appvoid commited on
Commit
3cce5c0
·
verified ·
1 Parent(s): 4e9de63

Qwen90M step 215000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +49 -49
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +250 -250
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 210000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "921781da103eac594d0d849dee7135488c4fb169435d08bae6486e287aaf16c7",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "1a7597fc28402f093934d8c6ac10a76d19bc865c277e196a6a417b18ff76bfba",
12
- "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "7f72f0a0e868a4baec06f238f922d0c5d7e397a51386c11ac32acc776d4d3522",
14
- "shard_scheduler.json": "12ca185f760b35fc0306bf401e7f24f0c1ef6c13f220f2c3f65d1588dbf74548",
15
- "coverage.json": "0784ef6420a747808e25c92ea80f2676691796080a1e53c62ef8213031463920",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 215000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "fba9a241ccd9333dd7a1e6f8ef4e6f607cb802bac67c51d4e5ab08a9819714a1",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "e6f406062f59ad512e7807d3f45cba2fde70c4efea152b8334e10eaac4b3bc33",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "7f72f0a0e868a4baec06f238f922d0c5d7e397a51386c11ac32acc776d4d3522",
14
+ "shard_scheduler.json": "a39a540a474143f77dfcb67ff46b317dfd9988fd48b0e36afd807af861668444",
15
+ "coverage.json": "774fb0d49d25ba12bc1cc0687f04fde6ae04f629088fa242a46809dcd920b474",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 210000,
3
  "stats": {
4
- "climbmix/blocks": 7560000,
5
- "climbmix/targets": 12635674730,
6
- "climbmix/input_tokens": 12636475801,
7
- "ultra_qa/blocks": 2835000,
8
- "ultra_qa/targets": 4625316379,
9
- "ultra_qa/input_tokens": 4625325318,
10
- "cortex/blocks": 2835000,
11
- "cortex/targets": 107843015,
12
- "cortex/input_tokens": 4506362531,
13
- "ultra_style/blocks": 2835000,
14
- "ultra_style/targets": 4484394327,
15
- "ultra_style/input_tokens": 4484629586,
16
- "openmath/blocks": 1890000,
17
- "openmath/targets": 2920105808,
18
- "openmath/input_tokens": 2921079037,
19
- "rewrite6/blocks": 945000,
20
- "rewrite6/targets": 1549174599,
21
- "rewrite6/input_tokens": 1550873200
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1035,
26
- "row": 3928,
27
  "start": {
28
- "file": 50,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 4,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 4243288
38
  },
39
  "rewrite6": {
40
- "page": 39,
41
- "row": 727,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 159744
47
  }
48
  },
49
- "epoch": 11,
50
- "offset": 0,
51
- "documents": 2419915
52
  },
53
  "openmath": {
54
- "page": 647,
55
- "row": 820,
56
  "start": {
57
  "file": 6,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 30,
61
- "row": 212
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 2650932
67
  },
68
  "ultra_qa": {
69
- "page": 279,
70
- "row": 3264,
71
  "start": {
72
  "file": 2,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 1,
76
- "row": 38970
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 1146048
82
  },
83
  "ultra_style": {
84
- "page": 427,
85
- "row": 231,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
- "row": 38971
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 1749223
97
  },
98
  "cortex": {
99
- "page": 54982,
100
- "row": 96,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 7037792
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 215000,
3
  "stats": {
4
+ "climbmix/blocks": 7740000,
5
+ "climbmix/targets": 12937648715,
6
+ "climbmix/input_tokens": 12938468575,
7
+ "ultra_qa/blocks": 2902500,
8
+ "ultra_qa/targets": 4735458775,
9
+ "ultra_qa/input_tokens": 4735467933,
10
+ "cortex/blocks": 2902500,
11
+ "cortex/targets": 110418665,
12
+ "cortex/input_tokens": 4613613081,
13
+ "ultra_style/blocks": 2902500,
14
+ "ultra_style/targets": 4591380949,
15
+ "ultra_style/input_tokens": 4591621680,
16
+ "openmath/blocks": 1935000,
17
+ "openmath/targets": 2989665915,
18
+ "openmath/input_tokens": 2990662278,
19
+ "rewrite6/blocks": 967500,
20
+ "rewrite6/targets": 1586083890,
21
+ "rewrite6/input_tokens": 1587822751
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1060,
26
+ "row": 844,
27
  "start": {
28
+ "file": 51,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 23,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 16384,
37
+ "documents": 4342604
38
  },
39
  "rewrite6": {
40
+ "page": 3,
41
+ "row": 133,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 12288
47
  }
48
  },
49
+ "epoch": 12,
50
+ "offset": 2048,
51
+ "documents": 2477269
52
  },
53
  "openmath": {
54
+ "page": 662,
55
+ "row": 2443,
56
  "start": {
57
  "file": 6,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 91,
61
+ "row": 652
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 2713995
67
  },
68
  "ultra_qa": {
69
+ "page": 286,
70
+ "row": 1881,
71
  "start": {
72
  "file": 2,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 2,
76
+ "row": 3058
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 1173337
82
  },
83
  "ultra_style": {
84
+ "page": 437,
85
+ "row": 479,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
+ "row": 79931
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 1790431
97
  },
98
  "cortex": {
99
+ "page": 56290,
100
+ "row": 85,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 7205205
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 210000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "921781da103eac594d0d849dee7135488c4fb169435d08bae6486e287aaf16c7", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "1a7597fc28402f093934d8c6ac10a76d19bc865c277e196a6a417b18ff76bfba", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "7f72f0a0e868a4baec06f238f922d0c5d7e397a51386c11ac32acc776d4d3522", "shard_scheduler.json": "12ca185f760b35fc0306bf401e7f24f0c1ef6c13f220f2c3f65d1588dbf74548", "coverage.json": "0784ef6420a747808e25c92ea80f2676691796080a1e53c62ef8213031463920", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000210000"}
 
1
+ {"step": 215000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "fba9a241ccd9333dd7a1e6f8ef4e6f607cb802bac67c51d4e5ab08a9819714a1", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "e6f406062f59ad512e7807d3f45cba2fde70c4efea152b8334e10eaac4b3bc33", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "7f72f0a0e868a4baec06f238f922d0c5d7e397a51386c11ac32acc776d4d3522", "shard_scheduler.json": "a39a540a474143f77dfcb67ff46b317dfd9988fd48b0e36afd807af861668444", "coverage.json": "774fb0d49d25ba12bc1cc0687f04fde6ae04f629088fa242a46809dcd920b474", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000215000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:921781da103eac594d0d849dee7135488c4fb169435d08bae6486e287aaf16c7
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fba9a241ccd9333dd7a1e6f8ef4e6f607cb802bac67c51d4e5ab08a9819714a1
3
  size 360622744
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
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": 74616.38181288075,
26
- "required_pages_per_second_upper_estimate": 0.006467558862049758
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
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": 9327.047726610093,
40
- "required_pages_per_second_upper_estimate": 0.0037206987899354132
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
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": 18654.095453220187,
54
- "required_pages_per_second_upper_estimate": 0.004315296682235381
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
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": 27981.14317983028,
68
- "required_pages_per_second_upper_estimate": 0.0017039048139391429
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
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": 27981.14317983028,
82
- "required_pages_per_second_upper_estimate": 0.00215406585874511
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
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": 27981.14317983028,
96
- "required_pages_per_second_upper_estimate": 0.3474321514313953
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,413 +155,413 @@
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": 688946,
161
- "producer_batch_seconds": 0.11913276657806193,
162
- "producer_p95_seconds": 0.32301659399854543,
163
- "gpu_update_seconds": 0.9880940111099384,
164
- "producer_batches_per_second": 8.393996284345066,
165
- "gpu_batches_per_second": 1.0120494495019632,
166
- "producer_headroom": 8.294057457840436,
167
  "sustainable": true,
168
- "buffer_seconds": 60.273734677706244,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 86,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 355577.373587176,
174
- "upload_bytes_per_second": 5217106.874362177,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 76.29862695418795,
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": 0.9532060609999462,
184
- "bytes": 7422959,
185
- "pages": 29,
186
- "shards": 3
187
- },
188
- {
189
- "seconds": 1.2087593340002059,
190
- "bytes": 6148456,
191
- "pages": 35,
192
  "shards": 2
193
  },
194
  {
195
- "seconds": 1.2602108099999896,
196
- "bytes": 2561145,
197
- "pages": 35,
198
- "shards": 3
199
  },
200
  {
201
- "seconds": 1.5218778089993066,
202
- "bytes": 10001472,
203
- "pages": 31,
204
  "shards": 3
205
  },
206
  {
207
- "seconds": 1.655338631999257,
208
- "bytes": 6858339,
209
- "pages": 37,
210
  "shards": 2
211
  },
212
  {
213
- "seconds": 1.3612198569990142,
214
- "bytes": 7511604,
215
- "pages": 29,
216
- "shards": 3
217
- },
218
- {
219
- "seconds": 1.111224463000326,
220
- "bytes": 5878363,
221
  "pages": 35,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 1.1176849410003342,
226
- "bytes": 5784032,
227
- "pages": 33,
228
  "shards": 2
229
  },
230
  {
231
- "seconds": 1.2254727769995952,
232
- "bytes": 7561832,
233
- "pages": 34,
234
- "shards": 3
235
  },
236
  {
237
- "seconds": 1.1802483909996226,
238
- "bytes": 1055971,
239
- "pages": 33,
240
  "shards": 1
241
  },
242
  {
243
- "seconds": 1.59867495799881,
244
- "bytes": 13042114,
245
- "pages": 35,
246
- "shards": 4
247
  },
248
  {
249
- "seconds": 1.1432860509994498,
250
- "bytes": 5906480,
251
- "pages": 31,
252
  "shards": 2
253
  },
254
  {
255
- "seconds": 1.2417195279995212,
256
- "bytes": 5623378,
257
- "pages": 33,
258
  "shards": 2
259
  },
260
  {
261
- "seconds": 0.9634915199985699,
262
- "bytes": 7597611,
263
- "pages": 38,
264
  "shards": 3
265
  },
266
  {
267
- "seconds": 0.761399758999687,
268
- "bytes": 1061264,
269
- "pages": 32,
270
- "shards": 1
271
  },
272
  {
273
- "seconds": 1.0512442660001398,
274
- "bytes": 5862459,
275
- "pages": 28,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.2080343199995696,
280
- "bytes": 17359027,
281
- "pages": 39,
282
- "shards": 5
283
  },
284
  {
285
- "seconds": 1.0087294929999189,
286
- "bytes": 1073260,
287
- "pages": 33,
288
- "shards": 1
289
  },
290
  {
291
- "seconds": 1.4433996169991588,
292
- "bytes": 7239065,
293
  "pages": 29,
294
- "shards": 3
295
  },
296
  {
297
- "seconds": 0.894821537000098,
298
- "bytes": 1079304,
299
- "pages": 31,
300
- "shards": 1
301
  },
302
  {
303
- "seconds": 1.1567702649990679,
304
- "bytes": 10579644,
305
- "pages": 35,
306
- "shards": 3
307
  },
308
  {
309
- "seconds": 1.6480453410003975,
310
- "bytes": 7697533,
311
- "pages": 38,
312
  "shards": 3
313
  },
314
  {
315
- "seconds": 1.2853153289997863,
316
- "bytes": 6668992,
317
- "pages": 31,
 
 
 
 
 
 
318
  "shards": 2
319
  },
320
  {
321
- "seconds": 1.2017098229989642,
322
- "bytes": 7521074,
323
- "pages": 38,
324
- "shards": 3
325
  },
326
  {
327
- "seconds": 1.335906720998537,
328
- "bytes": 5362985,
329
- "pages": 26,
330
- "shards": 2
331
  },
332
  {
333
- "seconds": 1.582250990999455,
334
- "bytes": 5900156,
335
- "pages": 27,
336
  "shards": 2
337
  },
338
  {
339
- "seconds": 1.2782940739998594,
340
- "bytes": 7640268,
341
- "pages": 37,
342
- "shards": 3
343
  },
344
  {
345
- "seconds": 0.723643687000731,
346
- "bytes": 1209843,
347
  "pages": 34,
348
  "shards": 1
349
  },
350
  {
351
- "seconds": 1.586552576000031,
352
- "bytes": 9572867,
353
- "pages": 30,
354
- "shards": 3
355
- },
356
- {
357
- "seconds": 1.2363076699984958,
358
- "bytes": 8177813,
359
- "pages": 33,
360
- "shards": 3
361
- },
362
- {
363
- "seconds": 1.3350522609998734,
364
- "bytes": 6097996,
365
- "pages": 35,
366
  "shards": 2
367
  },
368
  {
369
- "seconds": 0.9499315409993869,
370
- "bytes": 5779343,
371
- "pages": 35,
372
- "shards": 2
373
  },
374
  {
375
- "seconds": 1.612262433000069,
376
- "bytes": 7066450,
377
- "pages": 31,
378
- "shards": 3
379
  },
380
  {
381
- "seconds": 1.120285896000496,
382
- "bytes": 6092850,
383
- "pages": 35,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 1.532518773999982,
388
- "bytes": 2518854,
389
- "pages": 32,
390
  "shards": 3
391
  },
392
  {
393
- "seconds": 1.2675222210000356,
394
- "bytes": 11869039,
395
- "pages": 39,
396
  "shards": 3
397
  },
398
  {
399
- "seconds": 1.6028917950006871,
400
- "bytes": 10461677,
401
- "pages": 31,
402
  "shards": 3
403
  },
404
  {
405
- "seconds": 1.3869329340013792,
406
- "bytes": 2462257,
407
- "pages": 29,
408
- "shards": 2
409
  },
410
  {
411
- "seconds": 1.3128174869998475,
412
- "bytes": 5972623,
413
- "pages": 34,
414
  "shards": 2
415
  },
416
  {
417
- "seconds": 0.8817397439997876,
418
- "bytes": 1063209,
419
- "pages": 34,
420
- "shards": 1
421
- },
422
- {
423
- "seconds": 1.6997340180005267,
424
- "bytes": 11573294,
425
- "pages": 33,
426
- "shards": 4
427
  },
428
  {
429
- "seconds": 1.4719399390014587,
430
- "bytes": 11935199,
431
- "pages": 35,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 0.9283750220001821,
436
- "bytes": 2657482,
437
- "pages": 36,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 0.9979908650002471,
442
- "bytes": 5962542,
443
- "pages": 27,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 1.4787909560000116,
448
- "bytes": 11055954,
449
- "pages": 41,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 1.0256725109993567,
454
- "bytes": 2318063,
455
- "pages": 28,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 0.9941544879984576,
460
- "bytes": 6059545,
461
  "pages": 35,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.1885242320004181,
466
- "bytes": 7012163,
467
- "pages": 35,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.3779209780004749,
472
- "bytes": 11061746,
473
- "pages": 35,
474
  "shards": 4
475
  },
476
  {
477
- "seconds": 0.9865021240002534,
478
- "bytes": 862370,
479
- "pages": 26,
480
- "shards": 1
481
  },
482
  {
483
- "seconds": 1.1108503920004296,
484
- "bytes": 7694144,
485
- "pages": 39,
486
- "shards": 3
487
  },
488
  {
489
- "seconds": 2.1076598479994573,
490
- "bytes": 6179293,
491
- "pages": 35,
492
- "shards": 2
493
  },
494
  {
495
- "seconds": 1.0649322700010089,
496
- "bytes": 5823579,
497
- "pages": 32,
498
- "shards": 2
499
  },
500
  {
501
- "seconds": 1.1941432809999242,
502
- "bytes": 7137236,
503
- "pages": 22,
504
  "shards": 3
505
  },
506
  {
507
- "seconds": 1.292083440001079,
508
- "bytes": 7124472,
509
- "pages": 44,
510
  "shards": 2
511
  },
512
  {
513
- "seconds": 1.2287268149993906,
514
- "bytes": 6045685,
515
- "pages": 35,
516
- "shards": 2
517
  },
518
  {
519
- "seconds": 1.297586668999429,
520
- "bytes": 7207958,
521
- "pages": 28,
522
- "shards": 3
523
  },
524
  {
525
- "seconds": 1.0006021399985912,
526
- "bytes": 4975022,
527
- "pages": 36,
528
- "shards": 2
529
  },
530
  {
531
- "seconds": 1.1054393590002292,
532
- "bytes": 5686965,
533
- "pages": 29,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 0.952104601999963,
538
- "bytes": 2531508,
539
- "pages": 33,
540
- "shards": 2
541
  },
542
  {
543
- "seconds": 1.6972403340005258,
544
- "bytes": 16066476,
545
- "pages": 41,
546
  "shards": 4
547
  },
548
  {
549
- "seconds": 1.3755702170001314,
550
- "bytes": 7451679,
551
- "pages": 30,
552
  "shards": 3
553
  },
554
  {
555
- "seconds": 1.598954668999795,
556
- "bytes": 1262837,
557
  "pages": 34,
558
  "shards": 1
559
  },
560
  {
561
- "seconds": 1.0224317270003667,
562
- "bytes": 6094784,
 
 
 
 
 
 
563
  "pages": 35,
 
 
 
 
 
 
564
  "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
565
  }
566
  ],
567
  "cache_key": {
 
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": 75396.39935795023,
26
+ "required_pages_per_second_upper_estimate": 0.006535168805919986
27
  },
28
  {
29
  "source": "rewrite6",
 
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": 9424.549919743778,
40
+ "required_pages_per_second_upper_estimate": 0.003759593872564137
41
  },
42
  {
43
  "source": "openmath",
 
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": 18849.099839487557,
54
+ "required_pages_per_second_upper_estimate": 0.004360407515038354
55
  },
56
  {
57
  "source": "ultra_qa",
 
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": 28273.649759231335,
68
+ "required_pages_per_second_upper_estimate": 0.00172171692996125
69
  },
70
  {
71
  "source": "ultra_style",
 
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": 28273.649759231335,
82
+ "required_pages_per_second_upper_estimate": 0.002176583824937439
83
  },
84
  {
85
  "source": "cortex",
 
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": 22,
95
+ "required_token_positions_per_second": 28273.649759231335,
96
+ "required_pages_per_second_upper_estimate": 0.35106410419100953
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
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": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
+ "producer_batch_seconds": 0.06690109020320278,
162
+ "producer_p95_seconds": 0.21632020699689747,
163
+ "gpu_update_seconds": 0.977871630845004,
164
+ "producer_batches_per_second": 14.947439525464214,
165
+ "gpu_batches_per_second": 1.0226291145555315,
166
+ "producer_headroom": 14.616677065722763,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.62804111239024,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 81,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 294982.2025964519,
174
+ "upload_bytes_per_second": 5236085.920943062,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 63.29631384296577,
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": 0.9283750220001821,
184
+ "bytes": 2657482,
185
+ "pages": 36,
 
 
 
 
 
 
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 0.9979908650002471,
190
+ "bytes": 5962542,
191
+ "pages": 27,
192
+ "shards": 2
193
  },
194
  {
195
+ "seconds": 1.4787909560000116,
196
+ "bytes": 11055954,
197
+ "pages": 41,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 1.0256725109993567,
202
+ "bytes": 2318063,
203
+ "pages": 28,
204
  "shards": 2
205
  },
206
  {
207
+ "seconds": 0.9941544879984576,
208
+ "bytes": 6059545,
 
 
 
 
 
 
209
  "pages": 35,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.1885242320004181,
214
+ "bytes": 7012163,
215
+ "pages": 35,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.3779209780004749,
220
+ "bytes": 11061746,
221
+ "pages": 35,
222
+ "shards": 4
223
  },
224
  {
225
+ "seconds": 0.9865021240002534,
226
+ "bytes": 862370,
227
+ "pages": 26,
228
  "shards": 1
229
  },
230
  {
231
+ "seconds": 1.1108503920004296,
232
+ "bytes": 7694144,
233
+ "pages": 39,
234
+ "shards": 3
235
  },
236
  {
237
+ "seconds": 2.1076598479994573,
238
+ "bytes": 6179293,
239
+ "pages": 35,
240
  "shards": 2
241
  },
242
  {
243
+ "seconds": 1.0649322700010089,
244
+ "bytes": 5823579,
245
+ "pages": 32,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.1941432809999242,
250
+ "bytes": 7137236,
251
+ "pages": 22,
252
  "shards": 3
253
  },
254
  {
255
+ "seconds": 1.292083440001079,
256
+ "bytes": 7124472,
257
+ "pages": 44,
258
+ "shards": 2
259
  },
260
  {
261
+ "seconds": 1.2287268149993906,
262
+ "bytes": 6045685,
263
+ "pages": 35,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.297586668999429,
268
+ "bytes": 7207958,
269
+ "pages": 28,
270
+ "shards": 3
271
  },
272
  {
273
+ "seconds": 1.0006021399985912,
274
+ "bytes": 4975022,
275
+ "pages": 36,
276
+ "shards": 2
277
  },
278
  {
279
+ "seconds": 1.1054393590002292,
280
+ "bytes": 5686965,
281
  "pages": 29,
282
+ "shards": 2
283
  },
284
  {
285
+ "seconds": 0.952104601999963,
286
+ "bytes": 2531508,
287
+ "pages": 33,
288
+ "shards": 2
289
  },
290
  {
291
+ "seconds": 1.6972403340005258,
292
+ "bytes": 16066476,
293
+ "pages": 41,
294
+ "shards": 4
295
  },
296
  {
297
+ "seconds": 1.3755702170001314,
298
+ "bytes": 7451679,
299
+ "pages": 30,
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 1.598954668999795,
304
+ "bytes": 1262837,
305
+ "pages": 34,
306
+ "shards": 1
307
+ },
308
+ {
309
+ "seconds": 1.0224317270003667,
310
+ "bytes": 6094784,
311
+ "pages": 35,
312
  "shards": 2
313
  },
314
  {
315
+ "seconds": 1.438195700000506,
316
+ "bytes": 12388203,
317
+ "pages": 31,
318
+ "shards": 4
319
  },
320
  {
321
+ "seconds": 0.9514959919997636,
322
+ "bytes": 336067,
323
+ "pages": 9,
324
+ "shards": 1
325
  },
326
  {
327
+ "seconds": 1.108248578999337,
328
+ "bytes": 861803,
329
+ "pages": 26,
330
  "shards": 2
331
  },
332
  {
333
+ "seconds": 1.207310141000562,
334
+ "bytes": 13710377,
335
+ "pages": 39,
336
+ "shards": 4
337
  },
338
  {
339
+ "seconds": 0.9586756399985461,
340
+ "bytes": 1163538,
341
  "pages": 34,
342
  "shards": 1
343
  },
344
  {
345
+ "seconds": 1.222915964999629,
346
+ "bytes": 5976864,
347
+ "pages": 26,
 
 
 
 
 
 
 
 
 
 
 
 
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 1.3758100840004772,
352
+ "bytes": 11655433,
353
+ "pages": 40,
354
+ "shards": 4
355
  },
356
  {
357
+ "seconds": 1.3469852179987356,
358
+ "bytes": 1264752,
359
+ "pages": 33,
360
+ "shards": 1
361
  },
362
  {
363
+ "seconds": 1.1859561970013601,
364
+ "bytes": 6123105,
365
+ "pages": 28,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 0.9817681739987165,
370
+ "bytes": 8356010,
371
+ "pages": 35,
372
  "shards": 3
373
  },
374
  {
375
+ "seconds": 1.2058971119986381,
376
+ "bytes": 10431329,
377
+ "pages": 35,
378
  "shards": 3
379
  },
380
  {
381
+ "seconds": 1.240897143001348,
382
+ "bytes": 7984887,
383
+ "pages": 38,
384
  "shards": 3
385
  },
386
  {
387
+ "seconds": 1.0015397960014525,
388
+ "bytes": 971215,
389
+ "pages": 33,
390
+ "shards": 1
391
  },
392
  {
393
+ "seconds": 1.235119819000829,
394
+ "bytes": 5680772,
395
+ "pages": 24,
396
  "shards": 2
397
  },
398
  {
399
+ "seconds": 1.104966183000215,
400
+ "bytes": 6761570,
401
+ "pages": 43,
402
+ "shards": 3
 
 
 
 
 
 
403
  },
404
  {
405
+ "seconds": 1.8051689160020032,
406
+ "bytes": 11661330,
407
+ "pages": 30,
408
  "shards": 3
409
  },
410
  {
411
+ "seconds": 1.1013407509999524,
412
+ "bytes": 6091571,
413
+ "pages": 35,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 0.8600726500008022,
418
+ "bytes": 2195975,
419
+ "pages": 20,
420
  "shards": 2
421
  },
422
  {
423
+ "seconds": 1.1491692820018216,
424
+ "bytes": 11269048,
425
+ "pages": 45,
426
  "shards": 3
427
  },
428
  {
429
+ "seconds": 1.0205925500013109,
430
+ "bytes": 2657105,
431
+ "pages": 36,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.0683662000010372,
436
+ "bytes": 6645182,
437
  "pages": 35,
438
  "shards": 2
439
  },
440
  {
441
+ "seconds": 1.0938033080019522,
442
+ "bytes": 6719573,
443
+ "pages": 31,
444
  "shards": 2
445
  },
446
  {
447
+ "seconds": 1.3891484439991473,
448
+ "bytes": 11561860,
449
+ "pages": 34,
450
  "shards": 4
451
  },
452
  {
453
+ "seconds": 1.268800834997819,
454
+ "bytes": 6327049,
455
+ "pages": 36,
456
+ "shards": 2
457
  },
458
  {
459
+ "seconds": 1.941673843000899,
460
+ "bytes": 1008328,
461
+ "pages": 30,
462
+ "shards": 1
463
  },
464
  {
465
+ "seconds": 1.119837071000802,
466
+ "bytes": 7224601,
467
+ "pages": 32,
468
+ "shards": 3
469
  },
470
  {
471
+ "seconds": 1.2717098209977848,
472
+ "bytes": 11010766,
473
+ "pages": 39,
474
+ "shards": 3
475
  },
476
  {
477
+ "seconds": 1.1728442359999462,
478
+ "bytes": 8097014,
479
+ "pages": 29,
480
  "shards": 3
481
  },
482
  {
483
+ "seconds": 1.3505811920003907,
484
+ "bytes": 5904996,
485
+ "pages": 27,
486
  "shards": 2
487
  },
488
  {
489
+ "seconds": 0.9558858569980657,
490
+ "bytes": 1199021,
491
+ "pages": 33,
492
+ "shards": 1
493
  },
494
  {
495
+ "seconds": 1.896848256001249,
496
+ "bytes": 11337881,
497
+ "pages": 42,
498
+ "shards": 4
499
  },
500
  {
501
+ "seconds": 1.1246725650016742,
502
+ "bytes": 878351,
503
+ "pages": 26,
504
+ "shards": 1
505
  },
506
  {
507
+ "seconds": 1.1035908939993533,
508
+ "bytes": 6360663,
509
+ "pages": 35,
510
  "shards": 2
511
  },
512
  {
513
+ "seconds": 0.9343603500019526,
514
+ "bytes": 2109863,
515
+ "pages": 20,
516
+ "shards": 3
517
  },
518
  {
519
+ "seconds": 1.3331431920014438,
520
+ "bytes": 16665888,
521
+ "pages": 43,
522
  "shards": 4
523
  },
524
  {
525
+ "seconds": 1.675639938999666,
526
+ "bytes": 7647558,
527
+ "pages": 37,
528
  "shards": 3
529
  },
530
  {
531
+ "seconds": 0.968150565000542,
532
+ "bytes": 1141925,
533
  "pages": 34,
534
  "shards": 1
535
  },
536
  {
537
+ "seconds": 1.0775915699996403,
538
+ "bytes": 5849379,
539
+ "pages": 27,
540
+ "shards": 2
541
+ },
542
+ {
543
+ "seconds": 1.3228814079993754,
544
+ "bytes": 11625452,
545
  "pages": 35,
546
+ "shards": 4
547
+ },
548
+ {
549
+ "seconds": 1.1623286959984398,
550
+ "bytes": 7130450,
551
+ "pages": 38,
552
  "shards": 2
553
+ },
554
+ {
555
+ "seconds": 1.6525030769989826,
556
+ "bytes": 8063307,
557
+ "pages": 38,
558
+ "shards": 3
559
+ },
560
+ {
561
+ "seconds": 1.0988706719981565,
562
+ "bytes": 771997,
563
+ "pages": 25,
564
+ "shards": 1
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1a7597fc28402f093934d8c6ac10a76d19bc865c277e196a6a417b18ff76bfba
3
- size 721768575
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e6f406062f59ad512e7807d3f45cba2fde70c4efea152b8334e10eaac4b3bc33
3
+ size 721752255
training_config.json CHANGED
@@ -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,
 
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
+ "prefetch_batches": 62,
50
+ "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,