appvoid commited on
Commit
e710dd9
·
verified ·
1 Parent(s): 89f3496

Qwen90M step 400000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 397106,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "a47c91999dd0ebcdc282336b24f357db9c84c594fb3b2642a3894a2b6a3275cc",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "f7cd41c0ad064fdb32d106a04f2baeeeeed5930763f4ca1400aa877b835cb4d4",
12
- "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
- "performance.json": "e3923fa86352f6d878fd8d37b6b9fc374577f52be522203b0b865badf630da15",
14
- "shard_scheduler.json": "73efbc129590a5a3737fdb30c517ab49d4777d335d99a458b160010d41de4d70",
15
- "coverage.json": "861a5c400ef29ce2ebe8dbeb84b40a4ce490ac63f7b453e5e8d6bfed66907dad",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 400000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "a131696398b13dbcafb16c35950eb04cbd669d17fb0d4873cd34a9793ddb041d",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "0883783165f69503eba3b62bf8c9c7a6e664527b2bcaeee310413d02bf169eb3",
12
+ "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6",
13
+ "performance.json": "a6ae17d5ca4d76a29c6e069dddb352125dfbf6f18e59e9553ab94c741a962b18",
14
+ "shard_scheduler.json": "c40659570cd21160457c25f9debd82b3f0d1ed40efab78c7456bc3d67e11e8d3",
15
+ "coverage.json": "ccc6e483b973caea3627baa83892ef110132ad12b30133bfa601b18a54ee08eb",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 397106,
3
  "stats": {
4
- "climbmix/blocks": 14295816,
5
- "climbmix/targets": 23897781997,
6
- "climbmix/input_tokens": 23899295733,
7
- "ultra_qa/blocks": 5360931,
8
- "ultra_qa/targets": 8745343329,
9
- "ultra_qa/input_tokens": 8745360186,
10
- "cortex/blocks": 5360931,
11
- "cortex/targets": 203917069,
12
- "cortex/input_tokens": 8521943593,
13
- "ultra_style/blocks": 5360931,
14
- "ultra_style/targets": 8479672034,
15
- "ultra_style/input_tokens": 8480116529,
16
- "openmath/blocks": 3573954,
17
- "openmath/targets": 5522855410,
18
- "openmath/input_tokens": 5524695746,
19
- "rewrite6/blocks": 1786977,
20
- "rewrite6/targets": 2929490049,
21
- "rewrite6/input_tokens": 2932701855
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1957,
26
- "row": 3523,
27
  "start": {
28
- "file": 94,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 60,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 0,
37
- "documents": 8019395
38
  },
39
  "rewrite6": {
40
- "page": 13,
41
- "row": 3512,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 53248
47
  }
48
  },
49
  "epoch": 22,
50
  "offset": 0,
51
- "documents": 4575648
52
  },
53
  "openmath": {
54
- "page": 1223,
55
- "row": 2487,
56
  "start": {
57
  "file": 11,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 206,
61
- "row": 258
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 5011895
67
  },
68
  "ultra_qa": {
69
- "page": 529,
70
- "row": 178,
71
  "start": {
72
  "file": 4,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
- "row": 23796
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 4096,
81
- "documents": 2166962
82
  },
83
  "ultra_style": {
84
- "page": 807,
85
- "row": 1879,
86
  "start": {
87
  "file": 4,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 6,
91
- "row": 55903
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 3307351
97
  },
98
  "cortex": {
99
- "page": 103976,
100
- "row": 115,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 13309043
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 400000,
3
  "stats": {
4
+ "climbmix/blocks": 14400000,
5
+ "climbmix/targets": 24072071047,
6
+ "climbmix/input_tokens": 24073595873,
7
+ "ultra_qa/blocks": 5400000,
8
+ "ultra_qa/targets": 8809069980,
9
+ "ultra_qa/input_tokens": 8809086942,
10
+ "cortex/blocks": 5400000,
11
+ "cortex/targets": 205392043,
12
+ "cortex/input_tokens": 8584044366,
13
+ "ultra_style/blocks": 5400000,
14
+ "ultra_style/targets": 8541509026,
15
+ "ultra_style/input_tokens": 8541956871,
16
+ "openmath/blocks": 3600000,
17
+ "openmath/targets": 5563081132,
18
+ "openmath/input_tokens": 5564934909,
19
+ "rewrite6/blocks": 1800000,
20
+ "rewrite6/targets": 2950834067,
21
+ "rewrite6/input_tokens": 2954069608
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1972,
26
+ "row": 642,
27
  "start": {
28
+ "file": 95,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 38,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 0,
37
+ "documents": 8077954
38
  },
39
  "rewrite6": {
40
+ "page": 22,
41
+ "row": 377,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 90112
47
  }
48
  },
49
  "epoch": 22,
50
  "offset": 0,
51
+ "documents": 4609377
52
  },
53
  "openmath": {
54
+ "page": 1232,
55
+ "row": 2237,
56
  "start": {
57
  "file": 11,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 243,
61
+ "row": 122
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 5048509
67
  },
68
  "ultra_qa": {
69
+ "page": 532,
70
+ "row": 3708,
71
  "start": {
72
  "file": 4,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
+ "row": 36084
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 0,
81
+ "documents": 2182780
82
  },
83
  "ultra_style": {
84
+ "page": 813,
85
+ "row": 1709,
86
  "start": {
87
  "file": 4,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 6,
91
+ "row": 80479
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 4096,
96
+ "documents": 3331757
97
  },
98
  "cortex": {
99
+ "page": 104734,
100
+ "row": 30,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 13405982
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 397106, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "a47c91999dd0ebcdc282336b24f357db9c84c594fb3b2642a3894a2b6a3275cc", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "f7cd41c0ad064fdb32d106a04f2baeeeeed5930763f4ca1400aa877b835cb4d4", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "e3923fa86352f6d878fd8d37b6b9fc374577f52be522203b0b865badf630da15", "shard_scheduler.json": "73efbc129590a5a3737fdb30c517ab49d4777d335d99a458b160010d41de4d70", "coverage.json": "861a5c400ef29ce2ebe8dbeb84b40a4ce490ac63f7b453e5e8d6bfed66907dad", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000397106"}
 
1
+ {"step": 400000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "a131696398b13dbcafb16c35950eb04cbd669d17fb0d4873cd34a9793ddb041d", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "0883783165f69503eba3b62bf8c9c7a6e664527b2bcaeee310413d02bf169eb3", "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6", "performance.json": "a6ae17d5ca4d76a29c6e069dddb352125dfbf6f18e59e9553ab94c741a962b18", "shard_scheduler.json": "c40659570cd21160457c25f9debd82b3f0d1ed40efab78c7456bc3d67e11e8d3", "coverage.json": "ccc6e483b973caea3627baa83892ef110132ad12b30133bfa601b18a54ee08eb", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000400000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a47c91999dd0ebcdc282336b24f357db9c84c594fb3b2642a3894a2b6a3275cc
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a131696398b13dbcafb16c35950eb04cbd669d17fb0d4873cd34a9793ddb041d
3
  size 360622744
performance.json CHANGED
@@ -39,148 +39,148 @@
39
  "compile_model": true,
40
  "autotune_microbatch": true,
41
  "attention_backend": "flash",
42
- "vram_fraction": 0.95
43
  },
44
  "source": "e6d135a4487a9fd1907e12903cda79e75fd61305fad149fda7ec5281279dc18b",
45
  "requested_microbatch_ceiling": 90
46
  },
47
- "cache_reused": true,
48
  "attention": "PyTorch FlashAttention (forced; forward/backward verified)",
49
  "selected": {
50
- "microbatch": 86,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9009280120000085,
53
- "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9565700770000376,
56
- "input_tokens_per_second": 195496.19687038695
57
  },
58
  "measurements": [
59
  {
60
  "microbatch": 8,
61
  "mode": "eager",
62
- "microbatch_seconds": 0.15782159200000478,
63
- "peak_gb": 14.079077243804932,
64
  "fits_budget": true,
65
- "estimated_update_seconds": 1.793421506500053,
66
- "input_tokens_per_second": 103813.42497165725
67
  },
68
  {
69
  "microbatch": 16,
70
  "mode": "eager",
71
- "microbatch_seconds": 0.3450626039999918,
72
- "peak_gb": 27.397329807281494,
73
  "fits_budget": true,
74
- "estimated_update_seconds": 1.9234990809999601,
75
- "input_tokens_per_second": 94962.47817106481
76
  },
77
  {
78
  "microbatch": 32,
79
  "mode": "eager",
80
- "microbatch_seconds": 0.7818640030000097,
81
- "peak_gb": 54.14767265319824,
82
  "fits_budget": true,
83
- "estimated_update_seconds": 2.185234111500023,
84
- "input_tokens_per_second": 83820.20370363462
85
  },
86
  {
87
  "microbatch": 48,
88
  "mode": "eager",
89
- "microbatch_seconds": 1.1685763079999845,
90
- "peak_gb": 80.94732475280762,
91
  "fits_budget": true,
92
- "estimated_update_seconds": 2.190914332499986,
93
- "input_tokens_per_second": 84122.87612457851
94
  },
95
  {
96
  "microbatch": 8,
97
  "mode": "compiled",
98
- "microbatch_seconds": 0.08803204299999834,
99
- "peak_gb": 8.428553104400635,
100
  "fits_budget": true,
101
- "estimated_update_seconds": 1.0060347614999898,
102
- "input_tokens_per_second": 186114.0494035826
103
  },
104
  {
105
  "microbatch": 16,
106
  "mode": "compiled",
107
- "microbatch_seconds": 0.1704058820000114,
108
- "peak_gb": 16.191233158111572,
109
  "fits_budget": true,
110
- "estimated_update_seconds": 0.9576335650000658,
111
- "input_tokens_per_second": 192293.8317352086
112
  },
113
  {
114
  "microbatch": 32,
115
  "mode": "compiled",
116
- "microbatch_seconds": 0.34082671699997036,
117
- "peak_gb": 31.810625553131104,
118
  "fits_budget": true,
119
- "estimated_update_seconds": 0.9493586249999737,
120
- "input_tokens_per_second": 192285.39527904938
121
  },
122
  {
123
  "microbatch": 48,
124
  "mode": "compiled",
125
- "microbatch_seconds": 0.5061537779999981,
126
- "peak_gb": 47.43971490859985,
127
  "fits_budget": true,
128
- "estimated_update_seconds": 0.9468006980000041,
129
- "input_tokens_per_second": 194217.6553308279
130
  },
131
  {
132
  "microbatch": 64,
133
  "mode": "compiled",
134
- "microbatch_seconds": 0.6658512800000267,
135
- "peak_gb": 63.073320388793945,
136
  "fits_budget": true,
137
- "estimated_update_seconds": 0.9295456565000109,
138
- "input_tokens_per_second": 196848.76178355434
139
  },
140
  {
141
  "microbatch": 72,
142
  "mode": "compiled",
143
- "microbatch_seconds": 0.7430151979999664,
144
- "peak_gb": 70.97307014465332,
145
  "fits_budget": true,
146
- "estimated_update_seconds": 0.9250642054999503,
147
- "input_tokens_per_second": 198456.23669195347
148
  },
149
  {
150
  "microbatch": 76,
151
  "mode": "compiled",
152
- "microbatch_seconds": 0.78068104700003,
153
- "peak_gb": 74.84185981750488,
154
  "fits_budget": true,
155
- "estimated_update_seconds": 0.9241739715000392,
156
- "input_tokens_per_second": 199374.63654089966
157
  },
158
  {
159
  "microbatch": 80,
160
  "mode": "compiled",
161
- "microbatch_seconds": 0.8210239410000213,
162
- "peak_gb": 78.72572612762451,
163
  "fits_budget": true,
164
- "estimated_update_seconds": 0.9229792625000073,
165
- "input_tokens_per_second": 199555.69115370748
166
  },
167
  {
168
  "microbatch": 86,
169
  "mode": "compiled",
170
- "microbatch_seconds": 0.8818629400000191,
171
- "peak_gb": 84.78368663787842,
172
  "fits_budget": true,
173
- "estimated_update_seconds": 0.9349941874999956,
174
- "input_tokens_per_second": 199722.64624250587
175
  },
176
  {
177
  "microbatch": 90,
178
  "mode": "compiled",
179
- "microbatch_seconds": 0.9259720020000373,
180
- "peak_gb": 88.6924238204956,
181
- "fits_budget": false,
182
- "estimated_update_seconds": 0.9259720020000373,
183
- "input_tokens_per_second": 199055.69455867045
184
  }
185
  ],
186
  "failures": [
@@ -216,8 +216,7 @@
216
  }
217
  ],
218
  "selection_policy": "max_batch",
219
- "budget_gb": 89.6964394569397,
220
  "reserve_gb": 2.1791439056396484,
221
- "note": "Forward/backward throughput; excludes optimizer, data, logging and checkpoint time.",
222
- "legacy_seed_verified": false
223
  }
 
39
  "compile_model": true,
40
  "autotune_microbatch": true,
41
  "attention_backend": "flash",
42
+ "vram_fraction": 0.98
43
  },
44
  "source": "e6d135a4487a9fd1907e12903cda79e75fd61305fad149fda7ec5281279dc18b",
45
  "requested_microbatch_ceiling": 90
46
  },
47
+ "cache_reused": false,
48
  "attention": "PyTorch FlashAttention (forced; forward/backward verified)",
49
  "selected": {
50
+ "microbatch": 90,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9302335780000135,
53
+ "peak_gb": 89.3772840499878,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9302335780000135,
56
+ "input_tokens_per_second": 198143.7827650609
57
  },
58
  "measurements": [
59
  {
60
  "microbatch": 8,
61
  "mode": "eager",
62
+ "microbatch_seconds": 0.15844926500000156,
63
+ "peak_gb": 14.764304637908936,
64
  "fits_budget": true,
65
+ "estimated_update_seconds": 1.8007989265000077,
66
+ "input_tokens_per_second": 103402.18365796673
67
  },
68
  {
69
  "microbatch": 16,
70
  "mode": "eager",
71
+ "microbatch_seconds": 0.34631374599999276,
72
+ "peak_gb": 28.084960460662842,
73
  "fits_budget": true,
74
+ "estimated_update_seconds": 1.929804849999968,
75
+ "input_tokens_per_second": 94619.40329680326
76
  },
77
  {
78
  "microbatch": 32,
79
  "mode": "eager",
80
+ "microbatch_seconds": 0.7831460159999892,
81
+ "peak_gb": 54.827155113220215,
82
  "fits_budget": true,
83
+ "estimated_update_seconds": 2.188403235499976,
84
+ "input_tokens_per_second": 83682.98971210101
85
  },
86
  {
87
  "microbatch": 48,
88
  "mode": "eager",
89
+ "microbatch_seconds": 1.1706683780000162,
90
+ "peak_gb": 81.63183498382568,
91
  "fits_budget": true,
92
+ "estimated_update_seconds": 2.1949380675000043,
93
+ "input_tokens_per_second": 83972.54239321277
94
  },
95
  {
96
  "microbatch": 8,
97
  "mode": "compiled",
98
+ "microbatch_seconds": 0.08827481200000875,
99
+ "peak_gb": 9.113307476043701,
100
  "fits_budget": true,
101
+ "estimated_update_seconds": 1.008851873500106,
102
+ "input_tokens_per_second": 185602.20779624404
103
  },
104
  {
105
  "microbatch": 16,
106
  "mode": "compiled",
107
+ "microbatch_seconds": 0.17047579999996287,
108
+ "peak_gb": 16.87287473678589,
109
  "fits_budget": true,
110
+ "estimated_update_seconds": 0.9571819019997747,
111
+ "input_tokens_per_second": 192214.96540862185
112
  },
113
  {
114
  "microbatch": 32,
115
  "mode": "compiled",
116
+ "microbatch_seconds": 0.34263174000000163,
117
+ "peak_gb": 32.49641752243042,
118
  "fits_budget": true,
119
+ "estimated_update_seconds": 0.9525396829999977,
120
+ "input_tokens_per_second": 191272.41393339593
121
  },
122
  {
123
  "microbatch": 48,
124
  "mode": "compiled",
125
+ "microbatch_seconds": 0.5075005140000144,
126
+ "peak_gb": 48.123157024383545,
127
  "fits_budget": true,
128
+ "estimated_update_seconds": 0.9488392155000156,
129
+ "input_tokens_per_second": 193702.2668709991
130
  },
131
  {
132
  "microbatch": 64,
133
  "mode": "compiled",
134
+ "microbatch_seconds": 0.6643874849999634,
135
+ "peak_gb": 63.757601737976074,
136
  "fits_budget": true,
137
+ "estimated_update_seconds": 0.926719353999971,
138
+ "input_tokens_per_second": 197282.46386219515
139
  },
140
  {
141
  "microbatch": 72,
142
  "mode": "compiled",
143
+ "microbatch_seconds": 0.7446585950000326,
144
+ "peak_gb": 71.65071392059326,
145
  "fits_budget": true,
146
+ "estimated_update_seconds": 0.9283865335000314,
147
+ "input_tokens_per_second": 198018.26097232322
148
  },
149
  {
150
  "microbatch": 76,
151
  "mode": "compiled",
152
+ "microbatch_seconds": 0.7858128550000174,
153
+ "peak_gb": 75.5242338180542,
154
  "fits_budget": true,
155
+ "estimated_update_seconds": 0.9300540555000225,
156
+ "input_tokens_per_second": 198072.60597689834
157
  },
158
  {
159
  "microbatch": 80,
160
  "mode": "compiled",
161
+ "microbatch_seconds": 0.8269688110000288,
162
+ "peak_gb": 79.40062236785889,
163
  "fits_budget": true,
164
+ "estimated_update_seconds": 0.9293605420000404,
165
+ "input_tokens_per_second": 198121.13567121493
166
  },
167
  {
168
  "microbatch": 86,
169
  "mode": "compiled",
170
+ "microbatch_seconds": 0.8882591199999865,
171
+ "peak_gb": 85.4687147140503,
172
  "fits_budget": true,
173
+ "estimated_update_seconds": 0.9413692904999778,
174
+ "input_tokens_per_second": 198284.4825730612
175
  },
176
  {
177
  "microbatch": 90,
178
  "mode": "compiled",
179
+ "microbatch_seconds": 0.9302335780000135,
180
+ "peak_gb": 89.3772840499878,
181
+ "fits_budget": true,
182
+ "estimated_update_seconds": 0.9302335780000135,
183
+ "input_tokens_per_second": 198143.7827650609
184
  }
185
  ],
186
  "failures": [
 
216
  }
217
  ],
218
  "selection_policy": "max_batch",
219
+ "budget_gb": 92.52170804977416,
220
  "reserve_gb": 2.1791439056396484,
221
+ "note": "Forward/backward throughput; excludes optimizer, data, logging and checkpoint time."
 
222
  }
shard_scheduler.json CHANGED
@@ -1,99 +1,99 @@
1
  {
2
  "version": 1,
3
  "startup_batch_seconds": [
4
- 17.862166522000052,
5
- 0.01599195200003578,
6
- 0.012736249999989013,
7
- 0.10922379700002693,
8
- 0.015912060000005113,
9
- 0.014246264000007614,
10
- 0.01397435499995936,
11
- 0.10559937200002878
12
  ],
13
  "source_benchmarks": [
14
  {
15
  "source": "climbmix",
16
  "rows": 256,
17
- "tokens": 721064,
18
- "bytes": 302776,
19
- "generation_seconds": 0.6739276440000594,
20
- "write_seconds": 0.00442227799999273,
21
- "read_seconds": 0.04565144100001817,
22
- "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 74809.39388603426,
26
- "required_pages_per_second_upper_estimate": 0.006484288659366078
27
  },
28
  {
29
  "source": "rewrite6",
30
  "rows": 256,
31
- "tokens": 156675,
32
- "bytes": 68694,
33
- "generation_seconds": 0.12810849499999222,
34
- "write_seconds": 0.001732062999963091,
35
- "read_seconds": 0.002349199000036606,
36
- "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
- "required_token_positions_per_second": 9351.174235754283,
40
- "required_pages_per_second_upper_estimate": 0.0037303232151564875
41
  },
42
  {
43
  "source": "openmath",
44
  "rows": 256,
45
- "tokens": 270174,
46
- "bytes": 90063,
47
- "generation_seconds": 0.6564187429999038,
48
- "write_seconds": 0.002473517999987962,
49
- "read_seconds": 0.0032515120000198294,
50
- "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
- "required_token_positions_per_second": 18702.348471508565,
54
- "required_pages_per_second_upper_estimate": 0.004326459168792279
55
  },
56
  {
57
  "source": "ultra_qa",
58
  "rows": 256,
59
- "tokens": 1026361,
60
- "bytes": 365167,
61
- "generation_seconds": 0.46967248600003586,
62
- "write_seconds": 0.004728889000034542,
63
- "read_seconds": 0.0050225659999796335,
64
- "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
- "required_token_positions_per_second": 28053.522707262844,
68
- "required_pages_per_second_upper_estimate": 0.0017083123474137538
69
  },
70
  {
71
  "source": "ultra_style",
72
  "rows": 256,
73
- "tokens": 811870,
74
- "bytes": 338806,
75
- "generation_seconds": 0.5208254859999215,
76
- "write_seconds": 0.004081093000081637,
77
- "read_seconds": 0.004787273000033565,
78
- "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
- "required_token_positions_per_second": 28053.522707262844,
82
- "required_pages_per_second_upper_estimate": 0.0021596378351262245
83
  },
84
  {
85
  "source": "cortex",
86
  "rows": 128,
87
- "tokens": 80537,
88
- "bytes": 22745,
89
- "generation_seconds": 0.07537080299994159,
90
- "write_seconds": 0.003962095000019872,
91
- "read_seconds": 0.042780299000014566,
92
- "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 21,
95
- "required_token_positions_per_second": 28053.522707262844,
96
- "required_pages_per_second_upper_estimate": 0.34833086292341214
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -108,460 +108,220 @@
108
  "local_shard_size_benchmark": [
109
  {
110
  "target_mib": 16,
111
- "actual_bytes": 16107520,
112
- "pages": 6,
113
- "packing_seconds": 0.03508346299997811,
114
- "read_seconds": 0.0015898530000413302,
115
- "io_bytes_per_second": 439216350.1111124,
116
  "target_reached": true
117
  },
118
  {
119
  "target_mib": 32,
120
- "actual_bytes": 18657280,
121
- "pages": 8,
122
- "packing_seconds": 0.04401178199998412,
123
- "read_seconds": 0.0016850460000341627,
124
- "io_bytes_per_second": 408283918.5247723,
125
  "target_reached": false
126
  },
127
  {
128
  "target_mib": 64,
129
- "actual_bytes": 18657280,
130
- "pages": 8,
131
- "packing_seconds": 0.04403130000002875,
132
- "read_seconds": 0.0015284270000393008,
133
- "io_bytes_per_second": 409512550.4148024,
134
  "target_reached": false
135
  },
136
  {
137
  "target_mib": 128,
138
- "actual_bytes": 18657280,
139
- "pages": 8,
140
- "packing_seconds": 0.04190044699998907,
141
- "read_seconds": 0.001583749000019452,
142
- "io_bytes_per_second": 429058870.0316856,
143
  "target_reached": false
144
  },
145
  {
146
  "target_mib": 256,
147
- "actual_bytes": 18657280,
148
- "pages": 8,
149
- "packing_seconds": 0.042332877000035296,
150
- "read_seconds": 0.0015621619999137693,
151
- "io_bytes_per_second": 425043021.37700915,
152
  "target_reached": false
153
  }
154
  ],
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
- "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
157
  "selected": {
158
- "prefetch_batches": 61,
159
  "prefetch_capacity_limit": 512,
160
- "batch_memory_estimate_bytes": 717798,
161
- "producer_batch_seconds": 0.32073739904717513,
162
- "producer_p95_seconds": 0.39812204299960285,
163
- "gpu_update_seconds": 0.9855446778825442,
164
- "producer_batches_per_second": 3.1178153934362878,
165
- "gpu_batches_per_second": 1.01466734328931,
166
- "producer_headroom": 3.072746367621404,
167
  "sustainable": true,
168
- "buffer_seconds": 60.118225350835196,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 78,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 370873.33853366604,
174
- "upload_bytes_per_second": 5789859.621793583,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 79.58078495986449,
179
  "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
180
  },
181
  "upload_measurements": [
182
  {
183
- "seconds": 1.0140557359991362,
184
- "bytes": 5030935,
185
- "pages": 25,
186
- "shards": 3
187
- },
188
- {
189
- "seconds": 1.2526760300024762,
190
- "bytes": 11691054,
191
- "pages": 30,
192
- "shards": 3
193
- },
194
- {
195
- "seconds": 0.9398190409992822,
196
- "bytes": 2819009,
197
- "pages": 37,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.1499773260002257,
202
- "bytes": 6017503,
203
- "pages": 37,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 1.6606815510021988,
208
- "bytes": 10233181,
209
- "pages": 23,
210
- "shards": 3
211
- },
212
- {
213
- "seconds": 1.0018192189963884,
214
- "bytes": 366522,
215
- "pages": 10,
216
  "shards": 1
217
  },
218
  {
219
- "seconds": 0.8059965610009385,
220
- "bytes": 2452391,
221
- "pages": 29,
222
- "shards": 2
223
- },
224
- {
225
- "seconds": 0.9446101400026237,
226
- "bytes": 6258704,
227
- "pages": 37,
228
- "shards": 2
229
- },
230
- {
231
- "seconds": 1.1053826549978112,
232
- "bytes": 8176480,
233
- "pages": 32,
234
- "shards": 3
235
  },
236
  {
237
- "seconds": 1.249701370994444,
238
- "bytes": 10043413,
239
- "pages": 32,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 1.0213671140008955,
244
- "bytes": 6189327,
245
- "pages": 36,
246
- "shards": 2
247
- },
248
- {
249
- "seconds": 1.0084248750063125,
250
- "bytes": 2332449,
251
- "pages": 29,
252
- "shards": 2
253
- },
254
- {
255
- "seconds": 0.9006169920030516,
256
- "bytes": 6118351,
257
- "pages": 36,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.270539192002616,
262
- "bytes": 5502511,
263
- "pages": 29,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 1.512666299000557,
268
- "bytes": 13201899,
269
- "pages": 33,
270
- "shards": 4
271
- },
272
- {
273
- "seconds": 1.000070736001362,
274
- "bytes": 6089030,
275
- "pages": 37,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 1.0753602199984016,
280
- "bytes": 871659,
281
  "pages": 26,
282
  "shards": 1
283
  },
284
  {
285
- "seconds": 1.484286307997536,
286
- "bytes": 11527011,
287
- "pages": 36,
288
- "shards": 4
289
- },
290
- {
291
- "seconds": 0.9214101749967085,
292
- "bytes": 1181839,
293
- "pages": 33,
294
- "shards": 1
295
- },
296
- {
297
- "seconds": 1.1708070859967847,
298
- "bytes": 7634853,
299
- "pages": 39,
300
- "shards": 3
301
- },
302
- {
303
- "seconds": 1.219036098998913,
304
- "bytes": 6506587,
305
- "pages": 23,
306
- "shards": 2
307
- },
308
- {
309
- "seconds": 1.234236194002733,
310
- "bytes": 11430289,
311
- "pages": 42,
312
  "shards": 3
313
  },
314
  {
315
- "seconds": 1.0073759259976214,
316
- "bytes": 7448462,
317
- "pages": 30,
318
  "shards": 3
319
  },
320
  {
321
- "seconds": 0.7627367220047745,
322
- "bytes": 1172696,
323
- "pages": 36,
324
  "shards": 1
325
  },
326
  {
327
- "seconds": 1.1723295699994196,
328
- "bytes": 5714080,
329
- "pages": 28,
330
- "shards": 2
331
- },
332
- {
333
- "seconds": 1.1845541919974494,
334
- "bytes": 7669056,
335
- "pages": 38,
336
  "shards": 3
337
  },
338
  {
339
- "seconds": 0.9817348009964917,
340
- "bytes": 4526983,
341
- "pages": 22,
342
- "shards": 2
343
- },
344
- {
345
- "seconds": 1.2601451889931923,
346
- "bytes": 13743089,
347
- "pages": 49,
348
  "shards": 4
349
  },
350
  {
351
- "seconds": 0.8212157870002557,
352
- "bytes": 890133,
353
- "pages": 26,
354
  "shards": 1
355
  },
356
  {
357
- "seconds": 1.1462097159965197,
358
- "bytes": 10417894,
359
- "pages": 36,
360
- "shards": 3
361
- },
362
- {
363
- "seconds": 1.1751894160042866,
364
- "bytes": 7506454,
365
- "pages": 38,
366
  "shards": 3
367
  },
368
  {
369
- "seconds": 1.0263809830066748,
370
- "bytes": 924946,
371
- "pages": 27,
372
- "shards": 1
373
- },
374
- {
375
- "seconds": 1.1823451020027278,
376
- "bytes": 13453015,
377
- "pages": 42,
378
- "shards": 4
379
- },
380
- {
381
- "seconds": 1.2355715770027018,
382
- "bytes": 9750066,
383
- "pages": 35,
384
- "shards": 4
385
- },
386
- {
387
- "seconds": 1.382024664999335,
388
- "bytes": 840789,
389
- "pages": 26,
390
- "shards": 1
391
- },
392
- {
393
- "seconds": 1.7087262210043264,
394
- "bytes": 7732105,
395
- "pages": 39,
396
  "shards": 3
397
  },
398
  {
399
- "seconds": 1.1756209270024556,
400
- "bytes": 879254,
401
- "pages": 27,
402
  "shards": 1
403
  },
404
  {
405
- "seconds": 1.4754508929981967,
406
- "bytes": 10625968,
407
- "pages": 36,
408
- "shards": 3
409
- },
410
- {
411
- "seconds": 0.9398709419983788,
412
- "bytes": 2450871,
413
- "pages": 30,
414
- "shards": 2
415
- },
416
- {
417
- "seconds": 1.1338124570029322,
418
- "bytes": 11954791,
419
- "pages": 38,
420
- "shards": 3
421
- },
422
- {
423
- "seconds": 1.1746257199993124,
424
- "bytes": 7528355,
425
- "pages": 38,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 1.2720379320016946,
430
- "bytes": 4790673,
431
- "pages": 27,
432
- "shards": 2
433
- },
434
- {
435
- "seconds": 0.9963719520019367,
436
- "bytes": 6449131,
437
- "pages": 37,
438
- "shards": 2
439
- },
440
- {
441
- "seconds": 0.9008199800009606,
442
- "bytes": 2360873,
443
- "pages": 29,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 1.193771635000303,
448
- "bytes": 12283347,
449
- "pages": 43,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 0.9776089399965713,
454
- "bytes": 9862558,
455
- "pages": 28,
456
  "shards": 3
457
  },
458
  {
459
- "seconds": 0.8356411559943808,
460
- "bytes": 2333118,
461
- "pages": 29,
462
- "shards": 2
463
- },
464
- {
465
- "seconds": 1.1188001900009112,
466
- "bytes": 6081727,
467
- "pages": 28,
468
- "shards": 2
469
- },
470
- {
471
- "seconds": 0.9660166879984899,
472
- "bytes": 2720285,
473
- "pages": 38,
474
- "shards": 2
475
  },
476
  {
477
- "seconds": 1.24898188199586,
478
- "bytes": 10600974,
479
  "pages": 37,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 1.208974320004927,
484
- "bytes": 5767450,
485
- "pages": 28,
486
- "shards": 2
487
- },
488
- {
489
- "seconds": 1.1749305560006178,
490
- "bytes": 8197083,
491
- "pages": 33,
492
- "shards": 3
493
- },
494
- {
495
- "seconds": 0.9447351959970547,
496
- "bytes": 6335429,
497
- "pages": 37,
498
- "shards": 2
499
- },
500
- {
501
- "seconds": 1.0603171299953829,
502
- "bytes": 4753623,
503
- "pages": 29,
504
  "shards": 2
505
  },
506
  {
507
- "seconds": 1.2698575760005042,
508
- "bytes": 7643809,
509
- "pages": 39,
510
  "shards": 3
511
  },
512
  {
513
- "seconds": 1.0007858850003686,
514
- "bytes": 5847294,
515
  "pages": 29,
516
- "shards": 2
517
- },
518
- {
519
- "seconds": 0.8758766960017965,
520
- "bytes": 2703346,
521
- "pages": 37,
522
- "shards": 2
523
- },
524
- {
525
- "seconds": 1.3512026129974402,
526
- "bytes": 16969598,
527
- "pages": 37,
528
- "shards": 4
529
- },
530
- {
531
- "seconds": 1.1678855200007092,
532
- "bytes": 919043,
533
- "pages": 27,
534
- "shards": 1
535
- },
536
- {
537
- "seconds": 1.127943243001937,
538
- "bytes": 7825410,
539
- "pages": 38,
540
  "shards": 3
541
  },
542
  {
543
- "seconds": 1.1976684740002383,
544
- "bytes": 5884805,
545
- "pages": 31,
546
- "shards": 2
547
- },
548
- {
549
- "seconds": 0.8530800510052359,
550
- "bytes": 740338,
551
- "pages": 24,
552
  "shards": 1
553
  },
554
  {
555
- "seconds": 1.4493056469946168,
556
- "bytes": 12401988,
557
- "pages": 45,
558
  "shards": 4
559
- },
560
- {
561
- "seconds": 0.9416848679975374,
562
- "bytes": 6631553,
563
- "pages": 29,
564
- "shards": 2
565
  }
566
  ],
567
  "cache_key": {
@@ -602,11 +362,7 @@
602
  "source": "f8722b66ff2f92bca174f36564336149e85c65c48b964d05ff3a4d5d85da308f",
603
  "cpu": "unknown",
604
  "cpu_count": 20,
605
- "microbatch": 86
606
  },
607
- "cache_reused": true,
608
- "resume_probe_seconds": [
609
- 3.301654667999969,
610
- 0.11550889399995867
611
- ]
612
  }
 
1
  {
2
  "version": 1,
3
  "startup_batch_seconds": [
4
+ 2.3260779750000324,
5
+ 0.017838069999982054,
6
+ 0.0158013189999906,
7
+ 0.015721255000016754,
8
+ 0.09197544800002788,
9
+ 0.014327330000014626,
10
+ 0.013824727999974584,
11
+ 0.10075297500003444
12
  ],
13
  "source_benchmarks": [
14
  {
15
  "source": "climbmix",
16
  "rows": 256,
17
+ "tokens": 903410,
18
+ "bytes": 372598,
19
+ "generation_seconds": 1.5044533369999726,
20
+ "write_seconds": 0.005470204000005197,
21
+ "read_seconds": 0.004841493000014907,
22
+ "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 76764.09997329117,
26
+ "required_pages_per_second_upper_estimate": 0.00531071855340399
27
  },
28
  {
29
  "source": "rewrite6",
30
  "rows": 256,
31
+ "tokens": 149279,
32
+ "bytes": 66870,
33
+ "generation_seconds": 17.489590872999997,
34
+ "write_seconds": 0.0027630569999814725,
35
+ "read_seconds": 0.003109784999992371,
36
+ "tokens_per_second": 8533.957213384612,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
+ "required_token_positions_per_second": 9595.512496661397,
40
+ "required_pages_per_second_upper_estimate": 0.004017440705265558
41
  },
42
  {
43
  "source": "openmath",
44
  "rows": 256,
45
+ "tokens": 300086,
46
+ "bytes": 97823,
47
+ "generation_seconds": 0.7871533299999669,
48
+ "write_seconds": 0.0022467860000006112,
49
+ "read_seconds": 0.0028803160000165917,
50
+ "tokens_per_second": 380144.35761751566,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
+ "required_token_positions_per_second": 19191.024993322793,
54
+ "required_pages_per_second_upper_estimate": 0.003996984404746221
55
  },
56
  {
57
  "source": "ultra_qa",
58
  "rows": 256,
59
+ "tokens": 1042562,
60
+ "bytes": 369139,
61
+ "generation_seconds": 23.827993708999998,
62
+ "write_seconds": 0.0062629279999555365,
63
+ "read_seconds": 0.006445001999964006,
64
+ "tokens_per_second": 43742.16556775435,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
+ "required_token_positions_per_second": 28786.537489984188,
68
+ "required_pages_per_second_upper_estimate": 0.0017257089680268527
69
  },
70
  {
71
  "source": "ultra_style",
72
  "rows": 256,
73
+ "tokens": 803904,
74
+ "bytes": 337820,
75
+ "generation_seconds": 36.50458967899999,
76
+ "write_seconds": 0.00519779599994763,
77
+ "read_seconds": 0.005391688000031536,
78
+ "tokens_per_second": 22018.86276523721,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
+ "required_token_positions_per_second": 28786.537489984188,
82
+ "required_pages_per_second_upper_estimate": 0.002238026671249318
83
  },
84
  {
85
  "source": "cortex",
86
  "rows": 128,
87
+ "tokens": 77961,
88
+ "bytes": 43256,
89
+ "generation_seconds": 0.08065045199998622,
90
+ "write_seconds": 0.004991313999994418,
91
+ "read_seconds": 0.04061180899998362,
92
+ "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 23,
95
+ "required_token_positions_per_second": 28786.537489984188,
96
+ "required_pages_per_second_upper_estimate": 0.3692427943456881
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
108
  "local_shard_size_benchmark": [
109
  {
110
  "target_mib": 16,
111
+ "actual_bytes": 15554560,
112
+ "pages": 7,
113
+ "packing_seconds": 0.03195600199990167,
114
+ "read_seconds": 0.0013301719999390116,
115
+ "io_bytes_per_second": 467297923.7588089,
116
  "target_reached": true
117
  },
118
  {
119
  "target_mib": 32,
120
+ "actual_bytes": 18083840,
121
+ "pages": 9,
122
+ "packing_seconds": 0.0402841879999869,
123
+ "read_seconds": 0.0016377959999545055,
124
+ "io_bytes_per_second": 431368897.0451703,
125
  "target_reached": false
126
  },
127
  {
128
  "target_mib": 64,
129
+ "actual_bytes": 18083840,
130
+ "pages": 9,
131
+ "packing_seconds": 0.0399535819999528,
132
+ "read_seconds": 0.0013251859999172666,
133
+ "io_bytes_per_second": 438090594.17802685,
134
  "target_reached": false
135
  },
136
  {
137
  "target_mib": 128,
138
+ "actual_bytes": 18083840,
139
+ "pages": 9,
140
+ "packing_seconds": 0.04405787199993938,
141
+ "read_seconds": 0.0014645830000290516,
142
+ "io_bytes_per_second": 397250983.05907583,
143
  "target_reached": false
144
  },
145
  {
146
  "target_mib": 256,
147
+ "actual_bytes": 18083840,
148
+ "pages": 9,
149
+ "packing_seconds": 0.038781767000045875,
150
+ "read_seconds": 0.0013218610000649278,
151
+ "io_bytes_per_second": 450927781.39548963,
152
  "target_reached": false
153
  }
154
  ],
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
+ "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
+ "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
+ "batch_memory_estimate_bytes": 704566,
161
+ "producer_batch_seconds": 0.044507669203127875,
162
+ "producer_p95_seconds": 0.14738937200036162,
163
+ "gpu_update_seconds": 0.9604489601995264,
164
+ "producer_batches_per_second": 22.468037933779797,
165
+ "gpu_batches_per_second": 1.0411797413912105,
166
+ "producer_headroom": 21.579403671222323,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.50828449257016,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 83,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 289702.9369927744,
174
+ "upload_bytes_per_second": 5452255.476764406,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 62.16350633943009,
179
  "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
180
  },
181
  "upload_measurements": [
182
  {
183
+ "seconds": 0.9708695179999722,
184
+ "bytes": 43075,
185
+ "pages": 1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 0.8249086619999844,
190
+ "bytes": 105910,
191
+ "pages": 3,
192
+ "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
193
  },
194
  {
195
+ "seconds": 1.08548185799998,
196
+ "bytes": 2013983,
197
+ "pages": 18,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 1.0607301400000324,
202
+ "bytes": 873319,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
203
  "pages": 26,
204
  "shards": 1
205
  },
206
  {
207
+ "seconds": 1.4327509930000133,
208
+ "bytes": 9634876,
209
+ "pages": 38,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.0578138549999494,
214
+ "bytes": 7567895,
215
+ "pages": 37,
216
  "shards": 3
217
  },
218
  {
219
+ "seconds": 0.8248641469999711,
220
+ "bytes": 1159359,
221
+ "pages": 34,
222
  "shards": 1
223
  },
224
  {
225
+ "seconds": 0.9606663819999994,
226
+ "bytes": 10228939,
227
+ "pages": 23,
 
 
 
 
 
 
228
  "shards": 3
229
  },
230
  {
231
+ "seconds": 1.4356544900001609,
232
+ "bytes": 13537180,
233
+ "pages": 46,
 
 
 
 
 
 
234
  "shards": 4
235
  },
236
  {
237
+ "seconds": 0.9362600690001273,
238
+ "bytes": 1025985,
239
+ "pages": 34,
240
  "shards": 1
241
  },
242
  {
243
+ "seconds": 1.7697618709999006,
244
+ "bytes": 7336838,
245
+ "pages": 30,
 
 
 
 
 
 
246
  "shards": 3
247
  },
248
  {
249
+ "seconds": 1.264844782000182,
250
+ "bytes": 9996183,
251
+ "pages": 40,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
252
  "shards": 3
253
  },
254
  {
255
+ "seconds": 0.9787746179999885,
256
+ "bytes": 1132423,
257
+ "pages": 33,
258
  "shards": 1
259
  },
260
  {
261
+ "seconds": 0.8866051049999442,
262
+ "bytes": 7112636,
263
+ "pages": 23,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
264
  "shards": 3
265
  },
266
  {
267
+ "seconds": 1.117540433999693,
268
+ "bytes": 7055686,
269
+ "pages": 40,
 
 
 
 
 
 
 
 
 
 
 
 
270
  "shards": 2
271
  },
272
  {
273
+ "seconds": 1.0181373959999291,
274
+ "bytes": 10411775,
275
+ "pages": 36,
276
  "shards": 3
277
  },
278
  {
279
+ "seconds": 1.183017940999889,
280
+ "bytes": 7545162,
281
+ "pages": 32,
282
  "shards": 3
283
  },
284
  {
285
+ "seconds": 0.9269013600001017,
286
+ "bytes": 1114116,
287
+ "pages": 34,
288
+ "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
289
  },
290
  {
291
+ "seconds": 1.037127201000203,
292
+ "bytes": 7612192,
293
  "pages": 37,
294
  "shards": 3
295
  },
296
  {
297
+ "seconds": 0.9605875500001275,
298
+ "bytes": 5330342,
299
+ "pages": 25,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
300
  "shards": 2
301
  },
302
  {
303
+ "seconds": 1.2280645959999674,
304
+ "bytes": 12018597,
305
+ "pages": 45,
306
  "shards": 3
307
  },
308
  {
309
+ "seconds": 1.1235430140000062,
310
+ "bytes": 7362856,
311
  "pages": 29,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
312
  "shards": 3
313
  },
314
  {
315
+ "seconds": 0.8067839350001123,
316
+ "bytes": 1162221,
317
+ "pages": 35,
 
 
 
 
 
 
318
  "shards": 1
319
  },
320
  {
321
+ "seconds": 1.3463661700002376,
322
+ "bytes": 11675037,
323
+ "pages": 32,
324
  "shards": 4
 
 
 
 
 
 
325
  }
326
  ],
327
  "cache_key": {
 
362
  "source": "f8722b66ff2f92bca174f36564336149e85c65c48b964d05ff3a4d5d85da308f",
363
  "cpu": "unknown",
364
  "cpu_count": 20,
365
+ "microbatch": 90
366
  },
367
+ "cache_reused": false
 
 
 
 
368
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f7cd41c0ad064fdb32d106a04f2baeeeeed5930763f4ca1400aa877b835cb4d4
3
- size 721771839
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0883783165f69503eba3b62bf8c9c7a6e664527b2bcaeee310413d02bf169eb3
3
+ size 721779455
training_config.json CHANGED
@@ -4,7 +4,7 @@
4
  "run_name": "void-byte-v1",
5
  "context": 2048,
6
  "global_batch": 90,
7
- "micro_batch": 86,
8
  "microbatch_candidates": [
9
  8,
10
  16,
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 61,
50
- "prefetch_low_water": 30,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,
@@ -66,7 +66,7 @@
66
  "benchmark_compile": true,
67
  "autotune_microbatch": true,
68
  "attention_backend": "flash",
69
- "vram_fraction": 0.95,
70
  "upload": true,
71
  "hub_resume": true,
72
  "work_dir": "/marimo/qwen90m_workspace/runs/void-byte-v1",
 
4
  "run_name": "void-byte-v1",
5
  "context": 2048,
6
  "global_batch": 90,
7
+ "micro_batch": 90,
8
  "microbatch_candidates": [
9
  8,
10
  16,
 
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
+ "prefetch_batches": 63,
50
+ "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,
 
66
  "benchmark_compile": true,
67
  "autotune_microbatch": true,
68
  "attention_backend": "flash",
69
+ "vram_fraction": 0.98,
70
  "upload": true,
71
  "hub_resume": true,
72
  "work_dir": "/marimo/qwen90m_workspace/runs/void-byte-v1",