951', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 44, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s44', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed44'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 14:30:24,644 INFO MainThread:69 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 14:30:25,191 INFO MainThread:69 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 14:30:25,562 INFO MainThread:69 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 14:30:25,565 INFO MainThread:69 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 14:30:25,585 INFO MainThread:69 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 14:30:25,591 INFO MainThread:69 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 14:30:25,790 INFO MainThread:69 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 14:30:25,790 INFO MainThread:69 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 14:30:25,790 INFO MainThread:69 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 14:30:25,791 INFO MainThread:69 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 14:30:25,792 INFO MainThread:69 [wandb_init.py:init():1094] run started, returning control to user process
epfl_final/SparseAtt/wandb/offline-run-20260921_143024-kbn66izc/run-kbn66izc.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f8d7767727a9fb0c9c652f571b990bbae7a5074ddfc36a5ebeed9b64e493455
3
+ size 1376256
epfl_final/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/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/offline-run-20260921_143024-zgrsvkbi/logs/debug-internal.log ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-09-21T14:30:25.206626644Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-09-21T14:30:25.20665084Z","level":"INFO","msg":"stream: starting","core version":"0.26.1"}
3
+ {"time":"2026-09-21T14:30:25.554203978Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
4
+ {"time":"2026-09-21T14:30:25.554264238Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
5
+ {"time":"2026-09-21T14:30:25.554298309Z","level":"INFO","msg":"stream: created new stream","id":"zgrsvkbi"}
6
+ {"time":"2026-09-21T14:30:25.554400832Z","level":"INFO","msg":"handler: started"}
7
+ {"time":"2026-09-21T14:30:25.556590301Z","level":"INFO","msg":"stream: started"}
8
+ {"time":"2026-09-21T14:30:25.556631613Z","level":"INFO","msg":"writer: started","stream_id":"zgrsvkbi"}
9
+ {"time":"2026-09-21T14:30:25.556676189Z","level":"INFO","msg":"sender: started"}
10
+ {"time":"2026-09-21T14:30:25.588291171Z","level":"WARN","msg":"featurechecker: GraphQL client is nil, skipping feature loading"}
11
+ {"time":"2026-09-21T14:30:25.588316949Z","level":"WARN","msg":"runupserter: server does not expand metric globs but the x_server_side_expand_glob_metrics setting is set; ignoring"}
epfl_final/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/logs/debug.log ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-09-21 14:30:24,658 INFO MainThread:68 [wandb_setup.py:_flush():81] Current SDK version is 0.26.1
2
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [wandb_setup.py:_flush():81] Configure stats pid to 68
3
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [wandb_init.py:setup_run_log_directory():723] Logging user logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/logs/debug.log
5
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [wandb_init.py:setup_run_log_directory():724] Logging internal logs to /mloscratch/homes/navasard/sparse/SparseAtt/wandb/offline-run-20260921_143024-zgrsvkbi/logs/debug-internal.log
6
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [wandb_init.py:init():850] calling init triggers
7
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [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/models/ALIBI_20260921/365m_alibi_s43/run_20260921_142951', 'resume_from_checkpoint': None, 'force_new_run': False, 'num_epochs': 1, 'max_steps': 44000, 'per_device_batch_size': 2, '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': 1000, 'eval_steps': 250, 'eval_batches': 50, 'logging_steps': 100, 'seed': 43, 'verbose': True, 'checkpoint_activations': False, 'resume_micro_step_override': None, 'strategy': 'ddp', 'mixed_precision': 'bf16', 'wandb_enabled': True, 'wandb_project': 'stochastic_sparsity', 'wandb_name': '365m alibi 8k s43', 'wandb_tags': ['qwen3', '365m', '8k', 'scaleladder', 'alibi', 'seed43'], '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, 'positional': 'alibi', 'attention_type': 'dense', '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': False, 'dense_step_intervals': None, '_wandb': {}}
9
+ 2026-09-21 14:30:24,659 INFO MainThread:68 [wandb_init.py:init():898] starting backend
10
+ 2026-09-21 14:30:25,191 INFO MainThread:68 [wandb_init.py:init():913] sending inform_init request
11
+ 2026-09-21 14:30:25,557 INFO MainThread:68 [wandb_init.py:init():918] backend started and connected
12
+ 2026-09-21 14:30:25,558 INFO MainThread:68 [wandb_init.py:init():988] updated telemetry
13
+ 2026-09-21 14:30:25,585 INFO MainThread:68 [wandb_init.py:init():1011] communicating run to backend with 90.0 second timeout
14
+ 2026-09-21 14:30:25,590 INFO MainThread:68 [wandb_init.py:init():1056] starting run threads in backend
15
+ 2026-09-21 14:30:25,804 INFO MainThread:68 [wandb_run.py:_console_start():2554] atexit reg
16
+ 2026-09-21 14:30:25,805 INFO MainThread:68 [wandb_run.py:_redirect():2403] redirect: wrap_raw
17
+ 2026-09-21 14:30:25,805 INFO MainThread:68 [wandb_run.py:_redirect():2472] Wrapping output streams.
18
+ 2026-09-21 14:30:25,805 INFO MainThread:68 [wandb_run.py:_redirect():2495] Redirects installed.
19
+ 2026-09-21 14:30:25,806 INFO MainThread:68 [wandb_init.py:init():1094] run started, returning control to user process