appvoid commited on
Commit
b5dbad5
·
verified ·
1 Parent(s): 56533fc

Qwen90M step 605000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,19 +1,19 @@
1
  {
2
- "step": 604474,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "844e788f1be5aab74d28e9ce7619f1b4e439c0fbb67830af44ada2cbb259588a",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "3d9f8c9ad474cdc8cb2c7641df33aedf49a7e6f0d7993cba7ca705b1e93d5865",
12
- "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6",
13
- "performance.json": "f50e1ba0b63842d885a4889b44af72c5a22921c6edb887fc4bd12e1c74dd5b96",
14
- "shard_scheduler.json": "4838b8feaac6a7934582c03c48aed0e704413c464260d4abc236fa450ab19c11",
15
- "coverage.json": "ad1b760456d2de7f27d4bf65fbe0d8199e32c419a7ad94f3dc662b7548924d2f",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
- "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
19
  }
 
1
  {
2
+ "step": 605000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "ae98fb2eb852ed9b2ef3d507885819a6d2ff073c8d6b5c3f2962d07318d92067",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "0d3af9a8da6b443eadd810e3c1064dd4eef75f364569d9e4096a54e70b5227d9",
12
+ "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb",
13
+ "performance.json": "539995c1e4203016bfc6259363445c7c25dfb51604d74cd18d2cc7ad51bfb71e",
14
+ "shard_scheduler.json": "73644496d4fc18122195c127b0dde4b315912feb170d8757f56b71ce52b9030d",
15
+ "coverage.json": "2533a345a29ab4b098939b907ba6cdee33e969c8784659c7ca62523942ef7752",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
+ "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"
19
  }
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 604474,
3
  "stats": {
4
- "climbmix/blocks": 21761064,
5
- "climbmix/targets": 36376593489,
6
- "climbmix/input_tokens": 36378899231,
7
- "ultra_qa/blocks": 8160399,
8
- "ultra_qa/targets": 13312971450,
9
- "ultra_qa/input_tokens": 13312997138,
10
- "cortex/blocks": 8160399,
11
- "cortex/targets": 310295416,
12
- "cortex/input_tokens": 12971523956,
13
- "ultra_style/blocks": 8160399,
14
- "ultra_style/targets": 12907233911,
15
- "ultra_style/input_tokens": 12907911297,
16
- "openmath/blocks": 5440266,
17
- "openmath/targets": 8406347806,
18
- "openmath/input_tokens": 8409147882,
19
- "rewrite6/blocks": 2720133,
20
- "rewrite6/targets": 4459340280,
21
- "rewrite6/input_tokens": 4464229442
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 2980,
26
- "row": 2621,
27
  "start": {
28
  "file": 144,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 18,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 12208701
38
  },
39
  "rewrite6": {
40
- "page": 45,
41
- "row": 2626,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 184320
47
  }
48
  },
49
  "epoch": 33,
50
  "offset": 0,
51
- "documents": 6965278
52
  },
53
  "openmath": {
54
- "page": 1862,
55
- "row": 1357,
56
  "start": {
57
  "file": 17,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 203,
61
- "row": 702
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 7628109
67
  },
68
  "ultra_qa": {
69
- "page": 805,
70
- "row": 1293,
71
  "start": {
72
  "file": 6,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 2,
76
- "row": 50531
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 3298573
82
  },
83
  "ultra_style": {
84
- "page": 1229,
85
- "row": 1230,
86
  "start": {
87
  "file": 7,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 2,
91
- "row": 68406
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 5035214
97
  },
98
  "cortex": {
99
- "page": 158263,
100
- "row": 99,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 20257763
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 605000,
3
  "stats": {
4
+ "climbmix/blocks": 21780000,
5
+ "climbmix/targets": 36408254077,
6
+ "climbmix/input_tokens": 36410561797,
7
+ "ultra_qa/blocks": 8167500,
8
+ "ultra_qa/targets": 13324585723,
9
+ "ultra_qa/input_tokens": 13324611432,
10
+ "cortex/blocks": 8167500,
11
+ "cortex/targets": 310565399,
12
+ "cortex/input_tokens": 12982796500,
13
+ "ultra_style/blocks": 8167500,
14
+ "ultra_style/targets": 12918082747,
15
+ "ultra_style/input_tokens": 12918760755,
16
+ "openmath/blocks": 5445000,
17
+ "openmath/targets": 8413669599,
18
+ "openmath/input_tokens": 8416472178,
19
+ "rewrite6/blocks": 2722500,
20
+ "rewrite6/targets": 4463218274,
21
+ "rewrite6/input_tokens": 4468111496
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 2983,
26
+ "row": 912,
27
  "start": {
28
  "file": 144,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 30,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 6144,
37
+ "documents": 12219280
38
  },
39
  "rewrite6": {
40
+ "page": 47,
41
+ "row": 282,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 192512
47
  }
48
  },
49
  "epoch": 33,
50
  "offset": 0,
51
+ "documents": 6971126
52
  },
53
  "openmath": {
54
+ "page": 1863,
55
+ "row": 3962,
56
  "start": {
57
  "file": 17,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 207,
61
+ "row": 798
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 7634810
67
  },
68
  "ultra_qa": {
69
+ "page": 806,
70
+ "row": 91,
71
  "start": {
72
  "file": 6,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 2,
76
+ "row": 54627
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 6144,
81
+ "documents": 3301467
82
  },
83
  "ultra_style": {
84
+ "page": 1230,
85
+ "row": 1900,
86
  "start": {
87
  "file": 7,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 2,
91
+ "row": 72502
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 5039980
97
  },
98
  "cortex": {
99
+ "page": 158401,
100
+ "row": 33,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 20275361
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 604474, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "844e788f1be5aab74d28e9ce7619f1b4e439c0fbb67830af44ada2cbb259588a", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "3d9f8c9ad474cdc8cb2c7641df33aedf49a7e6f0d7993cba7ca705b1e93d5865", "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6", "performance.json": "f50e1ba0b63842d885a4889b44af72c5a22921c6edb887fc4bd12e1c74dd5b96", "shard_scheduler.json": "4838b8feaac6a7934582c03c48aed0e704413c464260d4abc236fa450ab19c11", "coverage.json": "ad1b760456d2de7f27d4bf65fbe0d8199e32c419a7ad94f3dc662b7548924d2f", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000604474"}
 
1
+ {"step": 605000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "ae98fb2eb852ed9b2ef3d507885819a6d2ff073c8d6b5c3f2962d07318d92067", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "0d3af9a8da6b443eadd810e3c1064dd4eef75f364569d9e4096a54e70b5227d9", "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb", "performance.json": "539995c1e4203016bfc6259363445c7c25dfb51604d74cd18d2cc7ad51bfb71e", "shard_scheduler.json": "73644496d4fc18122195c127b0dde4b315912feb170d8757f56b71ce52b9030d", "coverage.json": "2533a345a29ab4b098939b907ba6cdee33e969c8784659c7ca62523942ef7752", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000605000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:844e788f1be5aab74d28e9ce7619f1b4e439c0fbb67830af44ada2cbb259588a
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ae98fb2eb852ed9b2ef3d507885819a6d2ff073c8d6b5c3f2962d07318d92067
3
  size 360622744
performance.json CHANGED
@@ -41,146 +41,146 @@
41
  "attention_backend": "flash",
42
  "vram_fraction": 0.98
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": 90,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9418963939999685,
53
- "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9418963939999685,
56
- "input_tokens_per_second": 195690.31283498698
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": [
@@ -218,6 +218,5 @@
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
- "legacy_seed_verified": false
223
  }
 
41
  "attention_backend": "flash",
42
  "vram_fraction": 0.98
43
  },
44
+ "source": "d52752a64023cf29dc386d43835cf3173c5318ee093c3b375a812733c74023a1",
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": 0.9299377820000245,
53
+ "peak_gb": 89.3772840499878,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9299377820000245,
56
+ "input_tokens_per_second": 198206.80863571487
57
  },
58
  "measurements": [
59
  {
60
  "microbatch": 8,
61
  "mode": "eager",
62
+ "microbatch_seconds": 0.15785124100000303,
63
  "peak_gb": 14.764304637908936,
64
  "fits_budget": true,
65
+ "estimated_update_seconds": 1.7943123250000212,
66
+ "input_tokens_per_second": 103793.92582665654
67
  },
68
  {
69
  "microbatch": 16,
70
  "mode": "eager",
71
+ "microbatch_seconds": 0.3454984619999948,
72
  "peak_gb": 28.084960460662842,
73
  "fits_budget": true,
74
+ "estimated_update_seconds": 1.92592258199997,
75
+ "input_tokens_per_second": 94842.67979172798
76
  },
77
  {
78
  "microbatch": 32,
79
  "mode": "eager",
80
+ "microbatch_seconds": 0.7818738260000089,
81
  "peak_gb": 54.827155113220215,
82
  "fits_budget": true,
83
+ "estimated_update_seconds": 2.1863095480000254,
84
+ "input_tokens_per_second": 83819.15063620413
85
  },
86
  {
87
  "microbatch": 48,
88
  "mode": "eager",
89
+ "microbatch_seconds": 1.171907979999986,
90
  "peak_gb": 81.63183498382568,
91
  "fits_budget": true,
92
+ "estimated_update_seconds": 2.1970569614999818,
93
+ "input_tokens_per_second": 83883.71926608194
94
  },
95
  {
96
  "microbatch": 8,
97
  "mode": "compiled",
98
+ "microbatch_seconds": 0.0880895730000475,
99
  "peak_gb": 9.113307476043701,
100
  "fits_budget": true,
101
+ "estimated_update_seconds": 1.0067526180005189,
102
+ "input_tokens_per_second": 185992.50106469658
103
  },
104
  {
105
  "microbatch": 16,
106
  "mode": "compiled",
107
+ "microbatch_seconds": 0.1701795650000122,
108
  "peak_gb": 16.87287473678589,
109
  "fits_budget": true,
110
+ "estimated_update_seconds": 0.9559394930000451,
111
+ "input_tokens_per_second": 192549.55787434086
112
  },
113
  {
114
  "microbatch": 32,
115
  "mode": "compiled",
116
+ "microbatch_seconds": 0.34178767000003063,
117
  "peak_gb": 32.49641752243042,
118
  "fits_budget": true,
119
+ "estimated_update_seconds": 0.950241911000063,
120
+ "input_tokens_per_second": 191744.77534544803
121
  },
122
  {
123
  "microbatch": 48,
124
  "mode": "compiled",
125
+ "microbatch_seconds": 0.5061955770000282,
126
  "peak_gb": 48.123157024383545,
127
  "fits_budget": true,
128
+ "estimated_update_seconds": 0.9457143725000208,
129
+ "input_tokens_per_second": 194201.61784620755
130
  },
131
  {
132
  "microbatch": 64,
133
  "mode": "compiled",
134
+ "microbatch_seconds": 0.6595429280000076,
135
  "peak_gb": 63.757601737976074,
136
  "fits_budget": true,
137
+ "estimated_update_seconds": 0.9208904890000156,
138
+ "input_tokens_per_second": 198731.56762890573
139
  },
140
  {
141
  "microbatch": 72,
142
  "mode": "compiled",
143
+ "microbatch_seconds": 0.741346276999991,
144
  "peak_gb": 71.65071392059326,
145
  "fits_budget": true,
146
+ "estimated_update_seconds": 0.9243867199999727,
147
+ "input_tokens_per_second": 198903.00197730918
148
  },
149
  {
150
  "microbatch": 76,
151
  "mode": "compiled",
152
+ "microbatch_seconds": 0.7816785220000497,
153
  "peak_gb": 75.5242338180542,
154
  "fits_budget": true,
155
+ "estimated_update_seconds": 0.9251490005000278,
156
+ "input_tokens_per_second": 199120.22093398403
157
  },
158
  {
159
  "microbatch": 80,
160
  "mode": "compiled",
161
+ "microbatch_seconds": 0.8258502979999776,
162
  "peak_gb": 79.40062236785889,
163
  "fits_budget": true,
164
+ "estimated_update_seconds": 0.9283437199999867,
165
+ "input_tokens_per_second": 198389.46646478592
166
  },
167
  {
168
  "microbatch": 86,
169
  "mode": "compiled",
170
+ "microbatch_seconds": 0.8883623469999975,
171
  "peak_gb": 85.4687147140503,
172
  "fits_budget": true,
173
+ "estimated_update_seconds": 0.9419139469999607,
174
+ "input_tokens_per_second": 198261.44207347915
175
  },
176
  {
177
  "microbatch": 90,
178
  "mode": "compiled",
179
+ "microbatch_seconds": 0.9299377820000245,
180
  "peak_gb": 89.3772840499878,
181
  "fits_budget": true,
182
+ "estimated_update_seconds": 0.9299377820000245,
183
+ "input_tokens_per_second": 198206.80863571487
184
  }
185
  ],
186
  "failures": [
 
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
@@ -22,8 +22,8 @@
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": 76509.59968554687,
26
- "required_pages_per_second_upper_estimate": 0.005293111632975813
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
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": 9563.699960693359,
40
- "required_pages_per_second_upper_estimate": 0.004004121460777034
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
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": 19127.399921386717,
54
- "required_pages_per_second_upper_estimate": 0.0039837329801679176
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
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": 28691.099882080078,
68
- "required_pages_per_second_upper_estimate": 0.001719987629157791
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
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": 28691.099882080078,
82
- "required_pages_per_second_upper_estimate": 0.002230606817020446
83
  },
84
  {
85
  "source": "cortex",
@@ -92,8 +92,8 @@
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": 28691.099882080078,
96
- "required_pages_per_second_upper_estimate": 0.3680186231844137
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,182 +158,62 @@
158
  "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.09504132128127196,
162
- "producer_p95_seconds": 0.16387522999957582,
163
- "gpu_update_seconds": 0.9636437819962567,
164
- "producer_batches_per_second": 10.52173924477049,
165
- "gpu_batches_per_second": 1.0377278603182898,
166
- "producer_headroom": 10.139208599009073,
167
  "sustainable": true,
168
- "buffer_seconds": 60.70955826576417,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 79,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 279098.6073348513,
174
- "upload_bytes_per_second": 5870910.389131389,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 59.88806405100016,
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": 1.296470003999957,
184
- "bytes": 22520,
185
  "pages": 1,
186
  "shards": 1
187
  },
188
  {
189
- "seconds": 1.5001352909998786,
190
- "bytes": 11915224,
191
- "pages": 36,
192
- "shards": 3
193
- },
194
- {
195
- "seconds": 1.1385679780000828,
196
- "bytes": 2605060,
197
- "pages": 36,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.2088590650000697,
202
- "bytes": 9710127,
203
- "pages": 36,
204
- "shards": 3
205
- },
206
- {
207
- "seconds": 0.972143672999664,
208
- "bytes": 6102938,
209
- "pages": 35,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 0.7959529229997315,
214
- "bytes": 2320780,
215
- "pages": 28,
216
- "shards": 2
217
- },
218
- {
219
- "seconds": 1.0472984960001668,
220
- "bytes": 9808086,
221
- "pages": 37,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 1.0871454139996786,
226
- "bytes": 13468906,
227
- "pages": 38,
228
- "shards": 4
229
- },
230
- {
231
- "seconds": 0.8225870739997845,
232
- "bytes": 1093468,
233
- "pages": 34,
234
  "shards": 1
235
  },
236
  {
237
- "seconds": 1.3778674129998763,
238
- "bytes": 6041724,
239
- "pages": 32,
240
- "shards": 2
241
- },
242
- {
243
- "seconds": 1.0736160129999917,
244
- "bytes": 6799126,
245
- "pages": 32,
246
- "shards": 3
247
- },
248
- {
249
- "seconds": 1.1032466139999997,
250
- "bytes": 6317020,
251
- "pages": 35,
252
- "shards": 2
253
- },
254
- {
255
- "seconds": 1.0759855689998403,
256
- "bytes": 6102397,
257
- "pages": 35,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.0165042769999673,
262
- "bytes": 8195972,
263
- "pages": 32,
264
- "shards": 3
265
- },
266
- {
267
- "seconds": 0.9696753220000573,
268
- "bytes": 10144942,
269
- "pages": 38,
270
- "shards": 3
271
- },
272
- {
273
- "seconds": 1.035786852000001,
274
- "bytes": 2660447,
275
- "pages": 37,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 1.2393170549999013,
280
- "bytes": 5990595,
281
- "pages": 27,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 0.8981117179996545,
286
- "bytes": 970766,
287
  "pages": 28,
288
- "shards": 1
289
- },
290
- {
291
- "seconds": 1.1719487760001357,
292
- "bytes": 12751643,
293
- "pages": 48,
294
- "shards": 5
295
  },
296
  {
297
- "seconds": 1.1927614570004152,
298
- "bytes": 11797840,
299
  "pages": 30,
300
  "shards": 3
301
  },
302
  {
303
- "seconds": 1.1542902260007395,
304
- "bytes": 2688701,
305
- "pages": 36,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 0.9894740659992749,
310
- "bytes": 5828242,
311
  "pages": 28,
312
  "shards": 2
313
- },
314
- {
315
- "seconds": 1.3037635500004399,
316
- "bytes": 10399515,
317
- "pages": 37,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 1.1070935670004474,
322
- "bytes": 2763911,
323
- "pages": 37,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 1.3125206449994948,
328
- "bytes": 6101597,
329
- "pages": 27,
330
- "shards": 2
331
- },
332
- {
333
- "seconds": 0.9716803040000741,
334
- "bytes": 6849385,
335
- "pages": 34,
336
- "shards": 2
337
  }
338
  ],
339
  "cache_key": {
@@ -378,7 +258,7 @@
378
  },
379
  "cache_reused": true,
380
  "resume_probe_seconds": [
381
- 3.1246708119999766,
382
- 0.0175833739999689
383
  ]
384
  }
 
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": 77247.77051665467,
26
+ "required_pages_per_second_upper_estimate": 0.005344180003864155
27
  },
28
  {
29
  "source": "rewrite6",
 
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": 9655.971314581833,
40
+ "required_pages_per_second_upper_estimate": 0.00404275354980516
41
  },
42
  {
43
  "source": "openmath",
 
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": 19311.942629163666,
54
+ "required_pages_per_second_upper_estimate": 0.00402216835947938
55
  },
56
  {
57
  "source": "ultra_qa",
 
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": 28967.9139437455,
68
+ "required_pages_per_second_upper_estimate": 0.0017365822094840345
69
  },
70
  {
71
  "source": "ultra_style",
 
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": 28967.9139437455,
82
+ "required_pages_per_second_upper_estimate": 0.0022521278927385533
83
  },
84
  {
85
  "source": "cortex",
 
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": 28967.9139437455,
96
+ "required_pages_per_second_upper_estimate": 0.37156929674767514
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
158
  "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.06336448393750427,
162
+ "producer_p95_seconds": 0.12817057600000226,
163
+ "gpu_update_seconds": 0.9544353125907264,
164
+ "producer_batches_per_second": 15.781711423489057,
165
+ "gpu_batches_per_second": 1.0477399429884802,
166
+ "producer_headroom": 15.062622675694417,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.12942469321577,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 55,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 374353.52534401446,
174
+ "upload_bytes_per_second": 3622099.0301886583,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 80.32755203476262,
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": 1.386034201999962,
184
+ "bytes": 47132,
185
  "pages": 1,
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 1.0035069749999934,
190
+ "bytes": 155991,
191
+ "pages": 4,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192
  "shards": 1
193
  },
194
  {
195
+ "seconds": 0.9961135309999918,
196
+ "bytes": 5836820,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
197
  "pages": 28,
198
+ "shards": 2
 
 
 
 
 
 
199
  },
200
  {
201
+ "seconds": 1.1866138010000213,
202
+ "bytes": 7420176,
203
  "pages": 30,
204
  "shards": 3
205
  },
206
  {
207
+ "seconds": 1.0025722640000367,
208
+ "bytes": 6023360,
209
+ "pages": 35,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.4614262610000424,
214
+ "bytes": 6002577,
215
  "pages": 28,
216
  "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
217
  }
218
  ],
219
  "cache_key": {
 
258
  },
259
  "cache_reused": true,
260
  "resume_probe_seconds": [
261
+ 2.591754088000016,
262
+ 0.016849739999997837
263
  ]
264
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3d9f8c9ad474cdc8cb2c7641df33aedf49a7e6f0d7993cba7ca705b1e93d5865
3
- size 721780031
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d3af9a8da6b443eadd810e3c1064dd4eef75f364569d9e4096a54e70b5227d9
3
+ size 721766335
training_config.json CHANGED
@@ -32,11 +32,18 @@
32
  "openmath": 10,
33
  "rewrite6": 5
34
  },
35
- "lr": 0.0003,
 
 
36
  "warmup_steps": 1000,
37
- "max_steps": null,
 
 
 
38
  "min_lr_ratio": 0.1,
39
- "weight_decay": 0.1,
 
 
40
  "grad_clip": 1.0,
41
  "seed": 20260925,
42
  "save_every": 5000,
@@ -71,7 +78,7 @@
71
  "hub_resume": true,
72
  "work_dir": "/marimo/qwen90m_workspace/runs/void-byte-v1",
73
  "data_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
74
- "trainer_recipe": "bc45094ca6fe6129cd1201f38cc5978b80014fa21be8d01ea684cb56d8d1644b",
75
  "archive_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
76
  "cortex_archive_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658"
77
  }
 
32
  "openmath": 10,
33
  "rewrite6": 5
34
  },
35
+ "lr": 0.0001,
36
+ "weight_decay": 0.1,
37
+ "lr_schedule": "auto",
38
  "warmup_steps": 1000,
39
+ "hold_steps": 0,
40
+ "cooldown_steps": 0,
41
+ "schedule_steps": null,
42
+ "schedule_start_step": 0,
43
  "min_lr_ratio": 0.1,
44
+ "max_steps": null,
45
+ "recovery_adjust_lr": false,
46
+ "optimization_file": "optimization.json",
47
  "grad_clip": 1.0,
48
  "seed": 20260925,
49
  "save_every": 5000,
 
78
  "hub_resume": true,
79
  "work_dir": "/marimo/qwen90m_workspace/runs/void-byte-v1",
80
  "data_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
81
+ "trainer_recipe": "cada09ddc22837a37a731a733b4db79f20cd0d25748c84f9e80a63b31e4a73d1",
82
  "archive_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
83
  "cortex_archive_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658"
84
  }