appvoid commited on
Commit
1f21150
·
verified ·
1 Parent(s): 90aac50

Qwen90M step 190000; 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 +246 -246
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 185000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "6c683f5fcca897a05f7e0948afb2a85c2f5f99d3a2df8fe3ad69849852360942",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "267a2d850846babea151e8cb265757c9351aed7191d41054613399dd488e4d56",
12
- "training_config.json": "65390fe09f12d0188f44175607f7a9d1665082a136656274cb5d054eb998e80b",
13
  "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08",
14
- "shard_scheduler.json": "fc6e7dce18328a420ccbbadf0eec3f835f85696a4e1a04ae07e86a4342d1ba5e",
15
- "coverage.json": "f74a28e021edc403e0a2077b92693ffc8ada4dd44851adc7cd4be18fc29269e7",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 190000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "5b29d7dbcb4b6dc66cbd141f4370adef39d9c5bf0d95a0cb23dbf0b3fbf77d71",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "003d5c36c105e74cd2f29e7718b5727a4a7950ad08dd430bb87b3b0378f6d788",
12
+ "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08",
14
+ "shard_scheduler.json": "6e7020fbb929d08abf5c36bb9ec83b6165c6ce2fb599b8042b1e42ee010831ac",
15
+ "coverage.json": "379675965a16132527a2a833bac7de4bb49e2af467d73478c93a42753679aff0",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 185000,
3
  "stats": {
4
- "climbmix/blocks": 6660000,
5
- "climbmix/targets": 11131092534,
6
- "climbmix/input_tokens": 11131798123,
7
- "ultra_qa/blocks": 2497500,
8
- "ultra_qa/targets": 4074695441,
9
- "ultra_qa/input_tokens": 4074703291,
10
- "cortex/blocks": 2497500,
11
- "cortex/targets": 95006899,
12
- "cortex/input_tokens": 3969702317,
13
- "ultra_style/blocks": 2497500,
14
- "ultra_style/targets": 3950298536,
15
- "ultra_style/input_tokens": 3950506046,
16
- "openmath/blocks": 1665000,
17
- "openmath/targets": 2572479334,
18
- "openmath/input_tokens": 2573336685,
19
- "rewrite6/blocks": 832500,
20
- "rewrite6/targets": 1364722714,
21
- "rewrite6/input_tokens": 1366219165
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 912,
26
- "row": 2733,
27
  "start": {
28
- "file": 44,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 7,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 6144,
37
- "documents": 3738285
38
  },
39
  "rewrite6": {
40
- "page": 18,
41
- "row": 4079,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 73728
47
  }
48
  },
49
  "epoch": 10,
50
  "offset": 0,
51
- "documents": 2131847
52
  },
53
  "openmath": {
54
- "page": 570,
55
- "row": 592,
56
  "start": {
57
  "file": 5,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 151,
61
- "row": 470
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 2048,
66
- "documents": 2335312
67
  },
68
  "ultra_qa": {
69
- "page": 246,
70
- "row": 1981,
71
  "start": {
72
  "file": 1,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 7,
76
- "row": 36139
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 1009597
82
  },
83
  "ultra_style": {
84
- "page": 376,
85
- "row": 1396,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 1,
91
- "row": 84695
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 1541492
97
  },
98
  "cortex": {
99
- "page": 48431,
100
- "row": 115,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 6199283
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 190000,
3
  "stats": {
4
+ "climbmix/blocks": 6840000,
5
+ "climbmix/targets": 11431842491,
6
+ "climbmix/input_tokens": 11432567043,
7
+ "ultra_qa/blocks": 2565000,
8
+ "ultra_qa/targets": 4184969180,
9
+ "ultra_qa/input_tokens": 4184977239,
10
+ "cortex/blocks": 2565000,
11
+ "cortex/targets": 97573355,
12
+ "cortex/input_tokens": 4076931537,
13
+ "ultra_style/blocks": 2565000,
14
+ "ultra_style/targets": 4057172458,
15
+ "ultra_style/input_tokens": 4057385373,
16
+ "openmath/blocks": 1710000,
17
+ "openmath/targets": 2642162908,
18
+ "openmath/input_tokens": 2643043776,
19
+ "rewrite6/blocks": 855000,
20
+ "rewrite6/targets": 1401575108,
21
+ "rewrite6/input_tokens": 1403112022
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 937,
26
+ "row": 1183,
27
  "start": {
28
+ "file": 45,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 24,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 3839135
38
  },
39
  "rewrite6": {
40
+ "page": 33,
41
+ "row": 359,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 135168
47
  }
48
  },
49
  "epoch": 10,
50
  "offset": 0,
51
+ "documents": 2189567
52
  },
53
  "openmath": {
54
+ "page": 585,
55
+ "row": 2658,
56
  "start": {
57
  "file": 5,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 212,
61
+ "row": 910
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 0,
66
+ "documents": 2398818
67
  },
68
  "ultra_qa": {
69
+ "page": 253,
70
+ "row": 592,
71
  "start": {
72
  "file": 1,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 8,
76
+ "row": 410
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 1036880
82
  },
83
  "ultra_style": {
84
+ "page": 386,
85
+ "row": 1768,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 2,
91
+ "row": 40788
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 1582824
97
  },
98
  "cortex": {
99
+ "page": 49741,
100
+ "row": 11,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 6366859
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 185000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "6c683f5fcca897a05f7e0948afb2a85c2f5f99d3a2df8fe3ad69849852360942", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "267a2d850846babea151e8cb265757c9351aed7191d41054613399dd488e4d56", "training_config.json": "65390fe09f12d0188f44175607f7a9d1665082a136656274cb5d054eb998e80b", "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08", "shard_scheduler.json": "fc6e7dce18328a420ccbbadf0eec3f835f85696a4e1a04ae07e86a4342d1ba5e", "coverage.json": "f74a28e021edc403e0a2077b92693ffc8ada4dd44851adc7cd4be18fc29269e7", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000185000"}
 
1
+ {"step": 190000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "5b29d7dbcb4b6dc66cbd141f4370adef39d9c5bf0d95a0cb23dbf0b3fbf77d71", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "003d5c36c105e74cd2f29e7718b5727a4a7950ad08dd430bb87b3b0378f6d788", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08", "shard_scheduler.json": "6e7020fbb929d08abf5c36bb9ec83b6165c6ce2fb599b8042b1e42ee010831ac", "coverage.json": "379675965a16132527a2a833bac7de4bb49e2af467d73478c93a42753679aff0", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000190000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6c683f5fcca897a05f7e0948afb2a85c2f5f99d3a2df8fe3ad69849852360942
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b29d7dbcb4b6dc66cbd141f4370adef39d9c5bf0d95a0cb23dbf0b3fbf77d71
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": 76375.43340313055,
26
- "required_pages_per_second_upper_estimate": 0.006620028995617115
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": 9546.929175391318,
40
- "required_pages_per_second_upper_estimate": 0.0038084127873748677
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": 19093.858350782637,
54
- "required_pages_per_second_upper_estimate": 0.004417028089023795
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": 28640.787526173957,
68
- "required_pages_per_second_upper_estimate": 0.0017440736937450589
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": 28640.787526173957,
82
- "required_pages_per_second_upper_estimate": 0.002204847106539067
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": 22,
95
- "required_token_positions_per_second": 28640.787526173957,
96
- "required_pages_per_second_upper_estimate": 0.3556227265253729
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": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
- "producer_batch_seconds": 0.5055529695468408,
162
- "producer_p95_seconds": 0.33400089199858485,
163
- "gpu_update_seconds": 0.9653365842239262,
164
- "producer_batches_per_second": 1.9780320960162956,
165
- "gpu_batches_per_second": 1.0359081136492316,
166
- "producer_headroom": 1.9094667470536641,
167
  "sustainable": true,
168
- "buffer_seconds": 60.81620480610735,
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": 283448.8065840012,
174
- "upload_bytes_per_second": 4514531.554258477,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 60.821515542412065,
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": 1.2709916069979954,
184
- "bytes": 5895393,
185
- "pages": 35,
186
  "shards": 2
187
  },
188
  {
189
- "seconds": 1.2639917810010957,
190
- "bytes": 13243059,
191
- "pages": 34,
 
 
 
 
 
 
 
 
 
 
 
 
192
  "shards": 4
193
  },
194
  {
195
- "seconds": 1.182673250001244,
196
- "bytes": 5984869,
197
- "pages": 35,
198
  "shards": 2
199
  },
200
  {
201
- "seconds": 0.7554075879997981,
202
- "bytes": 949451,
203
- "pages": 27,
204
- "shards": 1
205
  },
206
  {
207
- "seconds": 1.1886153279992868,
208
- "bytes": 11793828,
209
- "pages": 35,
210
- "shards": 4
211
  },
212
  {
213
- "seconds": 1.1566395750014635,
214
- "bytes": 1136088,
215
- "pages": 34,
216
- "shards": 1
217
  },
218
  {
219
- "seconds": 1.3760789849984576,
220
- "bytes": 7774763,
221
- "pages": 37,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 1.7174801740002295,
226
- "bytes": 11887929,
227
- "pages": 34,
228
  "shards": 3
229
  },
230
  {
231
- "seconds": 1.0596165809984086,
232
- "bytes": 5348616,
233
- "pages": 34,
234
- "shards": 2
235
  },
236
  {
237
- "seconds": 1.3328053339973849,
238
- "bytes": 7544008,
239
- "pages": 30,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 0.952769712999725,
244
- "bytes": 1151127,
245
- "pages": 34,
246
  "shards": 1
247
  },
248
  {
249
- "seconds": 1.7508555999993405,
250
- "bytes": 8239169,
251
- "pages": 39,
252
  "shards": 3
253
  },
254
  {
255
- "seconds": 1.4154862130017136,
256
- "bytes": 9920178,
257
- "pages": 28,
258
- "shards": 3
259
  },
260
  {
261
- "seconds": 1.3325980400004482,
262
- "bytes": 6856977,
263
- "pages": 36,
264
- "shards": 2
265
  },
266
  {
267
- "seconds": 1.1581080529977044,
268
- "bytes": 7348060,
269
- "pages": 32,
270
- "shards": 3
271
  },
272
  {
273
- "seconds": 1.0286378179989697,
274
- "bytes": 1051160,
275
- "pages": 32,
276
  "shards": 1
277
  },
278
  {
279
- "seconds": 1.0477113340020878,
280
- "bytes": 9989334,
281
- "pages": 36,
282
  "shards": 3
283
  },
284
  {
285
- "seconds": 1.1226200919991243,
286
- "bytes": 7462663,
287
- "pages": 38,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 0.8068668020023324,
292
- "bytes": 982732,
293
  "pages": 26,
294
  "shards": 1
295
  },
296
  {
297
- "seconds": 1.4669019489992934,
298
- "bytes": 13788589,
299
- "pages": 39,
300
  "shards": 4
301
  },
302
  {
303
- "seconds": 1.348856757998874,
304
- "bytes": 5345336,
305
- "pages": 30,
 
 
 
 
 
 
306
  "shards": 3
307
  },
308
  {
309
- "seconds": 1.6173429520022182,
310
- "bytes": 6229619,
 
 
 
 
 
 
311
  "pages": 36,
312
  "shards": 2
313
  },
314
  {
315
- "seconds": 1.705037297000672,
316
- "bytes": 7666167,
317
- "pages": 37,
318
- "shards": 3
319
  },
320
  {
321
- "seconds": 0.7579026999992493,
322
- "bytes": 928826,
323
- "pages": 27,
324
- "shards": 1
325
  },
326
  {
327
- "seconds": 1.3881743410020135,
328
- "bytes": 10608461,
329
- "pages": 29,
330
- "shards": 3
331
  },
332
  {
333
- "seconds": 1.1401566910026304,
334
- "bytes": 8280901,
335
- "pages": 33,
336
  "shards": 3
337
  },
338
  {
339
- "seconds": 1.1491751960020338,
340
- "bytes": 6275126,
341
- "pages": 35,
342
  "shards": 2
343
  },
344
  {
345
- "seconds": 1.3218376940021699,
346
- "bytes": 6023489,
347
- "pages": 34,
348
  "shards": 2
349
  },
350
  {
351
- "seconds": 1.0548232209985144,
352
- "bytes": 2410136,
353
  "pages": 28,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.1507087749996572,
358
- "bytes": 10108802,
359
- "pages": 40,
360
- "shards": 3
361
- },
362
- {
363
- "seconds": 0.8633549010010029,
364
- "bytes": 2675452,
365
- "pages": 37,
366
  "shards": 2
367
  },
368
  {
369
- "seconds": 1.1101803599995037,
370
- "bytes": 11781038,
371
- "pages": 33,
372
- "shards": 3
373
  },
374
  {
375
- "seconds": 1.124476634999155,
376
- "bytes": 9801537,
377
- "pages": 30,
378
- "shards": 3
379
  },
380
  {
381
- "seconds": 0.9195323169988114,
382
- "bytes": 2651824,
383
- "pages": 37,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 1.357327001998783,
388
- "bytes": 5924306,
389
- "pages": 29,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 0.9212070710018452,
394
- "bytes": 2573619,
395
- "pages": 34,
396
  "shards": 2
397
  },
398
  {
399
- "seconds": 1.2234259870019741,
400
- "bytes": 10476491,
401
- "pages": 39,
402
  "shards": 3
403
  },
404
  {
405
- "seconds": 1.209228235999035,
406
- "bytes": 11665620,
407
- "pages": 34,
408
  "shards": 3
409
  },
410
  {
411
- "seconds": 0.8683807099987462,
412
- "bytes": 2266346,
413
- "pages": 28,
414
- "shards": 2
415
- },
416
- {
417
- "seconds": 0.9792534570005955,
418
- "bytes": 6085076,
419
  "pages": 36,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.2787357830020483,
424
- "bytes": 10476048,
425
  "pages": 31,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 1.3773436970004695,
430
- "bytes": 2708180,
431
- "pages": 37,
432
- "shards": 2
433
  },
434
  {
435
- "seconds": 1.157878524998523,
436
- "bytes": 6108198,
437
- "pages": 36,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 1.1510198380019574,
442
- "bytes": 2279083,
443
- "pages": 28,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 19.352378830000816,
448
- "bytes": 15868354,
449
- "pages": 40,
450
  "shards": 4
451
  },
452
  {
453
- "seconds": 1.1481948339969676,
454
- "bytes": 6679366,
455
- "pages": 36,
456
- "shards": 2
457
  },
458
  {
459
- "seconds": 1.3883708200010005,
460
- "bytes": 2398547,
461
- "pages": 33,
462
- "shards": 2
463
  },
464
  {
465
- "seconds": 1.0771264450013405,
466
- "bytes": 6009796,
467
- "pages": 32,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 0.8690516380011104,
472
- "bytes": 4841742,
473
  "pages": 29,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.4456373529974371,
478
- "bytes": 7578448,
479
- "pages": 35,
480
- "shards": 3
481
- },
482
- {
483
- "seconds": 1.2644371399983356,
484
- "bytes": 12142497,
485
- "pages": 38,
486
- "shards": 3
487
- },
488
- {
489
- "seconds": 0.8376236600015545,
490
- "bytes": 2410849,
491
- "pages": 28,
492
  "shards": 3
493
  },
494
  {
495
- "seconds": 1.585001501996885,
496
- "bytes": 10678100,
497
- "pages": 41,
498
- "shards": 3
499
  },
500
  {
501
- "seconds": 0.8894862470006046,
502
- "bytes": 957902,
503
- "pages": 28,
504
  "shards": 1
505
  },
506
  {
507
- "seconds": 1.1445572449993051,
508
- "bytes": 7733205,
509
- "pages": 37,
510
  "shards": 3
511
  },
512
  {
513
- "seconds": 1.1386810379990493,
514
- "bytes": 6058898,
515
- "pages": 29,
516
  "shards": 2
517
  },
518
  {
519
- "seconds": 1.2615420420006558,
520
- "bytes": 12864855,
521
- "pages": 34,
522
- "shards": 4
523
  },
524
  {
525
- "seconds": 0.9415042880027613,
526
- "bytes": 5881567,
527
- "pages": 28,
528
  "shards": 2
529
  },
530
  {
531
- "seconds": 0.8555704430000333,
532
- "bytes": 1226801,
533
- "pages": 35,
534
- "shards": 1
535
  },
536
  {
537
- "seconds": 0.896007103001466,
538
- "bytes": 7569788,
539
- "pages": 38,
540
- "shards": 3
541
  },
542
  {
543
- "seconds": 1.1828385400003754,
544
- "bytes": 9643641,
545
- "pages": 34,
546
- "shards": 3
547
  },
548
  {
549
- "seconds": 2.2667923209992296,
550
- "bytes": 913952,
551
- "pages": 26,
552
- "shards": 1
553
  },
554
  {
555
- "seconds": 1.163448106002761,
556
- "bytes": 13378237,
557
- "pages": 41,
558
- "shards": 4
559
  },
560
  {
561
- "seconds": 0.9504098300021724,
562
- "bytes": 828896,
563
- "pages": 27,
564
- "shards": 1
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": 74817.5457905918,
26
+ "required_pages_per_second_upper_estimate": 0.006484995245792312
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": 9352.193223823975,
40
+ "required_pages_per_second_upper_estimate": 0.003730729704732717
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": 18704.38644764795,
54
+ "required_pages_per_second_upper_estimate": 0.0043269306187049715
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": 28056.579671471925,
68
+ "required_pages_per_second_upper_estimate": 0.0017084985004954352
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": 28056.579671471925,
82
+ "required_pages_per_second_upper_estimate": 0.002159873168693258
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": 21,
95
+ "required_token_positions_per_second": 28056.579671471925,
96
+ "required_pages_per_second_upper_estimate": 0.3483688201878879
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": 61,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
+ "producer_batch_seconds": 0.15711286308589933,
162
+ "producer_p95_seconds": 0.3503623950018664,
163
+ "gpu_update_seconds": 0.9854372957696134,
164
+ "producer_batches_per_second": 6.364851230884027,
165
+ "gpu_batches_per_second": 1.0147779105711778,
166
+ "producer_headroom": 6.272161784938251,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.11167504194641,
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": 374399.8448116255,
174
+ "upload_bytes_per_second": 4221736.650489915,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 80.33749111424993,
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": 1.3773436970004695,
184
+ "bytes": 2708180,
185
+ "pages": 37,
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 1.157878524998523,
190
+ "bytes": 6108198,
191
+ "pages": 36,
192
+ "shards": 2
193
+ },
194
+ {
195
+ "seconds": 1.1510198380019574,
196
+ "bytes": 2279083,
197
+ "pages": 28,
198
+ "shards": 2
199
+ },
200
+ {
201
+ "seconds": 19.352378830000816,
202
+ "bytes": 15868354,
203
+ "pages": 40,
204
  "shards": 4
205
  },
206
  {
207
+ "seconds": 1.1481948339969676,
208
+ "bytes": 6679366,
209
+ "pages": 36,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.3883708200010005,
214
+ "bytes": 2398547,
215
+ "pages": 33,
216
+ "shards": 2
217
  },
218
  {
219
+ "seconds": 1.0771264450013405,
220
+ "bytes": 6009796,
221
+ "pages": 32,
222
+ "shards": 2
223
  },
224
  {
225
+ "seconds": 0.8690516380011104,
226
+ "bytes": 4841742,
227
+ "pages": 29,
228
+ "shards": 2
229
  },
230
  {
231
+ "seconds": 1.4456373529974371,
232
+ "bytes": 7578448,
233
+ "pages": 35,
234
  "shards": 3
235
  },
236
  {
237
+ "seconds": 1.2644371399983356,
238
+ "bytes": 12142497,
239
+ "pages": 38,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 0.8376236600015545,
244
+ "bytes": 2410849,
245
+ "pages": 28,
246
+ "shards": 3
247
  },
248
  {
249
+ "seconds": 1.585001501996885,
250
+ "bytes": 10678100,
251
+ "pages": 41,
252
  "shards": 3
253
  },
254
  {
255
+ "seconds": 0.8894862470006046,
256
+ "bytes": 957902,
257
+ "pages": 28,
258
  "shards": 1
259
  },
260
  {
261
+ "seconds": 1.1445572449993051,
262
+ "bytes": 7733205,
263
+ "pages": 37,
264
  "shards": 3
265
  },
266
  {
267
+ "seconds": 1.1386810379990493,
268
+ "bytes": 6058898,
269
+ "pages": 29,
270
+ "shards": 2
271
  },
272
  {
273
+ "seconds": 1.2615420420006558,
274
+ "bytes": 12864855,
275
+ "pages": 34,
276
+ "shards": 4
277
  },
278
  {
279
+ "seconds": 0.9415042880027613,
280
+ "bytes": 5881567,
281
+ "pages": 28,
282
+ "shards": 2
283
  },
284
  {
285
+ "seconds": 0.8555704430000333,
286
+ "bytes": 1226801,
287
+ "pages": 35,
288
  "shards": 1
289
  },
290
  {
291
+ "seconds": 0.896007103001466,
292
+ "bytes": 7569788,
293
+ "pages": 38,
294
  "shards": 3
295
  },
296
  {
297
+ "seconds": 1.1828385400003754,
298
+ "bytes": 9643641,
299
+ "pages": 34,
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 2.2667923209992296,
304
+ "bytes": 913952,
305
  "pages": 26,
306
  "shards": 1
307
  },
308
  {
309
+ "seconds": 1.163448106002761,
310
+ "bytes": 13378237,
311
+ "pages": 41,
312
  "shards": 4
313
  },
314
  {
315
+ "seconds": 0.9504098300021724,
316
+ "bytes": 828896,
317
+ "pages": 27,
318
+ "shards": 1
319
+ },
320
+ {
321
+ "seconds": 1.1922220630003721,
322
+ "bytes": 10349010,
323
+ "pages": 34,
324
  "shards": 3
325
  },
326
  {
327
+ "seconds": 0.9792584309980157,
328
+ "bytes": 2231089,
329
+ "pages": 24,
330
+ "shards": 2
331
+ },
332
+ {
333
+ "seconds": 1.3213498320001236,
334
+ "bytes": 5953362,
335
  "pages": 36,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 0.9344346500001848,
340
+ "bytes": 5832415,
341
+ "pages": 28,
342
+ "shards": 2
343
  },
344
  {
345
+ "seconds": 1.198589940999227,
346
+ "bytes": 12036533,
347
+ "pages": 42,
348
+ "shards": 4
349
  },
350
  {
351
+ "seconds": 0.8354435640030715,
352
+ "bytes": 5772648,
353
+ "pages": 28,
354
+ "shards": 2
355
  },
356
  {
357
+ "seconds": 1.2891920199981541,
358
+ "bytes": 7891753,
359
+ "pages": 40,
360
  "shards": 3
361
  },
362
  {
363
+ "seconds": 0.9012639180000406,
364
+ "bytes": 5714275,
365
+ "pages": 30,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.098721825001121,
370
+ "bytes": 5875494,
371
+ "pages": 37,
372
  "shards": 2
373
  },
374
  {
375
+ "seconds": 1.1078113340008713,
376
+ "bytes": 2283092,
377
  "pages": 28,
378
  "shards": 2
379
  },
380
  {
381
+ "seconds": 0.9954401819995837,
382
+ "bytes": 6010708,
383
+ "pages": 29,
 
 
 
 
 
 
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 1.1992983269992692,
388
+ "bytes": 13571269,
389
+ "pages": 40,
390
+ "shards": 4
391
  },
392
  {
393
+ "seconds": 1.1989007740012312,
394
+ "bytes": 5658429,
395
+ "pages": 39,
396
+ "shards": 2
397
  },
398
  {
399
+ "seconds": 1.222567180000624,
400
+ "bytes": 5708293,
401
+ "pages": 27,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 1.0141177349978534,
406
+ "bytes": 2660846,
407
+ "pages": 38,
408
  "shards": 2
409
  },
410
  {
411
+ "seconds": 1.5187353149995033,
412
+ "bytes": 5879969,
413
+ "pages": 35,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 1.1908320559996355,
418
+ "bytes": 10104452,
419
+ "pages": 27,
420
  "shards": 3
421
  },
422
  {
423
+ "seconds": 1.954254078998929,
424
+ "bytes": 8371700,
425
+ "pages": 38,
426
  "shards": 3
427
  },
428
  {
429
+ "seconds": 1.207307631000731,
430
+ "bytes": 6036390,
 
 
 
 
 
 
431
  "pages": 36,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.257034783000563,
436
+ "bytes": 2415905,
437
  "pages": 31,
438
  "shards": 3
439
  },
440
  {
441
+ "seconds": 1.2379260390007403,
442
+ "bytes": 10087351,
443
+ "pages": 39,
444
+ "shards": 3
445
  },
446
  {
447
+ "seconds": 1.0402059740008553,
448
+ "bytes": 6293747,
449
+ "pages": 28,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.1847501439988264,
454
+ "bytes": 2572744,
455
+ "pages": 33,
456
  "shards": 2
457
  },
458
  {
459
+ "seconds": 1.268939194000268,
460
+ "bytes": 16767391,
461
+ "pages": 42,
462
  "shards": 4
463
  },
464
  {
465
+ "seconds": 1.2854500899993582,
466
+ "bytes": 7334524,
467
+ "pages": 31,
468
+ "shards": 3
469
  },
470
  {
471
+ "seconds": 1.0401804770008312,
472
+ "bytes": 1254870,
473
+ "pages": 34,
474
+ "shards": 1
475
  },
476
  {
477
+ "seconds": 1.428139841002121,
478
+ "bytes": 6304931,
479
+ "pages": 37,
480
  "shards": 2
481
  },
482
  {
483
+ "seconds": 0.987766936996195,
484
+ "bytes": 2378061,
485
  "pages": 29,
486
  "shards": 2
487
  },
488
  {
489
+ "seconds": 1.0273188179999124,
490
+ "bytes": 9379394,
491
+ "pages": 33,
 
 
 
 
 
 
 
 
 
 
 
 
492
  "shards": 3
493
  },
494
  {
495
+ "seconds": 1.4798621720037772,
496
+ "bytes": 13649971,
497
+ "pages": 37,
498
+ "shards": 4
499
  },
500
  {
501
+ "seconds": 0.8202600639997399,
502
+ "bytes": 1256174,
503
+ "pages": 35,
504
  "shards": 1
505
  },
506
  {
507
+ "seconds": 1.0791088629994192,
508
+ "bytes": 10296860,
509
+ "pages": 35,
510
  "shards": 3
511
  },
512
  {
513
+ "seconds": 1.2926067090011202,
514
+ "bytes": 2391669,
515
+ "pages": 32,
516
  "shards": 2
517
  },
518
  {
519
+ "seconds": 1.033323986994219,
520
+ "bytes": 6001182,
521
+ "pages": 33,
522
+ "shards": 2
523
  },
524
  {
525
+ "seconds": 3.9172589509980753,
526
+ "bytes": 5890320,
527
+ "pages": 32,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.3426188599987654,
532
+ "bytes": 12956908,
533
+ "pages": 37,
534
+ "shards": 4
535
  },
536
  {
537
+ "seconds": 1.2346246180022717,
538
+ "bytes": 6315095,
539
+ "pages": 36,
540
+ "shards": 2
541
  },
542
  {
543
+ "seconds": 3.684220068003924,
544
+ "bytes": 2396656,
545
+ "pages": 29,
546
+ "shards": 2
547
  },
548
  {
549
+ "seconds": 1.1850955770059954,
550
+ "bytes": 6553717,
551
+ "pages": 37,
552
+ "shards": 2
553
  },
554
  {
555
+ "seconds": 1.714512132006348,
556
+ "bytes": 6167274,
557
+ "pages": 33,
558
+ "shards": 2
559
  },
560
  {
561
+ "seconds": 1.2202410780009814,
562
+ "bytes": 6552255,
563
+ "pages": 33,
564
+ "shards": 3
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:267a2d850846babea151e8cb265757c9351aed7191d41054613399dd488e4d56
3
- size 721772031
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:003d5c36c105e74cd2f29e7718b5727a4a7950ad08dd430bb87b3b0378f6d788
3
+ size 721770879
training_config.json CHANGED
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 63,
50
- "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,
 
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,