appvoid commited on
Commit
955d764
·
verified ·
1 Parent(s): d0b9d61

Qwen90M step 335000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +47 -47
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +240 -240
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 330000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "328bd745ba3eedb0c09045707f1c3645741429105fa758b46400ed4e42af0708",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "15e5868d1d2d9bba0011346b8b8fe6a7017e790885cbae0ddebf73fbc62c7dfe",
12
- "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a",
14
- "shard_scheduler.json": "9ea0c54003a1a89d00999afcc33c19cd9e20fd18d66970efa7421ab898bf1d77",
15
- "coverage.json": "8db65ea8720123ccbfed7d388c2266922b6f26d48d6d3b735a4714722ac5561c",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 335000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "7438707d5944ffca46843ff585ead722c9fa7bdc8dab736da1510f4bff5c4701",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "5f4a15472e9622ce649a08d61cb56adda487f4944d44a36aff803668640d8309",
12
+ "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a",
14
+ "shard_scheduler.json": "c1a7db519fe396e51051231707a3d26cd5328da0fb81bb9da77c36f5e680779e",
15
+ "coverage.json": "dddfaa0a3732c9d4616f58805f4c39815efab74bccdd566963924cd8fcbd4985",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 330000,
3
  "stats": {
4
- "climbmix/blocks": 11880000,
5
- "climbmix/targets": 19858829729,
6
- "climbmix/input_tokens": 19860088763,
7
- "ultra_qa/blocks": 4455000,
8
- "ultra_qa/targets": 7267913693,
9
- "ultra_qa/input_tokens": 7267927725,
10
- "cortex/blocks": 4455000,
11
- "cortex/targets": 169500703,
12
- "cortex/input_tokens": 7082136293,
13
- "ultra_style/blocks": 4455000,
14
- "ultra_style/targets": 7046969672,
15
- "ultra_style/input_tokens": 7047339282,
16
- "openmath/blocks": 2970000,
17
- "openmath/targets": 4589449827,
18
- "openmath/input_tokens": 4590979616,
19
- "rewrite6/blocks": 1485000,
20
- "rewrite6/targets": 2434435841,
21
- "rewrite6/input_tokens": 2437105213
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1627,
26
- "row": 1818,
27
  "start": {
28
- "file": 78,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 62,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 6666010
38
  },
39
  "rewrite6": {
40
- "page": 25,
41
- "row": 3161,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 102400
47
  }
48
  },
49
  "epoch": 18,
50
  "offset": 0,
51
- "documents": 3802833
52
  },
53
  "openmath": {
54
- "page": 1016,
55
- "row": 4063,
56
  "start": {
57
  "file": 9,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 231,
61
- "row": 686
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 4165599
67
  },
68
  "ultra_qa": {
69
- "page": 439,
70
- "row": 2714,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 3,
76
- "row": 45392
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 4096,
81
- "documents": 1800858
82
  },
83
  "ultra_style": {
84
- "page": 671,
85
- "row": 421,
86
  "start": {
87
  "file": 4,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 0,
91
- "row": 8645
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 2048,
96
- "documents": 2748837
97
  },
98
  "cortex": {
99
- "page": 86407,
100
- "row": 78,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 11060174
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 335000,
3
  "stats": {
4
+ "climbmix/blocks": 12060000,
5
+ "climbmix/targets": 20160288396,
6
+ "climbmix/input_tokens": 20161566115,
7
+ "ultra_qa/blocks": 4522500,
8
+ "ultra_qa/targets": 7378025318,
9
+ "ultra_qa/input_tokens": 7378039564,
10
+ "cortex/blocks": 4522500,
11
+ "cortex/targets": 172070651,
12
+ "cortex/input_tokens": 7189400225,
13
+ "ultra_style/blocks": 4522500,
14
+ "ultra_style/targets": 7153491246,
15
+ "ultra_style/input_tokens": 7153866702,
16
+ "openmath/blocks": 3015000,
17
+ "openmath/targets": 4658887005,
18
+ "openmath/input_tokens": 4660439825,
19
+ "rewrite6/blocks": 1507500,
20
+ "rewrite6/targets": 2471326187,
21
+ "rewrite6/input_tokens": 2474035818
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1651,
26
+ "row": 3325,
27
  "start": {
28
+ "file": 79,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 75,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 6765821
38
  },
39
  "rewrite6": {
40
+ "page": 39,
41
+ "row": 3247,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 159744
47
  }
48
  },
49
  "epoch": 18,
50
  "offset": 0,
51
+ "documents": 3860263
52
  },
53
  "openmath": {
54
+ "page": 1032,
55
+ "row": 1485,
56
  "start": {
57
  "file": 9,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 297,
61
+ "row": 222
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 4228557
67
  },
68
  "ultra_qa": {
69
+ "page": 446,
70
+ "row": 1301,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 4,
76
+ "row": 9563
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 2048,
81
+ "documents": 1828117
82
  },
83
  "ultra_style": {
84
+ "page": 681,
85
+ "row": 1612,
86
  "start": {
87
  "file": 4,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 0,
91
+ "row": 49605
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 0,
96
+ "documents": 2790988
97
  },
98
  "cortex": {
99
+ "page": 87716,
100
+ "row": 66,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 11227714
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 330000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "328bd745ba3eedb0c09045707f1c3645741429105fa758b46400ed4e42af0708", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "15e5868d1d2d9bba0011346b8b8fe6a7017e790885cbae0ddebf73fbc62c7dfe", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a", "shard_scheduler.json": "9ea0c54003a1a89d00999afcc33c19cd9e20fd18d66970efa7421ab898bf1d77", "coverage.json": "8db65ea8720123ccbfed7d388c2266922b6f26d48d6d3b735a4714722ac5561c", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000330000"}
 
1
+ {"step": 335000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "7438707d5944ffca46843ff585ead722c9fa7bdc8dab736da1510f4bff5c4701", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "5f4a15472e9622ce649a08d61cb56adda487f4944d44a36aff803668640d8309", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a", "shard_scheduler.json": "c1a7db519fe396e51051231707a3d26cd5328da0fb81bb9da77c36f5e680779e", "coverage.json": "dddfaa0a3732c9d4616f58805f4c39815efab74bccdd566963924cd8fcbd4985", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000335000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:328bd745ba3eedb0c09045707f1c3645741429105fa758b46400ed4e42af0708
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7438707d5944ffca46843ff585ead722c9fa7bdc8dab736da1510f4bff5c4701
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": 75630.60254063262,
26
- "required_pages_per_second_upper_estimate": 0.006555468944212357
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": 9453.825317579078,
40
- "required_pages_per_second_upper_estimate": 0.003771272266466841
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": 18907.650635158156,
54
- "required_pages_per_second_upper_estimate": 0.004373952211157938
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": 28361.475952737233,
68
- "required_pages_per_second_upper_estimate": 0.0017270650843573335
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": 28361.475952737233,
82
- "required_pages_per_second_upper_estimate": 0.002183344928431987
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": 28361.475952737233,
96
- "required_pages_per_second_upper_estimate": 0.3521546115789914
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": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 690670,
161
- "producer_batch_seconds": 0.0792378665312583,
162
- "producer_p95_seconds": 0.2171078139999736,
163
- "gpu_update_seconds": 0.9748434829722473,
164
- "producer_batches_per_second": 12.620228733764721,
165
- "gpu_batches_per_second": 1.0258056985220354,
166
- "producer_headroom": 12.302747734729635,
167
  "sustainable": true,
168
- "buffer_seconds": 60.440295944279335,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 83,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 523269.04561499885,
174
- "upload_bytes_per_second": 4017522.9884006456,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 112.28135610902285,
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": 1.6009747379994224,
184
- "bytes": 7396060,
185
  "pages": 37,
186
  "shards": 3
187
  },
188
  {
189
- "seconds": 0.9060076790001403,
190
- "bytes": 1026778,
191
- "pages": 33,
192
- "shards": 1
193
  },
194
  {
195
- "seconds": 1.30779888699999,
196
- "bytes": 10925777,
197
- "pages": 35,
198
  "shards": 3
199
  },
200
  {
201
- "seconds": 0.9117075719987042,
202
- "bytes": 2359249,
203
- "pages": 28,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 0.9851995050012192,
208
- "bytes": 5930447,
209
- "pages": 34,
210
  "shards": 2
211
  },
212
  {
213
- "seconds": 1.2710061699999642,
214
- "bytes": 13577169,
215
- "pages": 42,
216
- "shards": 4
217
- },
218
- {
219
- "seconds": 0.8450371549988631,
220
- "bytes": 895470,
221
- "pages": 25,
222
- "shards": 1
223
  },
224
  {
225
- "seconds": 32.774606008000774,
226
- "bytes": 10411787,
227
- "pages": 35,
228
- "shards": 3
229
  },
230
  {
231
- "seconds": 1.2439632269997674,
232
- "bytes": 7901436,
233
- "pages": 46,
234
- "shards": 3
235
  },
236
  {
237
- "seconds": 0.8556265790011821,
238
- "bytes": 1138175,
239
- "pages": 33,
240
- "shards": 1
241
  },
242
  {
243
- "seconds": 1.1632403840012557,
244
- "bytes": 11705717,
245
- "pages": 30,
246
- "shards": 3
247
  },
248
  {
249
- "seconds": 1.197896118999779,
250
- "bytes": 12261236,
251
- "pages": 40,
252
  "shards": 4
253
  },
254
  {
255
- "seconds": 0.8173304430001735,
256
- "bytes": 1136030,
257
- "pages": 33,
258
  "shards": 1
259
  },
260
  {
261
- "seconds": 1.1651587139986077,
262
- "bytes": 5801842,
263
- "pages": 20,
264
- "shards": 2
265
  },
266
  {
267
- "seconds": 1.0791701340003783,
268
- "bytes": 2256184,
269
- "pages": 26,
270
  "shards": 2
271
  },
272
  {
273
- "seconds": 1.5194503539987636,
274
- "bytes": 9946744,
 
 
 
 
 
 
275
  "pages": 38,
276
  "shards": 3
277
  },
278
  {
279
- "seconds": 1.188283532999776,
280
- "bytes": 2401103,
281
- "pages": 29,
282
- "shards": 2
283
  },
284
  {
285
- "seconds": 1.182281362998765,
286
- "bytes": 11896636,
287
- "pages": 38,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 0.8353424750002887,
292
- "bytes": 975468,
293
- "pages": 29,
294
  "shards": 1
295
  },
296
  {
297
- "seconds": 1.1824829600009252,
298
- "bytes": 12495213,
299
- "pages": 37,
300
  "shards": 4
301
  },
302
  {
303
- "seconds": 1.254413848000695,
304
- "bytes": 6063292,
305
- "pages": 35,
306
  "shards": 3
307
  },
308
  {
309
- "seconds": 0.9384880159996101,
310
- "bytes": 1115458,
311
- "pages": 34,
312
  "shards": 1
313
  },
314
  {
315
- "seconds": 1.1859958829991228,
316
- "bytes": 7375397,
317
- "pages": 29,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 0.9071520359993883,
322
- "bytes": 6539230,
323
- "pages": 25,
324
- "shards": 2
325
  },
326
  {
327
- "seconds": 1.2508507479997206,
328
- "bytes": 11653692,
329
- "pages": 45,
330
- "shards": 4
331
  },
332
  {
333
- "seconds": 1.2195578610007942,
334
- "bytes": 5944204,
335
- "pages": 27,
336
  "shards": 2
337
  },
338
  {
339
- "seconds": 0.9418964010001218,
340
- "bytes": 1018742,
341
- "pages": 33,
342
  "shards": 1
343
  },
344
  {
345
- "seconds": 1.3071213839994016,
346
- "bytes": 12190710,
347
- "pages": 40,
348
  "shards": 4
349
  },
350
  {
351
- "seconds": 0.9488432640009705,
352
- "bytes": 921326,
353
- "pages": 29,
354
- "shards": 1
355
  },
356
  {
357
- "seconds": 1.0550517599986051,
358
- "bytes": 5962735,
359
- "pages": 26,
360
  "shards": 2
361
  },
362
  {
363
- "seconds": 1.1898453599987988,
364
- "bytes": 13419202,
365
- "pages": 43,
366
- "shards": 4
367
  },
368
  {
369
- "seconds": 0.9201347599992005,
370
- "bytes": 838883,
371
- "pages": 26,
372
  "shards": 1
373
  },
374
  {
375
- "seconds": 1.1256958809990465,
376
- "bytes": 12103363,
377
- "pages": 38,
378
  "shards": 4
379
  },
380
  {
381
- "seconds": 1.0281697059999715,
382
- "bytes": 1191797,
383
  "pages": 34,
384
- "shards": 1
385
  },
386
  {
387
- "seconds": 0.9877725879996433,
388
- "bytes": 6235947,
389
  "pages": 36,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 1.2409586350004247,
394
- "bytes": 7163170,
395
- "pages": 28,
396
  "shards": 3
397
  },
398
  {
399
- "seconds": 1.270394237999426,
400
- "bytes": 10514630,
401
- "pages": 29,
402
- "shards": 3
403
  },
404
  {
405
- "seconds": 1.509260009999707,
406
- "bytes": 6276691,
407
  "pages": 36,
 
 
 
 
 
 
408
  "shards": 2
409
  },
410
  {
411
- "seconds": 1.1396728859999712,
412
- "bytes": 2416731,
413
- "pages": 27,
414
  "shards": 2
415
  },
416
  {
417
- "seconds": 0.9309691420003219,
418
- "bytes": 6123756,
419
- "pages": 33,
 
 
 
 
 
 
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.1473543519987288,
424
- "bytes": 5736316,
425
- "pages": 34,
426
  "shards": 2
427
  },
428
  {
429
- "seconds": 1.0078334160007216,
430
- "bytes": 7473663,
431
- "pages": 37,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 1.0918037890005508,
436
- "bytes": 11651730,
437
- "pages": 37,
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 1.0898228620008013,
442
- "bytes": 2320314,
443
- "pages": 28,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 0.9809586890005448,
448
- "bytes": 9894024,
449
- "pages": 31,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 1.0132985079999344,
454
- "bytes": 6275693,
455
- "pages": 36,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 0.9494441940005345,
460
- "bytes": 2648505,
461
  "pages": 35,
462
- "shards": 2
463
  },
464
  {
465
- "seconds": 1.0081078959992738,
466
- "bytes": 5874669,
467
- "pages": 36,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.0266657669999404,
472
- "bytes": 4655742,
473
- "pages": 12,
474
- "shards": 2
475
  },
476
  {
477
- "seconds": 1.4716465549990971,
478
- "bytes": 13803834,
479
- "pages": 48,
480
  "shards": 4
481
  },
482
  {
483
- "seconds": 0.940834745999382,
484
- "bytes": 6400703,
485
- "pages": 35,
486
- "shards": 2
487
- },
488
- {
489
- "seconds": 1.1989260670015938,
490
- "bytes": 2439937,
491
  "pages": 28,
492
- "shards": 2
493
- },
494
- {
495
- "seconds": 1.2518158220009354,
496
- "bytes": 10474054,
497
- "pages": 38,
498
- "shards": 4
499
  },
500
  {
501
- "seconds": 1.1628889589992468,
502
- "bytes": 1077101,
503
  "pages": 34,
504
  "shards": 1
505
  },
506
  {
507
- "seconds": 1.117781933000515,
508
- "bytes": 7831651,
509
- "pages": 22,
510
  "shards": 3
511
  },
512
  {
513
- "seconds": 1.0544361239990394,
514
- "bytes": 7130727,
515
- "pages": 44,
516
  "shards": 2
517
  },
518
  {
519
- "seconds": 1.1150904570004059,
520
- "bytes": 10599538,
521
- "pages": 29,
522
- "shards": 3
523
  },
524
  {
525
- "seconds": 1.0015506020008615,
526
- "bytes": 7818449,
527
- "pages": 38,
528
- "shards": 3
529
  },
530
  {
531
- "seconds": 0.8705007850003312,
532
- "bytes": 1070987,
533
  "pages": 34,
534
  "shards": 1
535
  },
536
  {
537
- "seconds": 0.9655942159988626,
538
- "bytes": 7298211,
539
  "pages": 29,
540
  "shards": 3
541
  },
542
  {
543
- "seconds": 0.8577407669999957,
544
- "bytes": 1078743,
 
 
 
 
 
 
545
  "pages": 30,
546
  "shards": 1
547
  },
548
  {
549
- "seconds": 1.3106396379989746,
550
- "bytes": 15458880,
551
- "pages": 39,
552
- "shards": 4
553
  },
554
  {
555
- "seconds": 0.9607973950005544,
556
- "bytes": 7512102,
557
- "pages": 38,
558
- "shards": 3
559
  },
560
  {
561
- "seconds": 2.3653303009996307,
562
- "bytes": 1160564,
563
- "pages": 33,
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": 73863.57629496043,
26
+ "required_pages_per_second_upper_estimate": 0.006402307587724566
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": 9232.947036870053,
40
+ "required_pages_per_second_upper_estimate": 0.0036831606178674217
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": 18465.894073740106,
54
+ "required_pages_per_second_upper_estimate": 0.004271759605323816
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": 27698.84111061016,
68
+ "required_pages_per_second_upper_estimate": 0.0016867140990481273
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": 27698.84111061016,
82
+ "required_pages_per_second_upper_estimate": 0.0021323334639944016
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": 27698.84111061016,
96
+ "required_pages_per_second_upper_estimate": 0.34392690453592956
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": 690670,
161
+ "producer_batch_seconds": 0.11786020207830461,
162
+ "producer_p95_seconds": 0.29016289100036374,
163
+ "gpu_update_seconds": 0.9981645040524572,
164
+ "producer_batches_per_second": 8.48462824911512,
165
+ "gpu_batches_per_second": 1.0018388711881567,
166
+ "producer_headroom": 8.469054748347462,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.88803474719989,
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": 418892.7533661353,
174
+ "upload_bytes_per_second": 5731055.085245443,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 89.88463354814571,
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": 1.0918037890005508,
184
+ "bytes": 11651730,
185
  "pages": 37,
186
  "shards": 3
187
  },
188
  {
189
+ "seconds": 1.0898228620008013,
190
+ "bytes": 2320314,
191
+ "pages": 28,
192
+ "shards": 2
193
  },
194
  {
195
+ "seconds": 0.9809586890005448,
196
+ "bytes": 9894024,
197
+ "pages": 31,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 1.0132985079999344,
202
+ "bytes": 6275693,
203
+ "pages": 36,
204
  "shards": 2
205
  },
206
  {
207
+ "seconds": 0.9494441940005345,
208
+ "bytes": 2648505,
209
+ "pages": 35,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.0081078959992738,
214
+ "bytes": 5874669,
215
+ "pages": 36,
216
+ "shards": 2
 
 
 
 
 
 
217
  },
218
  {
219
+ "seconds": 1.0266657669999404,
220
+ "bytes": 4655742,
221
+ "pages": 12,
222
+ "shards": 2
223
  },
224
  {
225
+ "seconds": 1.4716465549990971,
226
+ "bytes": 13803834,
227
+ "pages": 48,
228
+ "shards": 4
229
  },
230
  {
231
+ "seconds": 0.940834745999382,
232
+ "bytes": 6400703,
233
+ "pages": 35,
234
+ "shards": 2
235
  },
236
  {
237
+ "seconds": 1.1989260670015938,
238
+ "bytes": 2439937,
239
+ "pages": 28,
240
+ "shards": 2
241
  },
242
  {
243
+ "seconds": 1.2518158220009354,
244
+ "bytes": 10474054,
245
+ "pages": 38,
246
  "shards": 4
247
  },
248
  {
249
+ "seconds": 1.1628889589992468,
250
+ "bytes": 1077101,
251
+ "pages": 34,
252
  "shards": 1
253
  },
254
  {
255
+ "seconds": 1.117781933000515,
256
+ "bytes": 7831651,
257
+ "pages": 22,
258
+ "shards": 3
259
  },
260
  {
261
+ "seconds": 1.0544361239990394,
262
+ "bytes": 7130727,
263
+ "pages": 44,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.1150904570004059,
268
+ "bytes": 10599538,
269
+ "pages": 29,
270
+ "shards": 3
271
+ },
272
+ {
273
+ "seconds": 1.0015506020008615,
274
+ "bytes": 7818449,
275
  "pages": 38,
276
  "shards": 3
277
  },
278
  {
279
+ "seconds": 0.8705007850003312,
280
+ "bytes": 1070987,
281
+ "pages": 34,
282
+ "shards": 1
283
  },
284
  {
285
+ "seconds": 0.9655942159988626,
286
+ "bytes": 7298211,
287
+ "pages": 29,
288
  "shards": 3
289
  },
290
  {
291
+ "seconds": 0.8577407669999957,
292
+ "bytes": 1078743,
293
+ "pages": 30,
294
  "shards": 1
295
  },
296
  {
297
+ "seconds": 1.3106396379989746,
298
+ "bytes": 15458880,
299
+ "pages": 39,
300
  "shards": 4
301
  },
302
  {
303
+ "seconds": 0.9607973950005544,
304
+ "bytes": 7512102,
305
+ "pages": 38,
306
  "shards": 3
307
  },
308
  {
309
+ "seconds": 2.3653303009996307,
310
+ "bytes": 1160564,
311
+ "pages": 33,
312
  "shards": 1
313
  },
314
  {
315
+ "seconds": 1.3003170710017002,
316
+ "bytes": 11354768,
317
+ "pages": 32,
318
+ "shards": 4
 
 
 
 
 
 
319
  },
320
  {
321
+ "seconds": 0.953995298999871,
322
+ "bytes": 345798,
323
+ "pages": 9,
324
+ "shards": 1
325
  },
326
  {
327
+ "seconds": 1.1301041130009253,
328
+ "bytes": 5919931,
329
+ "pages": 35,
330
  "shards": 2
331
  },
332
  {
333
+ "seconds": 0.7547573850006302,
334
+ "bytes": 662301,
335
+ "pages": 19,
336
  "shards": 1
337
  },
338
  {
339
+ "seconds": 1.1137079630007065,
340
+ "bytes": 13478878,
341
+ "pages": 47,
342
  "shards": 4
343
  },
344
  {
345
+ "seconds": 1.1504248440014635,
346
+ "bytes": 10949848,
347
+ "pages": 32,
348
+ "shards": 3
349
  },
350
  {
351
+ "seconds": 0.9190237900002103,
352
+ "bytes": 2687904,
353
+ "pages": 36,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 0.9390472680024686,
358
+ "bytes": 6039556,
359
+ "pages": 27,
360
+ "shards": 2
361
  },
362
  {
363
+ "seconds": 1.1019230559977586,
364
+ "bytes": 1147220,
365
+ "pages": 34,
366
  "shards": 1
367
  },
368
  {
369
+ "seconds": 1.2733968520005874,
370
+ "bytes": 11422716,
371
+ "pages": 32,
372
  "shards": 4
373
  },
374
  {
375
+ "seconds": 1.048131932002434,
376
+ "bytes": 6792716,
377
  "pages": 34,
378
+ "shards": 2
379
  },
380
  {
381
+ "seconds": 0.9567362410016358,
382
+ "bytes": 6277518,
383
  "pages": 36,
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 1.2580463369995414,
388
+ "bytes": 7425150,
389
+ "pages": 37,
390
  "shards": 3
391
  },
392
  {
393
+ "seconds": 1.0858683609985746,
394
+ "bytes": 5956157,
395
+ "pages": 32,
396
+ "shards": 2
397
  },
398
  {
399
+ "seconds": 1.2466707620005764,
400
+ "bytes": 7844213,
401
  "pages": 36,
402
+ "shards": 3
403
+ },
404
+ {
405
+ "seconds": 1.0721679589987616,
406
+ "bytes": 5937253,
407
+ "pages": 29,
408
  "shards": 2
409
  },
410
  {
411
+ "seconds": 0.9170985109994945,
412
+ "bytes": 6599086,
413
+ "pages": 28,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 1.249970442000631,
418
+ "bytes": 7464066,
419
+ "pages": 32,
420
+ "shards": 3
421
+ },
422
+ {
423
+ "seconds": 1.004503175001446,
424
+ "bytes": 5154959,
425
+ "pages": 38,
426
  "shards": 2
427
  },
428
  {
429
+ "seconds": 1.2185796390003816,
430
+ "bytes": 6487750,
431
+ "pages": 35,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.3924951020017033,
436
+ "bytes": 2562340,
437
+ "pages": 31,
438
  "shards": 3
439
  },
440
  {
441
+ "seconds": 1.429218111999944,
442
+ "bytes": 10705946,
443
+ "pages": 39,
444
  "shards": 3
445
  },
446
  {
447
+ "seconds": 1.0223479320011393,
448
+ "bytes": 6489223,
449
+ "pages": 24,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.180141814002127,
454
+ "bytes": 7534736,
455
+ "pages": 35,
456
  "shards": 3
457
  },
458
  {
459
+ "seconds": 1.0304640750000544,
460
+ "bytes": 6585841,
461
+ "pages": 35,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 0.9793735439998272,
466
+ "bytes": 7348514,
467
  "pages": 35,
468
+ "shards": 3
469
  },
470
  {
471
+ "seconds": 1.1231975430018792,
472
+ "bytes": 5800198,
473
+ "pages": 27,
474
  "shards": 2
475
  },
476
  {
477
+ "seconds": 0.8728474010022182,
478
+ "bytes": 1156879,
479
+ "pages": 34,
480
+ "shards": 1
481
  },
482
  {
483
+ "seconds": 1.1430372309987433,
484
+ "bytes": 13489384,
485
+ "pages": 40,
486
  "shards": 4
487
  },
488
  {
489
+ "seconds": 1.410483870000462,
490
+ "bytes": 10063648,
 
 
 
 
 
 
491
  "pages": 28,
492
+ "shards": 3
 
 
 
 
 
 
493
  },
494
  {
495
+ "seconds": 1.0145048790000146,
496
+ "bytes": 1134517,
497
  "pages": 34,
498
  "shards": 1
499
  },
500
  {
501
+ "seconds": 3.9685035570000764,
502
+ "bytes": 7473998,
503
+ "pages": 38,
504
  "shards": 3
505
  },
506
  {
507
+ "seconds": 1.2958799169973645,
508
+ "bytes": 6217387,
509
+ "pages": 35,
510
  "shards": 2
511
  },
512
  {
513
+ "seconds": 0.9571121369990578,
514
+ "bytes": 4079115,
515
+ "pages": 20,
516
+ "shards": 2
517
  },
518
  {
519
+ "seconds": 1.3280163199997332,
520
+ "bytes": 13612044,
521
+ "pages": 47,
522
+ "shards": 4
523
  },
524
  {
525
+ "seconds": 0.827956348999578,
526
+ "bytes": 1165763,
527
  "pages": 34,
528
  "shards": 1
529
  },
530
  {
531
+ "seconds": 1.215623644999141,
532
+ "bytes": 7281440,
533
  "pages": 29,
534
  "shards": 3
535
  },
536
  {
537
+ "seconds": 1.1079499599982228,
538
+ "bytes": 10281570,
539
+ "pages": 37,
540
+ "shards": 3
541
+ },
542
+ {
543
+ "seconds": 0.9046422840001469,
544
+ "bytes": 1080144,
545
  "pages": 30,
546
  "shards": 1
547
  },
548
  {
549
+ "seconds": 1.0989511400002812,
550
+ "bytes": 7379943,
551
+ "pages": 33,
552
+ "shards": 3
553
  },
554
  {
555
+ "seconds": 1.0861735590005992,
556
+ "bytes": 6754175,
557
+ "pages": 31,
558
+ "shards": 2
559
  },
560
  {
561
+ "seconds": 1.2741842929972336,
562
+ "bytes": 12412930,
563
+ "pages": 39,
564
+ "shards": 4
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:15e5868d1d2d9bba0011346b8b8fe6a7017e790885cbae0ddebf73fbc62c7dfe
3
- size 721778495
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f4a15472e9622ce649a08d61cb56adda487f4944d44a36aff803668640d8309
3
+ size 721759039
training_config.json CHANGED
@@ -46,8 +46,8 @@
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,
 
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,