appvoid commited on
Commit
0fb2bf3
·
verified ·
1 Parent(s): e10c2a3

Qwen90M step 910000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +46 -46
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +248 -248
  6. trainer.pt +2 -2
  7. training_config.json +1 -1
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 905000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "2eafced5169e47d39f63c15f6d5c04b0ee60f62d179c1e2dbe090427243f8630",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "3c5628671e1b2c65e85ea6083662d90b0c22e2a2bf9df3cd5c38a446fd7e4fa4",
12
- "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
  "performance.json": "49b9042fb538cca213af1e6f33ceb5a89185fc6d2bab488be3f60e812871f343",
14
- "shard_scheduler.json": "b24e87ee5310f8cae440ce163c660971b4bd3a3c4202793a1fd52e7a8b911217",
15
- "coverage.json": "2b0b0982d723d593e8b8cd110f0581bc41bb25cbc26ab9aa84796b740f5c7b28",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
 
1
  {
2
+ "step": 910000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "b4df0ef53a5a343fee1e4dce6ab8a731c5f01c993a900224ba917205158c612a",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "8ba9bbe2eb54c264d2b9324ec47dc07ced3c51c8d254214772fea4aa3594fd98",
12
+ "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58",
13
  "performance.json": "49b9042fb538cca213af1e6f33ceb5a89185fc6d2bab488be3f60e812871f343",
14
+ "shard_scheduler.json": "dd11a2928b6fc6dbcfc1eafd5d86aa23cf716d5b4273f83412bf2508c81e3964",
15
+ "coverage.json": "26aa8059e8f72378a7c5a4bd2300bf3c7cff681ad595ec96d08d6b092bc45174",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 905000,
3
  "stats": {
4
- "climbmix/blocks": 32580000,
5
- "climbmix/targets": 54463088890,
6
- "climbmix/input_tokens": 54466540635,
7
- "ultra_qa/blocks": 12217500,
8
- "ultra_qa/targets": 19931980030,
9
- "ultra_qa/input_tokens": 19932018303,
10
- "cortex/blocks": 12217500,
11
- "cortex/targets": 464554652,
12
- "cortex/input_tokens": 19420649920,
13
- "ultra_style/blocks": 12217500,
14
- "ultra_style/targets": 19324167416,
15
- "ultra_style/input_tokens": 19325180301,
16
- "openmath/blocks": 8145000,
17
- "openmath/targets": 12585378739,
18
- "openmath/input_tokens": 12589571832,
19
- "rewrite6/blocks": 4072500,
20
- "rewrite6/targets": 6676302372,
21
- "rewrite6/input_tokens": 6683622035
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 4462,
26
- "row": 661,
27
  "start": {
28
- "file": 215,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 82,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 0,
37
- "documents": 18277013
38
  },
39
  "rewrite6": {
40
- "page": 38,
41
- "row": 2160,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 155648
47
  }
48
  },
49
- "epoch": 50,
50
- "offset": 0,
51
- "documents": 10428008
52
  },
53
  "openmath": {
54
- "page": 2788,
55
- "row": 2605,
56
  "start": {
57
  "file": 26,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 66,
61
- "row": 751
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 11422253
67
  },
68
  "ultra_qa": {
69
- "page": 1205,
70
- "row": 2927,
71
  "start": {
72
  "file": 9,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 4,
76
- "row": 744
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 4938607
82
  },
83
  "ultra_style": {
84
- "page": 1840,
85
- "row": 498,
86
  "start": {
87
  "file": 11,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 0,
91
- "row": 2284
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 0,
96
- "documents": 7537138
97
  },
98
  "cortex": {
99
- "page": 236942,
100
- "row": 28,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 30328604
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 910000,
3
  "stats": {
4
+ "climbmix/blocks": 32760000,
5
+ "climbmix/targets": 54764250124,
6
+ "climbmix/input_tokens": 54767720896,
7
+ "ultra_qa/blocks": 12285000,
8
+ "ultra_qa/targets": 20042043648,
9
+ "ultra_qa/input_tokens": 20042082120,
10
+ "cortex/blocks": 12285000,
11
+ "cortex/targets": 467130820,
12
+ "cortex/input_tokens": 19527965675,
13
+ "ultra_style/blocks": 12285000,
14
+ "ultra_style/targets": 19430317419,
15
+ "ultra_style/input_tokens": 19431336007,
16
+ "openmath/blocks": 8190000,
17
+ "openmath/targets": 12654930072,
18
+ "openmath/input_tokens": 12659146541,
19
+ "rewrite6/blocks": 4095000,
20
+ "rewrite6/targets": 6713205114,
21
+ "rewrite6/input_tokens": 6720565027
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 4486,
26
+ "row": 2973,
27
  "start": {
28
+ "file": 217,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 13,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 0,
37
+ "documents": 18377629
38
  },
39
  "rewrite6": {
40
+ "page": 2,
41
+ "row": 1549,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 8192
47
  }
48
  },
49
+ "epoch": 51,
50
+ "offset": 2048,
51
+ "documents": 10485345
52
  },
53
  "openmath": {
54
+ "page": 2804,
55
+ "row": 458,
56
  "start": {
57
  "file": 26,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 132,
61
+ "row": 287
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 11485642
67
  },
68
  "ultra_qa": {
69
+ "page": 1212,
70
+ "row": 1525,
71
  "start": {
72
  "file": 9,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 4,
76
+ "row": 29416
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 4965877
82
  },
83
  "ultra_style": {
84
+ "page": 1850,
85
+ "row": 2060,
86
  "start": {
87
  "file": 11,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 0,
91
+ "row": 43244
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 0,
96
+ "documents": 7579660
97
  },
98
  "cortex": {
99
+ "page": 238252,
100
+ "row": 11,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 30496267
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 905000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "2eafced5169e47d39f63c15f6d5c04b0ee60f62d179c1e2dbe090427243f8630", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "3c5628671e1b2c65e85ea6083662d90b0c22e2a2bf9df3cd5c38a446fd7e4fa4", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "49b9042fb538cca213af1e6f33ceb5a89185fc6d2bab488be3f60e812871f343", "shard_scheduler.json": "b24e87ee5310f8cae440ce163c660971b4bd3a3c4202793a1fd52e7a8b911217", "coverage.json": "2b0b0982d723d593e8b8cd110f0581bc41bb25cbc26ab9aa84796b740f5c7b28", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000905000"}
 
1
+ {"step": 910000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "b4df0ef53a5a343fee1e4dce6ab8a731c5f01c993a900224ba917205158c612a", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "8ba9bbe2eb54c264d2b9324ec47dc07ced3c51c8d254214772fea4aa3594fd98", "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58", "performance.json": "49b9042fb538cca213af1e6f33ceb5a89185fc6d2bab488be3f60e812871f343", "shard_scheduler.json": "dd11a2928b6fc6dbcfc1eafd5d86aa23cf716d5b4273f83412bf2508c81e3964", "coverage.json": "26aa8059e8f72378a7c5a4bd2300bf3c7cff681ad595ec96d08d6b092bc45174", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000910000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2eafced5169e47d39f63c15f6d5c04b0ee60f62d179c1e2dbe090427243f8630
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4df0ef53a5a343fee1e4dce6ab8a731c5f01c993a900224ba917205158c612a
3
  size 360622744
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": 76565.79414645793,
26
- "required_pages_per_second_upper_estimate": 0.005296999296170754
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": 9570.724268307242,
40
- "required_pages_per_second_upper_estimate": 0.0040070623916907445
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": 19141.448536614484,
54
- "required_pages_per_second_upper_estimate": 0.003986658936232964
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": 28712.172804921724,
68
- "required_pages_per_second_upper_estimate": 0.0017212509187056575
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": 28712.172804921724,
82
- "required_pages_per_second_upper_estimate": 0.0022322451440813925
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 23,
95
- "required_token_positions_per_second": 28712.172804921724,
96
- "required_pages_per_second_upper_estimate": 0.36828892401228464
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": 2.558590814491007,
157
  "selected": {
158
- "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.067691734976421,
162
- "producer_p95_seconds": 0.24887008000223432,
163
- "gpu_update_seconds": 0.9629365282748887,
164
- "producer_batches_per_second": 14.772852259560626,
165
- "gpu_batches_per_second": 1.03849004647431,
166
- "producer_headroom": 14.225319067539154,
167
  "sustainable": true,
168
- "buffer_seconds": 60.66500128131799,
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": 346296.08435846475,
174
- "upload_bytes_per_second": 5025204.241395261,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 74.30707834306199,
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": 1.3964128229999915,
184
- "bytes": 12688722,
185
- "pages": 35,
186
- "shards": 4
187
  },
188
  {
189
- "seconds": 1.4358708510008,
190
- "bytes": 6164418,
191
- "pages": 37,
192
  "shards": 2
193
  },
194
  {
195
- "seconds": 1.5897931969993806,
196
- "bytes": 7707081,
197
- "pages": 36,
198
  "shards": 3
199
  },
200
  {
201
- "seconds": 1.0018968719996337,
202
- "bytes": 927002,
203
- "pages": 29,
204
- "shards": 1
205
  },
206
  {
207
- "seconds": 1.3852189129993349,
208
- "bytes": 10288556,
209
- "pages": 35,
210
  "shards": 3
211
  },
212
  {
213
- "seconds": 1.2743636189989047,
214
- "bytes": 2631065,
215
- "pages": 37,
216
- "shards": 3
217
  },
218
  {
219
- "seconds": 1.369053440999778,
220
- "bytes": 11976832,
221
  "pages": 37,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 1.3680221860013262,
226
- "bytes": 5930978,
227
- "pages": 33,
228
  "shards": 2
229
  },
230
  {
231
- "seconds": 1.3103218420001213,
232
- "bytes": 6746870,
233
  "pages": 31,
234
- "shards": 3
235
- },
236
- {
237
- "seconds": 1.4409255560003658,
238
- "bytes": 6062763,
239
- "pages": 36,
240
  "shards": 2
241
  },
242
  {
243
- "seconds": 1.1668520560015168,
244
- "bytes": 7696896,
245
- "pages": 38,
246
- "shards": 3
247
- },
248
- {
249
- "seconds": 1.1897505889992317,
250
- "bytes": 1090330,
251
- "pages": 30,
252
- "shards": 1
253
  },
254
  {
255
- "seconds": 1.3477415189991007,
256
- "bytes": 15713674,
257
  "pages": 35,
258
- "shards": 4
259
  },
260
  {
261
- "seconds": 1.2074203040010616,
262
- "bytes": 7463350,
263
- "pages": 39,
264
- "shards": 3
265
  },
266
  {
267
- "seconds": 1.190478869000799,
268
- "bytes": 1226986,
269
- "pages": 34,
270
- "shards": 1
271
  },
272
  {
273
- "seconds": 1.5480925499996374,
274
- "bytes": 11721285,
275
- "pages": 32,
276
- "shards": 4
277
  },
278
  {
279
- "seconds": 1.0147747470000468,
280
- "bytes": 1162064,
281
- "pages": 33,
282
- "shards": 1
283
  },
284
  {
285
- "seconds": 1.132629542000359,
286
- "bytes": 6067377,
287
- "pages": 33,
288
  "shards": 2
289
  },
290
  {
291
- "seconds": 1.3739411900005507,
292
- "bytes": 13347857,
293
- "pages": 36,
294
- "shards": 4
295
  },
296
  {
297
- "seconds": 1.5033581300012884,
298
- "bytes": 979335,
299
- "pages": 32,
300
- "shards": 1
301
  },
302
  {
303
- "seconds": 1.363758608000353,
304
- "bytes": 11974676,
305
- "pages": 44,
306
- "shards": 4
307
  },
308
  {
309
- "seconds": 1.1600040840003203,
310
- "bytes": 6119662,
311
  "pages": 36,
312
- "shards": 2
313
  },
314
  {
315
- "seconds": 0.8502458040002239,
316
- "bytes": 891680,
317
- "pages": 27,
318
- "shards": 1
319
  },
320
  {
321
- "seconds": 1.5061208509996504,
322
- "bytes": 11821724,
323
- "pages": 33,
324
- "shards": 4
325
  },
326
  {
327
- "seconds": 1.1246318519988563,
328
- "bytes": 6660673,
329
- "pages": 27,
330
  "shards": 2
331
  },
332
  {
333
- "seconds": 1.2537715590005973,
334
- "bytes": 5914249,
335
- "pages": 27,
336
- "shards": 2
337
  },
338
  {
339
- "seconds": 1.3754841779991693,
340
- "bytes": 2754086,
341
  "pages": 37,
342
- "shards": 2
343
- },
344
- {
345
- "seconds": 1.24713432200042,
346
- "bytes": 5975341,
347
- "pages": 36,
348
- "shards": 2
349
  },
350
  {
351
- "seconds": 1.473726465999789,
352
- "bytes": 9735362,
353
- "pages": 31,
354
  "shards": 3
355
  },
356
  {
357
- "seconds": 1.1486626710011478,
358
- "bytes": 2585051,
359
- "pages": 37,
360
  "shards": 2
361
  },
362
  {
363
- "seconds": 1.5323144779995346,
364
- "bytes": 11971610,
365
- "pages": 34,
366
- "shards": 3
367
  },
368
  {
369
- "seconds": 1.079354624998814,
370
- "bytes": 2321724,
371
- "pages": 26,
372
  "shards": 2
373
  },
374
  {
375
- "seconds": 1.6746519440002885,
376
- "bytes": 10811116,
377
- "pages": 42,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 1.732792734999748,
382
- "bytes": 6311900,
383
- "pages": 36,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 1.1773680320002313,
388
- "bytes": 2462043,
389
- "pages": 33,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 1.2794374899985996,
394
- "bytes": 6045695,
395
- "pages": 32,
396
- "shards": 3
397
  },
398
  {
399
- "seconds": 1.710491572997853,
400
- "bytes": 17528313,
401
- "pages": 40,
402
- "shards": 5
403
  },
404
  {
405
- "seconds": 1.008943750999606,
406
- "bytes": 1209380,
407
- "pages": 35,
408
- "shards": 1
409
  },
410
  {
411
- "seconds": 1.5094498580001527,
412
- "bytes": 5827665,
413
- "pages": 28,
414
  "shards": 2
415
  },
416
  {
417
- "seconds": 1.0169636540013016,
418
- "bytes": 2368292,
419
- "pages": 29,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.3777095789992018,
424
- "bytes": 11141641,
425
- "pages": 38,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 3.434384232998127,
430
- "bytes": 5971898,
431
  "pages": 36,
432
  "shards": 2
433
  },
434
  {
435
- "seconds": 1.1304158860002644,
436
- "bytes": 8230136,
437
  "pages": 37,
438
- "shards": 3
439
  },
440
  {
441
- "seconds": 1.2011196589992323,
442
- "bytes": 6075797,
443
- "pages": 37,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 1.276424440999108,
448
- "bytes": 6070440,
449
- "pages": 25,
450
- "shards": 3
451
- },
452
- {
453
- "seconds": 1.2386367050021363,
454
- "bytes": 6009209,
455
- "pages": 36,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 1.1234037470021576,
460
- "bytes": 6172938,
461
- "pages": 37,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 0.8850959770024929,
466
- "bytes": 2438669,
467
- "pages": 31,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.6140179690009973,
472
- "bytes": 15868305,
473
- "pages": 31,
474
  "shards": 4
475
  },
476
  {
477
- "seconds": 1.348259312002483,
478
- "bytes": 5928296,
479
- "pages": 35,
480
  "shards": 2
481
  },
482
  {
483
- "seconds": 0.9825653119987692,
484
- "bytes": 2330980,
485
- "pages": 28,
486
  "shards": 2
487
  },
488
  {
489
- "seconds": 1.2888210940000135,
490
- "bytes": 6181272,
491
- "pages": 37,
492
  "shards": 2
493
  },
494
  {
495
- "seconds": 1.1400288160002674,
496
- "bytes": 6499509,
497
- "pages": 34,
498
  "shards": 3
499
  },
500
  {
501
- "seconds": 1.1078390350012342,
502
- "bytes": 6180333,
503
- "pages": 37,
504
- "shards": 2
505
  },
506
  {
507
- "seconds": 1.1313278150009864,
508
- "bytes": 6410101,
509
- "pages": 20,
510
- "shards": 2
511
  },
512
  {
513
- "seconds": 1.2548322069997084,
514
- "bytes": 7827361,
515
- "pages": 41,
516
  "shards": 3
517
  },
518
  {
519
- "seconds": 1.3341312570009904,
520
- "bytes": 10704029,
521
  "pages": 37,
522
  "shards": 3
523
  },
524
  {
525
- "seconds": 1.136130429997138,
526
- "bytes": 965107,
527
- "pages": 29,
528
- "shards": 1
529
  },
530
  {
531
- "seconds": 1.3451466709993838,
532
- "bytes": 7707115,
533
  "pages": 36,
534
  "shards": 3
535
  },
536
  {
537
- "seconds": 2.3437790619973384,
538
- "bytes": 11207805,
 
 
 
 
 
 
 
 
 
 
 
 
539
  "pages": 40,
540
- "shards": 3
541
  },
542
  {
543
- "seconds": 2.2792270009995264,
544
- "bytes": 8233418,
545
- "pages": 31,
 
 
 
 
 
 
 
 
 
 
 
 
546
  "shards": 3
547
  },
548
  {
549
- "seconds": 1.1861695589977899,
550
- "bytes": 6023568,
551
  "pages": 37,
552
  "shards": 2
553
  },
554
  {
555
- "seconds": 1.0451825370000734,
556
- "bytes": 965218,
557
- "pages": 28,
558
- "shards": 1
559
  },
560
  {
561
- "seconds": 1.2396269109995046,
562
- "bytes": 7721145,
563
- "pages": 37,
564
  "shards": 3
 
 
 
 
 
 
565
  }
566
  ],
567
  "cache_key": {
 
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": 76105.76941737354,
26
+ "required_pages_per_second_upper_estimate": 0.0052651737180082646
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": 9513.221177171692,
40
+ "required_pages_per_second_upper_estimate": 0.00398298704823338
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": 19026.442354343384,
54
+ "required_pages_per_second_upper_estimate": 0.003962706181382876
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": 28539.663531515074,
68
+ "required_pages_per_second_upper_estimate": 0.0017109092511713377
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": 28539.663531515074,
82
+ "required_pages_per_second_upper_estimate": 0.002218833306862128
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 22,
95
+ "required_token_positions_per_second": 28539.663531515074,
96
+ "required_pages_per_second_upper_estimate": 0.3660761602790507
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": 2.558590814491007,
157
  "selected": {
158
+ "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.1008662234296196,
162
+ "producer_p95_seconds": 0.2910494249990734,
163
+ "gpu_update_seconds": 0.9687570412128212,
164
+ "producer_batches_per_second": 9.914121556238891,
165
+ "gpu_batches_per_second": 1.0322505617590811,
166
+ "producer_headroom": 9.60437506504624,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.062936555194916,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 80,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 371269.35917070275,
174
+ "upload_bytes_per_second": 4902520.230367588,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 79.66576176968395,
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": 1.3777095789992018,
184
+ "bytes": 11141641,
185
+ "pages": 38,
186
+ "shards": 3
187
  },
188
  {
189
+ "seconds": 3.434384232998127,
190
+ "bytes": 5971898,
191
+ "pages": 36,
192
  "shards": 2
193
  },
194
  {
195
+ "seconds": 1.1304158860002644,
196
+ "bytes": 8230136,
197
+ "pages": 37,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 1.2011196589992323,
202
+ "bytes": 6075797,
203
+ "pages": 37,
204
+ "shards": 2
205
  },
206
  {
207
+ "seconds": 1.276424440999108,
208
+ "bytes": 6070440,
209
+ "pages": 25,
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.2386367050021363,
214
+ "bytes": 6009209,
215
+ "pages": 36,
216
+ "shards": 2
217
  },
218
  {
219
+ "seconds": 1.1234037470021576,
220
+ "bytes": 6172938,
221
  "pages": 37,
 
 
 
 
 
 
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 0.8850959770024929,
226
+ "bytes": 2438669,
227
  "pages": 31,
 
 
 
 
 
 
228
  "shards": 2
229
  },
230
  {
231
+ "seconds": 1.6140179690009973,
232
+ "bytes": 15868305,
233
+ "pages": 31,
234
+ "shards": 4
 
 
 
 
 
 
235
  },
236
  {
237
+ "seconds": 1.348259312002483,
238
+ "bytes": 5928296,
239
  "pages": 35,
240
+ "shards": 2
241
  },
242
  {
243
+ "seconds": 0.9825653119987692,
244
+ "bytes": 2330980,
245
+ "pages": 28,
246
+ "shards": 2
247
  },
248
  {
249
+ "seconds": 1.2888210940000135,
250
+ "bytes": 6181272,
251
+ "pages": 37,
252
+ "shards": 2
253
  },
254
  {
255
+ "seconds": 1.1400288160002674,
256
+ "bytes": 6499509,
257
+ "pages": 34,
258
+ "shards": 3
259
  },
260
  {
261
+ "seconds": 1.1078390350012342,
262
+ "bytes": 6180333,
263
+ "pages": 37,
264
+ "shards": 2
265
  },
266
  {
267
+ "seconds": 1.1313278150009864,
268
+ "bytes": 6410101,
269
+ "pages": 20,
270
  "shards": 2
271
  },
272
  {
273
+ "seconds": 1.2548322069997084,
274
+ "bytes": 7827361,
275
+ "pages": 41,
276
+ "shards": 3
277
  },
278
  {
279
+ "seconds": 1.3341312570009904,
280
+ "bytes": 10704029,
281
+ "pages": 37,
282
+ "shards": 3
283
  },
284
  {
285
+ "seconds": 1.136130429997138,
286
+ "bytes": 965107,
287
+ "pages": 29,
288
+ "shards": 1
289
  },
290
  {
291
+ "seconds": 1.3451466709993838,
292
+ "bytes": 7707115,
293
  "pages": 36,
294
+ "shards": 3
295
  },
296
  {
297
+ "seconds": 2.3437790619973384,
298
+ "bytes": 11207805,
299
+ "pages": 40,
300
+ "shards": 3
301
  },
302
  {
303
+ "seconds": 2.2792270009995264,
304
+ "bytes": 8233418,
305
+ "pages": 31,
306
+ "shards": 3
307
  },
308
  {
309
+ "seconds": 1.1861695589977899,
310
+ "bytes": 6023568,
311
+ "pages": 37,
312
  "shards": 2
313
  },
314
  {
315
+ "seconds": 1.0451825370000734,
316
+ "bytes": 965218,
317
+ "pages": 28,
318
+ "shards": 1
319
  },
320
  {
321
+ "seconds": 1.2396269109995046,
322
+ "bytes": 7721145,
323
  "pages": 37,
324
+ "shards": 3
 
 
 
 
 
 
325
  },
326
  {
327
+ "seconds": 1.3206068420004158,
328
+ "bytes": 11244076,
329
+ "pages": 40,
330
  "shards": 3
331
  },
332
  {
333
+ "seconds": 1.1782944210026471,
334
+ "bytes": 2347277,
335
+ "pages": 29,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 1.2341140790013014,
340
+ "bytes": 11781395,
341
+ "pages": 31,
342
+ "shards": 4
343
  },
344
  {
345
+ "seconds": 2.1994080930016935,
346
+ "bytes": 6022375,
347
+ "pages": 38,
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 1.5225848519985448,
352
+ "bytes": 6009508,
353
+ "pages": 32,
354
  "shards": 3
355
  },
356
  {
357
+ "seconds": 1.197222058999614,
358
+ "bytes": 6186205,
359
+ "pages": 37,
360
  "shards": 2
361
  },
362
  {
363
+ "seconds": 1.0562214180026785,
364
+ "bytes": 2441509,
365
+ "pages": 32,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.3156736879973323,
370
+ "bytes": 6047068,
371
+ "pages": 34,
372
+ "shards": 2
373
  },
374
  {
375
+ "seconds": 1.5381603519999771,
376
+ "bytes": 16328441,
377
+ "pages": 37,
378
+ "shards": 4
379
  },
380
  {
381
+ "seconds": 1.2096843409999565,
382
+ "bytes": 2710723,
383
+ "pages": 38,
384
+ "shards": 2
385
  },
386
  {
387
+ "seconds": 1.2958958110029926,
388
+ "bytes": 5955386,
389
+ "pages": 27,
390
  "shards": 2
391
  },
392
  {
393
+ "seconds": 1.584322863000125,
394
+ "bytes": 6087710,
395
+ "pages": 37,
396
  "shards": 2
397
  },
398
  {
399
+ "seconds": 1.2062028339969402,
400
+ "bytes": 5756694,
401
+ "pages": 35,
402
  "shards": 3
403
  },
404
  {
405
+ "seconds": 1.087463227002445,
406
+ "bytes": 6477149,
407
  "pages": 36,
408
  "shards": 2
409
  },
410
  {
411
+ "seconds": 1.7014945069968235,
412
+ "bytes": 13448331,
413
  "pages": 37,
414
+ "shards": 4
415
  },
416
  {
417
+ "seconds": 1.4291834129980998,
418
+ "bytes": 4870018,
419
+ "pages": 31,
420
  "shards": 2
421
  },
422
  {
423
+ "seconds": 1.6493652240023948,
424
+ "bytes": 6440305,
425
+ "pages": 37,
 
 
 
 
 
 
426
  "shards": 2
427
  },
428
  {
429
+ "seconds": 1.0658796659990912,
430
+ "bytes": 2378111,
431
+ "pages": 30,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.2226023729999724,
436
+ "bytes": 6024557,
437
+ "pages": 36,
438
  "shards": 2
439
  },
440
  {
441
+ "seconds": 2.320106137998664,
442
+ "bytes": 11400278,
443
+ "pages": 37,
444
  "shards": 4
445
  },
446
  {
447
+ "seconds": 1.142467518999183,
448
+ "bytes": 6818785,
449
+ "pages": 33,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.1884660100004112,
454
+ "bytes": 6414581,
455
+ "pages": 37,
456
  "shards": 2
457
  },
458
  {
459
+ "seconds": 0.931925422999484,
460
+ "bytes": 2347633,
461
+ "pages": 28,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 1.7378346169971337,
466
+ "bytes": 10782839,
467
+ "pages": 36,
468
  "shards": 3
469
  },
470
  {
471
+ "seconds": 1.3063333719983348,
472
+ "bytes": 7915290,
473
+ "pages": 39,
474
+ "shards": 3
475
  },
476
  {
477
+ "seconds": 1.0522420070010412,
478
+ "bytes": 1136593,
479
+ "pages": 36,
480
+ "shards": 1
481
  },
482
  {
483
+ "seconds": 1.2345661709987326,
484
+ "bytes": 11661576,
485
+ "pages": 28,
486
  "shards": 3
487
  },
488
  {
489
+ "seconds": 1.2733937859993603,
490
+ "bytes": 6517566,
491
  "pages": 37,
492
  "shards": 3
493
  },
494
  {
495
+ "seconds": 1.2599434170006134,
496
+ "bytes": 6272622,
497
+ "pages": 37,
498
+ "shards": 2
499
  },
500
  {
501
+ "seconds": 1.9948137850005878,
502
+ "bytes": 7735186,
503
  "pages": 36,
504
  "shards": 3
505
  },
506
  {
507
+ "seconds": 1.6032572549993347,
508
+ "bytes": 5165320,
509
+ "pages": 25,
510
+ "shards": 2
511
+ },
512
+ {
513
+ "seconds": 1.8941422179996152,
514
+ "bytes": 6019433,
515
+ "pages": 39,
516
+ "shards": 2
517
+ },
518
+ {
519
+ "seconds": 1.3825668060017051,
520
+ "bytes": 13491032,
521
  "pages": 40,
522
+ "shards": 5
523
  },
524
  {
525
+ "seconds": 0.9105340399983106,
526
+ "bytes": 1249369,
527
+ "pages": 35,
528
+ "shards": 1
529
+ },
530
+ {
531
+ "seconds": 1.295263830001204,
532
+ "bytes": 5736433,
533
+ "pages": 23,
534
+ "shards": 2
535
+ },
536
+ {
537
+ "seconds": 1.218638172998908,
538
+ "bytes": 7112834,
539
+ "pages": 43,
540
  "shards": 3
541
  },
542
  {
543
+ "seconds": 1.2346669860016846,
544
+ "bytes": 6391793,
545
  "pages": 37,
546
  "shards": 2
547
  },
548
  {
549
+ "seconds": 1.537398492997454,
550
+ "bytes": 7291338,
551
+ "pages": 30,
552
+ "shards": 3
553
  },
554
  {
555
+ "seconds": 1.5484525030005898,
556
+ "bytes": 10816659,
557
+ "pages": 38,
558
  "shards": 3
559
+ },
560
+ {
561
+ "seconds": 2.108060000999103,
562
+ "bytes": 5732876,
563
+ "pages": 29,
564
+ "shards": 2
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3c5628671e1b2c65e85ea6083662d90b0c22e2a2bf9df3cd5c38a446fd7e4fa4
3
- size 721773631
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8ba9bbe2eb54c264d2b9324ec47dc07ced3c51c8d254214772fea4aa3594fd98
3
+ size 721768447
training_config.json CHANGED
@@ -53,7 +53,7 @@
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
- "prefetch_batches": 63,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 62,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,