appvoid commited on
Commit
d50fc75
·
verified ·
1 Parent(s): 965b805

Qwen90M step 690000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,19 +1,19 @@
1
  {
2
- "step": 685000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "2f198abaec427b39945f67e2dc4ae368c1e1f3bc749571b34df3aa6e1fabdbbc",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "fef7bf8813882524f5ca30eee990ca62de8fcdd9a4aac08f179ed7a0bfb2f7cb",
12
- "training_config.json": "13fe2f3f4e97181994bd5fcebc103bd2c14b5ea2ce9da3213db6d9bf51f1066f",
13
- "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf",
14
- "shard_scheduler.json": "fa63f6790e325f60a69dde36bb42495987d8d2fe00e2528f2132581755f7dea1",
15
- "coverage.json": "327c66438413b6fcd5c9a5d61ad54b0268663e1e6f30a1e6729f0395214b9ec3",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
- "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"
19
  }
 
1
  {
2
+ "step": 690000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "060f0081aaffd4b21592f326ed75b02afab281b3ffe2de294ff5041c4fd8f70f",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "24ae65d790f92deee8ff8c64623716e72d101b0c319d3796bed4c28c51aec73b",
12
+ "training_config.json": "d8df67626674cd77078628cbede1a700b576ca242c6f22a365ea49ba7ebdeea5",
13
+ "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29",
14
+ "shard_scheduler.json": "dc02b0a99d28c9305b60760d51bc8f62064c31b5e9e3882cfe674a2fb80005a8",
15
+ "coverage.json": "fd32998195f03fdfef6945cb5182b881b8e2c1599651d3d35fa9078908d2cd99",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
+ "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac"
19
  }
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 685000,
3
  "stats": {
4
- "climbmix/blocks": 24660000,
5
- "climbmix/targets": 41224942465,
6
- "climbmix/input_tokens": 41227554983,
7
- "ultra_qa/blocks": 9247500,
8
- "ultra_qa/targets": 15086681728,
9
- "ultra_qa/input_tokens": 15086710887,
10
- "cortex/blocks": 9247500,
11
- "cortex/targets": 351611655,
12
- "cortex/input_tokens": 14699417911,
13
- "ultra_style/blocks": 9247500,
14
- "ultra_style/targets": 14626945782,
15
- "ultra_style/input_tokens": 14627713723,
16
- "openmath/blocks": 6165000,
17
- "openmath/targets": 9526167364,
18
- "openmath/input_tokens": 9529340892,
19
- "rewrite6/blocks": 3082500,
20
- "rewrite6/targets": 5053314593,
21
- "rewrite6/input_tokens": 5058855101
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3377,
26
- "row": 676,
27
  "start": {
28
- "file": 163,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 37,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 4096,
37
- "documents": 13832868
38
  },
39
  "rewrite6": {
40
- "page": 21,
41
- "row": 1829,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 86016
47
  }
48
  },
49
  "epoch": 38,
50
- "offset": 0,
51
- "documents": 7893197
52
  },
53
  "openmath": {
54
- "page": 2110,
55
- "row": 2518,
56
  "start": {
57
  "file": 19,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 346,
61
- "row": 210
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 8645078
67
  },
68
  "ultra_qa": {
69
- "page": 912,
70
- "row": 2704,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
- "row": 33616
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 3738256
82
  },
83
  "ultra_style": {
84
- "page": 1393,
85
- "row": 671,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 2,
91
- "row": 56143
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 5706399
97
  },
98
  "cortex": {
99
- "page": 179346,
100
- "row": 60,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 22956348
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 690000,
3
  "stats": {
4
+ "climbmix/blocks": 24840000,
5
+ "climbmix/targets": 41524942926,
6
+ "climbmix/input_tokens": 41527574676,
7
+ "ultra_qa/blocks": 9315000,
8
+ "ultra_qa/targets": 15196630228,
9
+ "ultra_qa/input_tokens": 15196659602,
10
+ "cortex/blocks": 9315000,
11
+ "cortex/targets": 354172942,
12
+ "cortex/input_tokens": 14806782028,
13
+ "ultra_style/blocks": 9315000,
14
+ "ultra_style/targets": 14733529377,
15
+ "ultra_style/input_tokens": 14734302959,
16
+ "openmath/blocks": 6210000,
17
+ "openmath/targets": 9595770485,
18
+ "openmath/input_tokens": 9598967555,
19
+ "rewrite6/blocks": 3105000,
20
+ "rewrite6/targets": 5090189858,
21
+ "rewrite6/input_tokens": 5095770661
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3402,
26
+ "row": 485,
27
  "start": {
28
+ "file": 164,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 54,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 12288,
37
+ "documents": 13935077
38
  },
39
  "rewrite6": {
40
+ "page": 35,
41
+ "row": 1992,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 143360
47
  }
48
  },
49
  "epoch": 38,
50
+ "offset": 2048,
51
+ "documents": 7950704
52
  },
53
  "openmath": {
54
+ "page": 2126,
55
+ "row": 549,
56
  "start": {
57
  "file": 19,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 411,
61
+ "row": 746
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 8708645
67
  },
68
  "ultra_qa": {
69
+ "page": 919,
70
+ "row": 1367,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
+ "row": 62288
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 3765591
82
  },
83
  "ultra_style": {
84
+ "page": 1403,
85
+ "row": 1688,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 3,
91
+ "row": 11049
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 5748376
97
  },
98
  "cortex": {
99
+ "page": 180655,
100
+ "row": 117,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 23123957
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 685000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "2f198abaec427b39945f67e2dc4ae368c1e1f3bc749571b34df3aa6e1fabdbbc", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "fef7bf8813882524f5ca30eee990ca62de8fcdd9a4aac08f179ed7a0bfb2f7cb", "training_config.json": "13fe2f3f4e97181994bd5fcebc103bd2c14b5ea2ce9da3213db6d9bf51f1066f", "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf", "shard_scheduler.json": "fa63f6790e325f60a69dde36bb42495987d8d2fe00e2528f2132581755f7dea1", "coverage.json": "327c66438413b6fcd5c9a5d61ad54b0268663e1e6f30a1e6729f0395214b9ec3", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000685000"}
 
1
+ {"step": 690000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "060f0081aaffd4b21592f326ed75b02afab281b3ffe2de294ff5041c4fd8f70f", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "24ae65d790f92deee8ff8c64623716e72d101b0c319d3796bed4c28c51aec73b", "training_config.json": "d8df67626674cd77078628cbede1a700b576ca242c6f22a365ea49ba7ebdeea5", "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29", "shard_scheduler.json": "dc02b0a99d28c9305b60760d51bc8f62064c31b5e9e3882cfe674a2fb80005a8", "coverage.json": "fd32998195f03fdfef6945cb5182b881b8e2c1599651d3d35fa9078908d2cd99", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac", "folder": "step-000000690000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2f198abaec427b39945f67e2dc4ae368c1e1f3bc749571b34df3aa6e1fabdbbc
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:060f0081aaffd4b21592f326ed75b02afab281b3ffe2de294ff5041c4fd8f70f
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9403511730000105,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9403511730000105,
56
- "input_tokens_per_second": 196011.87863887308
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9388940140000273,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9388940140000273,
56
+ "input_tokens_per_second": 196316.08813302612
57
  },
58
  "measurements": [
59
  {
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": 75042.20393901474,
26
- "required_pages_per_second_upper_estimate": 0.005191593790403495
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": 9380.275492376843,
40
- "required_pages_per_second_upper_estimate": 0.00392732546623137
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": 18760.550984753685,
54
- "required_pages_per_second_upper_estimate": 0.003907328021124296
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": 28140.826477130526,
68
- "required_pages_per_second_upper_estimate": 0.0016869995787499043
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": 28140.826477130526,
82
- "required_pages_per_second_upper_estimate": 0.0021878254801825318
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": 22,
95
- "required_token_positions_per_second": 28140.826477130526,
96
- "required_pages_per_second_upper_estimate": 0.3609603067832702
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,409 +158,61 @@
158
  "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.3391172111871015,
162
- "producer_p95_seconds": 0.355980565000209,
163
- "gpu_update_seconds": 0.9824871356379303,
164
- "producer_batches_per_second": 2.948832931538438,
165
- "gpu_batches_per_second": 1.0178250317249178,
166
- "producer_headroom": 2.897190420382001,
167
  "sustainable": true,
168
- "buffer_seconds": 60.91420240955168,
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": 385732.1530428105,
174
- "upload_bytes_per_second": 4949797.310566265,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 82.769140658028,
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": 1.6456713849984226,
184
- "bytes": 6071875,
185
- "pages": 38,
186
- "shards": 2
187
- },
188
- {
189
- "seconds": 1.52896361799867,
190
- "bytes": 13388189,
191
- "pages": 35,
192
- "shards": 4
193
- },
194
- {
195
- "seconds": 1.3202310080014286,
196
- "bytes": 5863529,
197
- "pages": 31,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.5651336049995734,
202
- "bytes": 6044120,
203
- "pages": 37,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 1.1207633359990723,
208
- "bytes": 2404167,
209
- "pages": 29,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.4597645020003256,
214
- "bytes": 6256363,
215
- "pages": 37,
216
- "shards": 2
217
- },
218
- {
219
- "seconds": 1.2800013509986456,
220
- "bytes": 667279,
221
- "pages": 18,
222
  "shards": 1
223
  },
224
  {
225
- "seconds": 1.4021912890020758,
226
- "bytes": 11567150,
227
- "pages": 32,
228
- "shards": 4
229
- },
230
- {
231
- "seconds": 1.280043183000089,
232
- "bytes": 12206708,
233
- "pages": 48,
234
- "shards": 3
235
- },
236
- {
237
- "seconds": 0.8664777250014595,
238
- "bytes": 2384518,
239
- "pages": 29,
240
- "shards": 2
241
- },
242
- {
243
- "seconds": 1.027337190000253,
244
- "bytes": 5987324,
245
- "pages": 24,
246
- "shards": 2
247
- },
248
- {
249
- "seconds": 1.2340771259987378,
250
- "bytes": 10655993,
251
- "pages": 48,
252
- "shards": 3
253
- },
254
- {
255
- "seconds": 1.3521600980020594,
256
- "bytes": 2290704,
257
- "pages": 29,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.766581726995355,
262
- "bytes": 12101436,
263
- "pages": 41,
264
- "shards": 3
265
- },
266
- {
267
- "seconds": 1.0664826619977248,
268
- "bytes": 6396203,
269
- "pages": 31,
270
- "shards": 3
271
- },
272
- {
273
- "seconds": 1.070789342003991,
274
- "bytes": 6236244,
275
- "pages": 37,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 1.5840121689980151,
280
- "bytes": 6547762,
281
- "pages": 36,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 1.2987593140060198,
286
- "bytes": 2393188,
287
- "pages": 30,
288
- "shards": 2
289
- },
290
- {
291
- "seconds": 1.808516741002677,
292
- "bytes": 10401411,
293
- "pages": 34,
294
- "shards": 3
295
- },
296
- {
297
- "seconds": 1.1967187990012462,
298
- "bytes": 13416312,
299
- "pages": 42,
300
- "shards": 4
301
- },
302
- {
303
- "seconds": 0.995916217994818,
304
- "bytes": 1003302,
305
- "pages": 27,
306
- "shards": 1
307
- },
308
- {
309
- "seconds": 0.9674175760010257,
310
- "bytes": 6098431,
311
  "pages": 38,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.0489661490064464,
316
- "bytes": 6157279,
317
- "pages": 33,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 1.4413105299972813,
322
- "bytes": 6425996,
323
- "pages": 38,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 1.128546699997969,
328
- "bytes": 5952550,
329
- "pages": 28,
330
  "shards": 3
331
  },
332
  {
333
- "seconds": 1.3368553659965983,
334
- "bytes": 12766871,
335
- "pages": 38,
336
- "shards": 4
337
- },
338
- {
339
- "seconds": 1.6370693080025376,
340
- "bytes": 5715454,
341
- "pages": 28,
342
- "shards": 2
343
- },
344
- {
345
- "seconds": 1.2990395020024152,
346
- "bytes": 2748829,
347
- "pages": 38,
348
- "shards": 2
349
- },
350
- {
351
- "seconds": 1.300822404999053,
352
- "bytes": 6046016,
353
  "pages": 37,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.4766436999998405,
358
- "bytes": 10238885,
359
- "pages": 34,
360
- "shards": 3
361
- },
362
- {
363
- "seconds": 1.4429230239984463,
364
- "bytes": 2385665,
365
- "pages": 28,
366
- "shards": 2
367
- },
368
- {
369
- "seconds": 1.539863748002972,
370
- "bytes": 11993569,
371
- "pages": 43,
372
- "shards": 3
373
- },
374
- {
375
- "seconds": 1.4481774710002355,
376
- "bytes": 7324064,
377
- "pages": 31,
378
- "shards": 3
379
- },
380
- {
381
- "seconds": 1.494102450000355,
382
- "bytes": 5356034,
383
- "pages": 33,
384
- "shards": 2
385
- },
386
- {
387
- "seconds": 1.1517664069979219,
388
- "bytes": 6283260,
389
- "pages": 38,
390
- "shards": 2
391
- },
392
- {
393
- "seconds": 1.5029984030043124,
394
- "bytes": 2302830,
395
- "pages": 28,
396
- "shards": 2
397
- },
398
- {
399
- "seconds": 1.5264721800049301,
400
- "bytes": 6345437,
401
- "pages": 38,
402
- "shards": 2
403
- },
404
- {
405
- "seconds": 1.7716346369998064,
406
- "bytes": 17764299,
407
- "pages": 34,
408
- "shards": 5
409
- },
410
- {
411
- "seconds": 1.4270490490016527,
412
- "bytes": 1152119,
413
- "pages": 35,
414
- "shards": 1
415
- },
416
- {
417
- "seconds": 1.5365510039991932,
418
- "bytes": 6323603,
419
- "pages": 37,
420
- "shards": 2
421
- },
422
- {
423
- "seconds": 1.5189877529992373,
424
- "bytes": 7424560,
425
- "pages": 31,
426
- "shards": 3
427
- },
428
- {
429
- "seconds": 1.2933890700005577,
430
- "bytes": 4945280,
431
- "pages": 34,
432
- "shards": 2
433
- },
434
- {
435
- "seconds": 1.5587586719993851,
436
- "bytes": 6311746,
437
- "pages": 37,
438
- "shards": 2
439
- },
440
- {
441
- "seconds": 1.5925084610062186,
442
- "bytes": 13513629,
443
- "pages": 37,
444
- "shards": 4
445
- },
446
- {
447
- "seconds": 1.9457880960035254,
448
- "bytes": 1015286,
449
- "pages": 33,
450
- "shards": 1
451
- },
452
- {
453
- "seconds": 1.624823366997589,
454
- "bytes": 11605166,
455
- "pages": 31,
456
- "shards": 4
457
- },
458
- {
459
- "seconds": 1.4567563370001153,
460
- "bytes": 1138092,
461
- "pages": 35,
462
- "shards": 1
463
- },
464
- {
465
- "seconds": 1.3836040919995867,
466
- "bytes": 6235413,
467
- "pages": 37,
468
- "shards": 2
469
- },
470
- {
471
- "seconds": 1.6395168350063614,
472
- "bytes": 7620206,
473
- "pages": 34,
474
- "shards": 3
475
- },
476
- {
477
- "seconds": 1.4781058270018548,
478
- "bytes": 11305199,
479
- "pages": 39,
480
- "shards": 3
481
- },
482
- {
483
- "seconds": 1.3714339009966352,
484
- "bytes": 7352671,
485
- "pages": 31,
486
- "shards": 3
487
- },
488
- {
489
- "seconds": 1.3506424060033169,
490
- "bytes": 6203770,
491
- "pages": 38,
492
- "shards": 2
493
- },
494
- {
495
- "seconds": 1.3150392690004082,
496
- "bytes": 914685,
497
  "pages": 26,
498
  "shards": 1
499
  },
500
  {
501
- "seconds": 1.6476991579984315,
502
- "bytes": 12016198,
503
- "pages": 39,
504
- "shards": 4
505
- },
506
- {
507
- "seconds": 1.4772214850017917,
508
- "bytes": 1146549,
509
- "pages": 36,
510
- "shards": 2
511
- },
512
- {
513
- "seconds": 1.649438596003165,
514
- "bytes": 11880426,
515
- "pages": 30,
516
- "shards": 3
517
- },
518
- {
519
- "seconds": 0.9699813629995333,
520
- "bytes": 2462244,
521
- "pages": 31,
522
- "shards": 2
523
- },
524
- {
525
- "seconds": 1.3995045109986677,
526
- "bytes": 10981344,
527
- "pages": 45,
528
- "shards": 3
529
- },
530
- {
531
- "seconds": 1.094142590998672,
532
- "bytes": 7571723,
533
- "pages": 31,
534
  "shards": 3
535
  },
536
  {
537
- "seconds": 1.1913003890003893,
538
- "bytes": 1341787,
539
- "pages": 37,
540
- "shards": 1
541
- },
542
- {
543
- "seconds": 1.4330705320026027,
544
- "bytes": 5823183,
545
- "pages": 26,
546
- "shards": 2
547
- },
548
- {
549
- "seconds": 1.3540045269983239,
550
- "bytes": 17287207,
551
- "pages": 44,
552
- "shards": 5
553
- },
554
- {
555
- "seconds": 1.2279312789978576,
556
- "bytes": 1252954,
557
- "pages": 35,
558
- "shards": 1
559
- },
560
- {
561
- "seconds": 1.2454847960034385,
562
- "bytes": 7379042,
563
- "pages": 32,
564
  "shards": 3
565
  }
566
  ],
@@ -606,7 +258,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 2.9105605440000204,
610
- 0.014060387999961677
611
  ]
612
  }
 
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": 75790.38877363384,
26
+ "required_pages_per_second_upper_estimate": 0.005243354953290438
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": 9473.79859670423,
40
+ "required_pages_per_second_upper_estimate": 0.0039664816370287475
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": 18947.59719340846,
54
+ "required_pages_per_second_upper_estimate": 0.00394628481364685
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": 28421.395790112692,
68
+ "required_pages_per_second_upper_estimate": 0.0017038192806586498
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": 28421.395790112692,
82
+ "required_pages_per_second_upper_estimate": 0.0022096385101729103
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": 22,
95
+ "required_token_positions_per_second": 28421.395790112692,
96
+ "required_pages_per_second_upper_estimate": 0.36455914867834804
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
158
  "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.15217620494523487,
162
+ "producer_p95_seconds": 0.1721363199994812,
163
+ "gpu_update_seconds": 0.9727882544606855,
164
+ "producer_batches_per_second": 6.571329600182103,
165
+ "gpu_batches_per_second": 1.0279729380104417,
166
+ "producer_headroom": 6.392512251246982,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.3128717765625,
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": 373719.53067025304,
174
+ "upload_bytes_per_second": 4658638.935595296,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 80.19151153641408,
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": 0.9768797280003128,
184
+ "bytes": 23252,
185
+ "pages": 1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 1.0272072289999414,
190
+ "bytes": 7383107,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
191
  "pages": 38,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 1.1484548910002559,
196
+ "bytes": 5982767,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
197
  "pages": 37,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.7043199660001846,
202
+ "bytes": 963598,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
203
  "pages": 26,
204
  "shards": 1
205
  },
206
  {
207
+ "seconds": 1.1770538490000035,
208
+ "bytes": 8003790,
209
+ "pages": 38,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.2935991909998847,
214
+ "bytes": 11779732,
215
+ "pages": 36,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
216
  "shards": 3
217
  }
218
  ],
 
258
  },
259
  "cache_reused": true,
260
  "resume_probe_seconds": [
261
+ 2.7244317560000013,
262
+ 0.015869192000025123
263
  ]
264
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fef7bf8813882524f5ca30eee990ca62de8fcdd9a4aac08f179ed7a0bfb2f7cb
3
- size 721756031
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24ae65d790f92deee8ff8c64623716e72d101b0c319d3796bed4c28c51aec73b
3
+ size 721776383
training_config.json CHANGED
@@ -32,7 +32,7 @@
32
  "openmath": 10,
33
  "rewrite6": 5
34
  },
35
- "lr": 0.0001,
36
  "weight_decay": 0.1,
37
  "lr_schedule": "auto",
38
  "warmup_steps": 1000,
 
32
  "openmath": 10,
33
  "rewrite6": 5
34
  },
35
+ "lr": 8e-05,
36
  "weight_decay": 0.1,
37
  "lr_schedule": "auto",
38
  "warmup_steps": 1000,