appvoid commited on
Commit
d130b46
·
verified ·
1 Parent(s): dc0d0e4

Qwen90M step 735000; 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 +271 -175
  6. trainer.pt +2 -2
  7. training_config.json +1 -1
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 730000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "8c53eaad59485c3e7f93abf3a4b58bce24607b53a618b83f0a58c30a6ceb0a27",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "1650174bff312eebb6355ae334a6607b73e2c555e90fe45a7a399e9567721584",
12
- "training_config.json": "75248af0a321738b405c3d51d1d55c7040dbc0785759d95f8dbdf6cd071a6687",
13
  "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379",
14
- "shard_scheduler.json": "1ca6461804592e6d06e70f93e961d9b78fbee44b41acd3825f26f6670f746dd6",
15
- "coverage.json": "a190366ff62a7a816e8e5cfff69e551cc2bb0b47d066c9abcec59ff0915c60cf",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63"
 
1
  {
2
+ "step": 735000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "a1c6719e35478e41b41395590fac06c68dd51c2c2fe17fc071f3f0d62497ad73",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "3ba531a5375e7fdc02ffb63552e7ec706eb03a955e5cb37672bc7eb52d7cc315",
12
+ "training_config.json": "a5b628e58d936dba526336e6ed1a3b16a703b7373430cc4aca66220c358ed037",
13
  "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379",
14
+ "shard_scheduler.json": "541a1f3f705d2b085cbb51195f8b00accd55d018941e78dc4445045fc4c03dcd",
15
+ "coverage.json": "fe7d1dc256a120fa7e42273b21198470f2acfeaaf06e57edd8a92aeb71326f7c",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 730000,
3
  "stats": {
4
- "climbmix/blocks": 26280000,
5
- "climbmix/targets": 43931838219,
6
- "climbmix/input_tokens": 43934622432,
7
- "ultra_qa/blocks": 9855000,
8
- "ultra_qa/targets": 16077362872,
9
- "ultra_qa/input_tokens": 16077393915,
10
- "cortex/blocks": 9855000,
11
- "cortex/targets": 374703742,
12
- "cortex/input_tokens": 15664993820,
13
- "ultra_style/blocks": 9855000,
14
- "ultra_style/targets": 15587572089,
15
- "ultra_style/input_tokens": 15588389892,
16
- "openmath/blocks": 6570000,
17
- "openmath/targets": 10152051955,
18
- "openmath/input_tokens": 10155434511,
19
- "rewrite6/blocks": 3285000,
20
- "rewrite6/targets": 5385363851,
21
- "rewrite6/input_tokens": 5391268095
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3599,
26
- "row": 1241,
27
  "start": {
28
- "file": 174,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 16,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 0,
37
- "documents": 14742745
38
  },
39
  "rewrite6": {
40
- "page": 47,
41
- "row": 2791,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 192512
47
  }
48
  },
49
- "epoch": 40,
50
- "offset": 2048,
51
- "documents": 8411463
52
  },
53
  "openmath": {
54
- "page": 2249,
55
- "row": 1624,
56
  "start": {
57
  "file": 21,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 42,
61
- "row": 254
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 9213528
67
  },
68
  "ultra_qa": {
69
- "page": 972,
70
- "row": 2405,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 5,
76
- "row": 21207
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 3983717
82
  },
83
  "ultra_style": {
84
- "page": 1484,
85
- "row": 2035,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 7,
91
- "row": 2817
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 6080499
97
  },
98
  "cortex": {
99
- "page": 191125,
100
- "row": 60,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 24464060
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 735000,
3
  "stats": {
4
+ "climbmix/blocks": 26460000,
5
+ "climbmix/targets": 44233581693,
6
+ "climbmix/input_tokens": 44236384555,
7
+ "ultra_qa/blocks": 9922500,
8
+ "ultra_qa/targets": 16187472584,
9
+ "ultra_qa/input_tokens": 16187503812,
10
+ "cortex/blocks": 9922500,
11
+ "cortex/targets": 377265616,
12
+ "cortex/input_tokens": 15772193657,
13
+ "ultra_style/blocks": 9922500,
14
+ "ultra_style/targets": 15694417562,
15
+ "ultra_style/input_tokens": 15695240794,
16
+ "openmath/blocks": 6615000,
17
+ "openmath/targets": 10221476545,
18
+ "openmath/input_tokens": 10224882264,
19
+ "rewrite6/blocks": 3307500,
20
+ "rewrite6/targets": 5422206078,
21
+ "rewrite6/input_tokens": 5428150619
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3623,
26
+ "row": 2520,
27
  "start": {
28
+ "file": 175,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 31,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 14842328
38
  },
39
  "rewrite6": {
40
+ "page": 11,
41
+ "row": 2259,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 45056
47
  }
48
  },
49
+ "epoch": 41,
50
+ "offset": 0,
51
+ "documents": 8468879
52
  },
53
  "openmath": {
54
+ "page": 2264,
55
+ "row": 3335,
56
  "start": {
57
  "file": 21,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 103,
61
+ "row": 694
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 9276679
67
  },
68
  "ultra_qa": {
69
+ "page": 979,
70
+ "row": 1035,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 5,
76
+ "row": 49879
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 4011019
82
  },
83
  "ultra_style": {
84
+ "page": 1494,
85
+ "row": 2372,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 7,
91
+ "row": 43777
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 6121796
97
  },
98
  "cortex": {
99
+ "page": 192433,
100
+ "row": 128,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 24631552
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 730000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "8c53eaad59485c3e7f93abf3a4b58bce24607b53a618b83f0a58c30a6ceb0a27", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "1650174bff312eebb6355ae334a6607b73e2c555e90fe45a7a399e9567721584", "training_config.json": "75248af0a321738b405c3d51d1d55c7040dbc0785759d95f8dbdf6cd071a6687", "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379", "shard_scheduler.json": "1ca6461804592e6d06e70f93e961d9b78fbee44b41acd3825f26f6670f746dd6", "coverage.json": "a190366ff62a7a816e8e5cfff69e551cc2bb0b47d066c9abcec59ff0915c60cf", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63", "folder": "step-000000730000"}
 
1
+ {"step": 735000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "a1c6719e35478e41b41395590fac06c68dd51c2c2fe17fc071f3f0d62497ad73", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "3ba531a5375e7fdc02ffb63552e7ec706eb03a955e5cb37672bc7eb52d7cc315", "training_config.json": "a5b628e58d936dba526336e6ed1a3b16a703b7373430cc4aca66220c358ed037", "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379", "shard_scheduler.json": "541a1f3f705d2b085cbb51195f8b00accd55d018941e78dc4445045fc4c03dcd", "coverage.json": "fe7d1dc256a120fa7e42273b21198470f2acfeaaf06e57edd8a92aeb71326f7c", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63", "folder": "step-000000735000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8c53eaad59485c3e7f93abf3a4b58bce24607b53a618b83f0a58c30a6ceb0a27
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1c6719e35478e41b41395590fac06c68dd51c2c2fe17fc071f3f0d62497ad73
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": 76313.75839753207,
26
- "required_pages_per_second_upper_estimate": 0.005279562878256555
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": 9539.21979969151,
40
- "required_pages_per_second_upper_estimate": 0.003993872128569453
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": 19078.43959938302,
54
- "required_pages_per_second_upper_estimate": 0.003973535836265066
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": 28617.659399074528,
68
- "required_pages_per_second_upper_estimate": 0.0017155849843387327
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": 28617.659399074528,
82
- "required_pages_per_second_upper_estimate": 0.002224897142497311
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": 28617.659399074528,
96
- "required_pages_per_second_upper_estimate": 0.3670766075226655
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,180 +155,30 @@
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": 0.10201518258595854,
162
- "producer_p95_seconds": 0.18670680400100537,
163
- "gpu_update_seconds": 0.9661167468117297,
164
- "producer_batches_per_second": 9.802462483046527,
165
- "gpu_batches_per_second": 1.035071592848471,
166
- "producer_headroom": 9.47032316486494,
167
  "sustainable": true,
168
- "buffer_seconds": 60.86535504913897,
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": 342045.38893318584,
174
- "upload_bytes_per_second": 4834421.459987242,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 73.39497805592234,
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": 2.1827129590001277,
184
- "bytes": 47377,
185
- "pages": 1,
186
- "shards": 1
187
- },
188
- {
189
- "seconds": 1.064991151999493,
190
- "bytes": 1003408,
191
- "pages": 31,
192
- "shards": 1
193
- },
194
- {
195
- "seconds": 1.010816134000379,
196
- "bytes": 5780671,
197
- "pages": 29,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.7082208169995283,
202
- "bytes": 12327681,
203
- "pages": 32,
204
- "shards": 4
205
- },
206
- {
207
- "seconds": 2.076565071999539,
208
- "bytes": 6113882,
209
- "pages": 36,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.1128287640003691,
214
- "bytes": 6043549,
215
- "pages": 35,
216
- "shards": 2
217
- },
218
- {
219
- "seconds": 1.0340174030006892,
220
- "bytes": 2678432,
221
- "pages": 36,
222
- "shards": 2
223
- },
224
- {
225
- "seconds": 1.1785862379992977,
226
- "bytes": 9722174,
227
- "pages": 29,
228
- "shards": 3
229
- },
230
- {
231
- "seconds": 1.4900774309999179,
232
- "bytes": 2616245,
233
- "pages": 36,
234
- "shards": 2
235
- },
236
- {
237
- "seconds": 1.2741041360004601,
238
- "bytes": 11482818,
239
- "pages": 21,
240
- "shards": 3
241
- },
242
- {
243
- "seconds": 1.076980086999356,
244
- "bytes": 1228924,
245
- "pages": 35,
246
- "shards": 1
247
- },
248
- {
249
- "seconds": 1.3698066919996563,
250
- "bytes": 11420874,
251
- "pages": 33,
252
- "shards": 4
253
- },
254
- {
255
- "seconds": 1.0434529929998462,
256
- "bytes": 6181688,
257
- "pages": 36,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.099565554000037,
262
- "bytes": 1093470,
263
- "pages": 35,
264
- "shards": 1
265
- },
266
- {
267
- "seconds": 1.3861621640007797,
268
- "bytes": 7683491,
269
- "pages": 35,
270
- "shards": 4
271
- },
272
- {
273
- "seconds": 1.2777914989992496,
274
- "bytes": 16450745,
275
- "pages": 38,
276
- "shards": 4
277
- },
278
- {
279
- "seconds": 0.9403941730006409,
280
- "bytes": 2445089,
281
- "pages": 28,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 1.042657115000111,
286
- "bytes": 6054602,
287
- "pages": 36,
288
- "shards": 2
289
- },
290
- {
291
- "seconds": 1.0668450510002003,
292
- "bytes": 1106491,
293
- "pages": 35,
294
- "shards": 1
295
- },
296
- {
297
- "seconds": 1.397956458000408,
298
- "bytes": 11076576,
299
- "pages": 34,
300
- "shards": 4
301
- },
302
- {
303
- "seconds": 1.1856196190001356,
304
- "bytes": 6180989,
305
- "pages": 35,
306
- "shards": 2
307
- },
308
- {
309
- "seconds": 1.3201266620008028,
310
- "bytes": 6761235,
311
- "pages": 30,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.2538779149999755,
316
- "bytes": 7546399,
317
- "pages": 30,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 1.2998433860002478,
322
- "bytes": 5962432,
323
- "pages": 35,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 1.2994795300000987,
328
- "bytes": 7606454,
329
- "pages": 39,
330
- "shards": 3
331
- },
332
  {
333
  "seconds": 1.54305899999963,
334
  "bytes": 5745307,
@@ -466,6 +316,252 @@
466
  "bytes": 10315993,
467
  "pages": 33,
468
  "shards": 3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
469
  }
470
  ],
471
  "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": 76078.75752915804,
26
+ "required_pages_per_second_upper_estimate": 0.0052633049729053005
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": 9509.844691144755,
40
+ "required_pages_per_second_upper_estimate": 0.003981573384042948
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": 19019.68938228951,
54
+ "required_pages_per_second_upper_estimate": 0.0039612997153919026
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": 28529.534073434268,
68
+ "required_pages_per_second_upper_estimate": 0.0017103020056261802
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": 28529.534073434268,
82
+ "required_pages_per_second_upper_estimate": 0.0022180457860511227
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": 28529.534073434268,
96
+ "required_pages_per_second_upper_estimate": 0.36594623046695485
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": 0.08382008823436138,
162
+ "producer_p95_seconds": 0.23780670700034534,
163
+ "gpu_update_seconds": 0.9691009999965221,
164
+ "producer_batches_per_second": 11.930314332335168,
165
+ "gpu_batches_per_second": 1.031884189577339,
166
+ "producer_headroom": 11.561679549738852,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.084261999784374,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 78,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 304411.78655841085,
174
+ "upload_bytes_per_second": 4856659.550780564,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 65.319683051722,
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": 1.54305899999963,
184
  "bytes": 5745307,
 
316
  "bytes": 10315993,
317
  "pages": 33,
318
  "shards": 3
319
+ },
320
+ {
321
+ "seconds": 1.4511640289983916,
322
+ "bytes": 7595757,
323
+ "pages": 38,
324
+ "shards": 3
325
+ },
326
+ {
327
+ "seconds": 1.2930020470012096,
328
+ "bytes": 547633,
329
+ "pages": 17,
330
+ "shards": 1
331
+ },
332
+ {
333
+ "seconds": 1.3445512510006665,
334
+ "bytes": 5828365,
335
+ "pages": 27,
336
+ "shards": 2
337
+ },
338
+ {
339
+ "seconds": 1.2113412039998366,
340
+ "bytes": 6839007,
341
+ "pages": 41,
342
+ "shards": 3
343
+ },
344
+ {
345
+ "seconds": 1.4825698250006099,
346
+ "bytes": 11644292,
347
+ "pages": 29,
348
+ "shards": 3
349
+ },
350
+ {
351
+ "seconds": 1.3249330049984565,
352
+ "bytes": 7496918,
353
+ "pages": 38,
354
+ "shards": 3
355
+ },
356
+ {
357
+ "seconds": 1.1885178210013692,
358
+ "bytes": 1081987,
359
+ "pages": 34,
360
+ "shards": 1
361
+ },
362
+ {
363
+ "seconds": 1.9165967070002807,
364
+ "bytes": 10316454,
365
+ "pages": 33,
366
+ "shards": 3
367
+ },
368
+ {
369
+ "seconds": 1.1245960120013478,
370
+ "bytes": 2666345,
371
+ "pages": 37,
372
+ "shards": 2
373
+ },
374
+ {
375
+ "seconds": 1.2064912770001683,
376
+ "bytes": 6004944,
377
+ "pages": 31,
378
+ "shards": 2
379
+ },
380
+ {
381
+ "seconds": 3.7895365029999084,
382
+ "bytes": 13350183,
383
+ "pages": 36,
384
+ "shards": 4
385
+ },
386
+ {
387
+ "seconds": 1.1515760519996547,
388
+ "bytes": 5491935,
389
+ "pages": 30,
390
+ "shards": 2
391
+ },
392
+ {
393
+ "seconds": 1.1848105680001026,
394
+ "bytes": 6290344,
395
+ "pages": 36,
396
+ "shards": 2
397
+ },
398
+ {
399
+ "seconds": 1.0463193560008222,
400
+ "bytes": 2621559,
401
+ "pages": 37,
402
+ "shards": 2
403
+ },
404
+ {
405
+ "seconds": 1.171350075999726,
406
+ "bytes": 6139992,
407
+ "pages": 36,
408
+ "shards": 2
409
+ },
410
+ {
411
+ "seconds": 1.280977876000179,
412
+ "bytes": 9975922,
413
+ "pages": 33,
414
+ "shards": 3
415
+ },
416
+ {
417
+ "seconds": 1.2528174399994896,
418
+ "bytes": 8102539,
419
+ "pages": 30,
420
+ "shards": 3
421
+ },
422
+ {
423
+ "seconds": 1.1965843840007437,
424
+ "bytes": 6380308,
425
+ "pages": 36,
426
+ "shards": 2
427
+ },
428
+ {
429
+ "seconds": 1.2874604959997669,
430
+ "bytes": 2530940,
431
+ "pages": 33,
432
+ "shards": 2
433
+ },
434
+ {
435
+ "seconds": 1.3693337599997903,
436
+ "bytes": 10798536,
437
+ "pages": 37,
438
+ "shards": 3
439
+ },
440
+ {
441
+ "seconds": 1.092897812000956,
442
+ "bytes": 6537422,
443
+ "pages": 28,
444
+ "shards": 2
445
+ },
446
+ {
447
+ "seconds": 0.9582660920004855,
448
+ "bytes": 2603502,
449
+ "pages": 36,
450
+ "shards": 2
451
+ },
452
+ {
453
+ "seconds": 1.5727497110001423,
454
+ "bytes": 11757823,
455
+ "pages": 31,
456
+ "shards": 3
457
+ },
458
+ {
459
+ "seconds": 1.1941701559990179,
460
+ "bytes": 5221736,
461
+ "pages": 34,
462
+ "shards": 2
463
+ },
464
+ {
465
+ "seconds": 1.6056268569991516,
466
+ "bytes": 7578967,
467
+ "pages": 37,
468
+ "shards": 3
469
+ },
470
+ {
471
+ "seconds": 1.20693444599965,
472
+ "bytes": 6453330,
473
+ "pages": 35,
474
+ "shards": 2
475
+ },
476
+ {
477
+ "seconds": 0.9651811909989192,
478
+ "bytes": 2682508,
479
+ "pages": 37,
480
+ "shards": 2
481
+ },
482
+ {
483
+ "seconds": 1.4213833419998991,
484
+ "bytes": 10754179,
485
+ "pages": 34,
486
+ "shards": 3
487
+ },
488
+ {
489
+ "seconds": 1.5129684479998105,
490
+ "bytes": 6722253,
491
+ "pages": 30,
492
+ "shards": 3
493
+ },
494
+ {
495
+ "seconds": 1.165229547999843,
496
+ "bytes": 7984438,
497
+ "pages": 38,
498
+ "shards": 3
499
+ },
500
+ {
501
+ "seconds": 1.001051560999258,
502
+ "bytes": 939443,
503
+ "pages": 30,
504
+ "shards": 1
505
+ },
506
+ {
507
+ "seconds": 1.4753997340012575,
508
+ "bytes": 11016252,
509
+ "pages": 34,
510
+ "shards": 4
511
+ },
512
+ {
513
+ "seconds": 1.6974678700007644,
514
+ "bytes": 6266676,
515
+ "pages": 36,
516
+ "shards": 2
517
+ },
518
+ {
519
+ "seconds": 1.6492898739998054,
520
+ "bytes": 1094803,
521
+ "pages": 34,
522
+ "shards": 1
523
+ },
524
+ {
525
+ "seconds": 1.8326161909990333,
526
+ "bytes": 13393499,
527
+ "pages": 34,
528
+ "shards": 4
529
+ },
530
+ {
531
+ "seconds": 1.4041516550005326,
532
+ "bytes": 11034703,
533
+ "pages": 37,
534
+ "shards": 3
535
+ },
536
+ {
537
+ "seconds": 1.1081106500005262,
538
+ "bytes": 1056276,
539
+ "pages": 28,
540
+ "shards": 1
541
+ },
542
+ {
543
+ "seconds": 1.211192590999417,
544
+ "bytes": 7370118,
545
+ "pages": 35,
546
+ "shards": 3
547
+ },
548
+ {
549
+ "seconds": 1.246054737999657,
550
+ "bytes": 1155172,
551
+ "pages": 34,
552
+ "shards": 1
553
+ },
554
+ {
555
+ "seconds": 1.6413428979994933,
556
+ "bytes": 11205821,
557
+ "pages": 33,
558
+ "shards": 4
559
+ },
560
+ {
561
+ "seconds": 1.3393069079993438,
562
+ "bytes": 11931034,
563
+ "pages": 38,
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 sha256:1650174bff312eebb6355ae334a6607b73e2c555e90fe45a7a399e9567721584
3
- size 721792959
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ba531a5375e7fdc02ffb63552e7ec706eb03a955e5cb37672bc7eb52d7cc315
3
+ size 721775167
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,