appvoid commited on
Commit
86f9632
·
verified ·
1 Parent(s): 8e223cf

Qwen90M step 320000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 315000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "7bad7905e71cbf5282827daa6e33886fd079a2b0c210ba5a1021c2c4c8ef5b70",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "0becabd5ddaf3503de95d1c46e0a2cab7a3ad153cc954b1a5c27591cd4e7fba4",
12
- "training_config.json": "65390fe09f12d0188f44175607f7a9d1665082a136656274cb5d054eb998e80b",
13
- "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba",
14
- "shard_scheduler.json": "2059ce5f880963ea177af3f92919853c35636a63fc6fe06b4f84a8f420e442a2",
15
- "coverage.json": "1a679a788576b42e3b6606d45a64495b0d61f4444204e639605bcbbd2ffce4ee",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 320000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "7af0a447c456f6760d4f6e84edc350c5e69416f8c3a7a61824547f910318f805",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "38c2f886ccf49b55cd34e215a6a6c1aa8127d46138bf70445d4ca70032231d6c",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
+ "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a",
14
+ "shard_scheduler.json": "d5343011e97e001a66563c4b9f78ab277480b08424f02858b0f8f0948b4c217f",
15
+ "coverage.json": "1f8b82802b7d23b4ac70b19b7e71ccea4fd98cb1ced9c8e431325fe81b59768d",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 315000,
3
  "stats": {
4
- "climbmix/blocks": 11340000,
5
- "climbmix/targets": 18956161933,
6
- "climbmix/input_tokens": 18957363575,
7
- "ultra_qa/blocks": 4252500,
8
- "ultra_qa/targets": 6937463581,
9
- "ultra_qa/input_tokens": 6937476970,
10
- "cortex/blocks": 4252500,
11
- "cortex/targets": 161788999,
12
- "cortex/input_tokens": 6760036876,
13
- "ultra_style/blocks": 4252500,
14
- "ultra_style/targets": 6726387227,
15
- "ultra_style/input_tokens": 6726740294,
16
- "openmath/blocks": 2835000,
17
- "openmath/targets": 4380942289,
18
- "openmath/input_tokens": 4382402714,
19
- "rewrite6/blocks": 1417500,
20
- "rewrite6/targets": 2323732855,
21
- "rewrite6/input_tokens": 2326280728
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1553,
26
- "row": 1737,
27
  "start": {
28
- "file": 75,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 13,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 18432,
37
- "documents": 6362825
38
  },
39
  "rewrite6": {
40
- "page": 33,
41
- "row": 2804,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 135168
47
  }
48
  },
49
  "epoch": 17,
50
- "offset": 0,
51
- "documents": 3629840
52
  },
53
  "openmath": {
54
- "page": 970,
55
- "row": 3337,
56
  "start": {
57
  "file": 9,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 43,
61
- "row": 270
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 2048,
66
- "documents": 3976457
67
  },
68
  "ultra_qa": {
69
- "page": 419,
70
- "row": 2676,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 2,
76
- "row": 28083
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 1718900
82
  },
83
  "ultra_style": {
84
- "page": 640,
85
- "row": 3189,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 6,
91
- "row": 56507
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 2624629
97
  },
98
  "cortex": {
99
- "page": 82476,
100
- "row": 126,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 10557054
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 320000,
3
  "stats": {
4
+ "climbmix/blocks": 11520000,
5
+ "climbmix/targets": 19257096294,
6
+ "climbmix/input_tokens": 19258316982,
7
+ "ultra_qa/blocks": 4320000,
8
+ "ultra_qa/targets": 7047544993,
9
+ "ultra_qa/input_tokens": 7047558585,
10
+ "cortex/blocks": 4320000,
11
+ "cortex/targets": 164372097,
12
+ "cortex/input_tokens": 6867485637,
13
+ "ultra_style/blocks": 4320000,
14
+ "ultra_style/targets": 6833023501,
15
+ "ultra_style/input_tokens": 6833381920,
16
+ "openmath/blocks": 2880000,
17
+ "openmath/targets": 4450371997,
18
+ "openmath/input_tokens": 4451855399,
19
+ "rewrite6/blocks": 1440000,
20
+ "rewrite6/targets": 2360725625,
21
+ "rewrite6/input_tokens": 2363313839
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1578,
26
+ "row": 267,
27
  "start": {
28
+ "file": 76,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 31,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 6463755
38
  },
39
  "rewrite6": {
40
+ "page": 47,
41
+ "row": 2882,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 192512
47
  }
48
  },
49
  "epoch": 17,
50
+ "offset": 2048,
51
+ "documents": 3687262
52
  },
53
  "openmath": {
54
+ "page": 986,
55
+ "row": 621,
56
  "start": {
57
  "file": 9,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 108,
61
+ "row": 806
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 0,
66
+ "documents": 4039277
67
  },
68
  "ultra_qa": {
69
+ "page": 426,
70
+ "row": 1436,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 2,
76
+ "row": 56755
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 1746332
82
  },
83
  "ultra_style": {
84
+ "page": 650,
85
+ "row": 3626,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 7,
91
+ "row": 10908
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 2666026
97
  },
98
  "cortex": {
99
+ "page": 83788,
100
+ "row": 117,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 10724981
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 315000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "7bad7905e71cbf5282827daa6e33886fd079a2b0c210ba5a1021c2c4c8ef5b70", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "0becabd5ddaf3503de95d1c46e0a2cab7a3ad153cc954b1a5c27591cd4e7fba4", "training_config.json": "65390fe09f12d0188f44175607f7a9d1665082a136656274cb5d054eb998e80b", "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba", "shard_scheduler.json": "2059ce5f880963ea177af3f92919853c35636a63fc6fe06b4f84a8f420e442a2", "coverage.json": "1a679a788576b42e3b6606d45a64495b0d61f4444204e639605bcbbd2ffce4ee", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000315000"}
 
1
+ {"step": 320000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "7af0a447c456f6760d4f6e84edc350c5e69416f8c3a7a61824547f910318f805", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "38c2f886ccf49b55cd34e215a6a6c1aa8127d46138bf70445d4ca70032231d6c", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a", "shard_scheduler.json": "d5343011e97e001a66563c4b9f78ab277480b08424f02858b0f8f0948b4c217f", "coverage.json": "1f8b82802b7d23b4ac70b19b7e71ccea4fd98cb1ced9c8e431325fe81b59768d", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000320000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7bad7905e71cbf5282827daa6e33886fd079a2b0c210ba5a1021c2c4c8ef5b70
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7af0a447c456f6760d4f6e84edc350c5e69416f8c3a7a61824547f910318f805
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.8979115860001912,
53
- "peak_gb": 85.39889430999756,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.953845084999557,
56
- "input_tokens_per_second": 196152.9428354681
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9004576890000067,
53
+ "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9568634569999972,
56
+ "input_tokens_per_second": 195598.30756245417
57
  },
58
  "measurements": [
59
  {
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": 76228.00974719698,
26
- "required_pages_per_second_upper_estimate": 0.006607250686762633
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": 9528.501218399622,
40
- "required_pages_per_second_upper_estimate": 0.00380106159980837
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": 19057.002436799245,
54
- "required_pages_per_second_upper_estimate": 0.004408502121965669
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": 28585.503655198867,
68
- "required_pages_per_second_upper_estimate": 0.0017407071960547304
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": 28585.503655198867,
82
- "required_pages_per_second_upper_estimate": 0.002200591201115855
83
  },
84
  {
85
  "source": "cortex",
@@ -92,8 +92,8 @@
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": 28585.503655198867,
96
- "required_pages_per_second_upper_estimate": 0.35493628587107623
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,413 +155,269 @@
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": 690670,
161
- "producer_batch_seconds": 0.11422511410142988,
162
- "producer_p95_seconds": 0.3618045629991684,
163
- "gpu_update_seconds": 0.9672035285259575,
164
- "producer_batches_per_second": 8.754642163123757,
165
- "gpu_batches_per_second": 1.033908552343709,
166
- "producer_headroom": 8.467520791155419,
167
  "sustainable": true,
168
- "buffer_seconds": 60.93382229713532,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 85,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 354783.7727048669,
174
- "upload_bytes_per_second": 5691532.768220302,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 76.12833867892746,
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": 1.11581381400174,
184
- "bytes": 5950902,
185
- "pages": 37,
186
- "shards": 2
187
  },
188
  {
189
- "seconds": 0.9459429260023171,
190
- "bytes": 2335478,
191
- "pages": 29,
192
- "shards": 2
193
  },
194
  {
195
- "seconds": 1.2669872590049636,
196
- "bytes": 5719604,
197
- "pages": 21,
198
- "shards": 2
199
  },
200
  {
201
- "seconds": 1.032783564995043,
202
- "bytes": 5149031,
203
- "pages": 45,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 1.4498847849972663,
208
- "bytes": 7603712,
209
- "pages": 36,
210
- "shards": 3
211
- },
212
- {
213
- "seconds": 1.3791402839997318,
214
- "bytes": 11555501,
215
- "pages": 30,
216
- "shards": 3
217
  },
218
  {
219
- "seconds": 1.030656152004667,
220
- "bytes": 6775920,
221
- "pages": 30,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 1.0988766239970573,
226
- "bytes": 6218344,
227
- "pages": 37,
228
  "shards": 2
229
  },
230
  {
231
- "seconds": 1.1409971860048245,
232
- "bytes": 830736,
233
- "pages": 27,
234
- "shards": 1
235
  },
236
  {
237
- "seconds": 1.1426053430041065,
238
- "bytes": 7670085,
239
- "pages": 37,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 1.1845013959973585,
244
- "bytes": 10539391,
245
- "pages": 31,
246
  "shards": 3
247
  },
248
  {
249
- "seconds": 0.8672450950034545,
250
- "bytes": 606235,
251
- "pages": 21,
252
  "shards": 1
253
  },
254
  {
255
- "seconds": 1.3488649450009689,
256
- "bytes": 13749450,
257
- "pages": 49,
258
- "shards": 4
259
  },
260
  {
261
- "seconds": 0.8790644170003361,
262
- "bytes": 1116619,
263
  "pages": 35,
264
- "shards": 1
265
- },
266
- {
267
- "seconds": 1.4015686020065914,
268
- "bytes": 11114007,
269
- "pages": 33,
270
- "shards": 4
271
- },
272
- {
273
- "seconds": 1.0096355339992442,
274
- "bytes": 5988067,
275
- "pages": 29,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.234346370001731,
280
- "bytes": 974463,
281
- "pages": 34,
282
- "shards": 1
283
- },
284
- {
285
- "seconds": 1.1329207589951693,
286
- "bytes": 7046164,
287
- "pages": 25,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 1.0467858490010258,
292
- "bytes": 11558244,
293
- "pages": 41,
294
- "shards": 4
295
- },
296
- {
297
- "seconds": 1.1082996040058788,
298
- "bytes": 6095964,
299
- "pages": 36,
300
- "shards": 2
301
- },
302
- {
303
- "seconds": 1.0661074119998375,
304
- "bytes": 7367457,
305
- "pages": 30,
306
  "shards": 3
307
  },
308
  {
309
- "seconds": 0.7073749000046519,
310
- "bytes": 1147032,
311
- "pages": 34,
312
  "shards": 1
313
  },
314
  {
315
- "seconds": 2.1967378000044846,
316
- "bytes": 12230231,
317
- "pages": 35,
318
  "shards": 4
319
  },
320
  {
321
- "seconds": 1.34678302899556,
322
- "bytes": 6766636,
323
- "pages": 30,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 0.8585810450022109,
328
- "bytes": 306974,
329
- "pages": 9,
330
  "shards": 1
331
  },
332
  {
333
- "seconds": 1.175691934004135,
334
- "bytes": 6239046,
335
- "pages": 37,
336
- "shards": 2
337
- },
338
- {
339
- "seconds": 1.251433448996977,
340
- "bytes": 7613150,
341
- "pages": 31,
342
  "shards": 3
343
  },
344
  {
345
- "seconds": 0.8646319059989764,
346
- "bytes": 4940868,
347
- "pages": 32,
348
- "shards": 2
349
- },
350
- {
351
- "seconds": 0.9655562699990696,
352
- "bytes": 7648817,
353
- "pages": 39,
354
  "shards": 3
355
  },
356
  {
357
- "seconds": 1.2715992000012193,
358
- "bytes": 6184683,
359
- "pages": 36,
360
- "shards": 2
361
- },
362
- {
363
- "seconds": 0.9518272349960171,
364
- "bytes": 6809793,
365
- "pages": 31,
366
- "shards": 2
367
- },
368
- {
369
- "seconds": 1.123792185993807,
370
- "bytes": 11983615,
371
  "pages": 37,
372
- "shards": 4
373
- },
374
- {
375
- "seconds": 1.1926247609953862,
376
- "bytes": 798107,
377
- "pages": 26,
378
- "shards": 1
379
- },
380
- {
381
- "seconds": 0.9836393130026408,
382
- "bytes": 5980946,
383
- "pages": 38,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 1.1501939880035934,
388
- "bytes": 7275318,
389
  "pages": 33,
390
  "shards": 3
391
  },
392
  {
393
- "seconds": 1.0238104009986273,
394
- "bytes": 4591810,
395
- "pages": 32,
396
- "shards": 2
397
- },
398
- {
399
- "seconds": 2.234572448003746,
400
- "bytes": 13274402,
401
- "pages": 35,
402
- "shards": 4
403
- },
404
- {
405
- "seconds": 1.07868888899975,
406
- "bytes": 1300513,
407
- "pages": 36,
408
- "shards": 1
409
- },
410
- {
411
- "seconds": 1.4324448540064623,
412
- "bytes": 9935435,
413
- "pages": 35,
414
  "shards": 3
415
  },
416
  {
417
- "seconds": 1.2297556650009938,
418
- "bytes": 7626325,
419
  "pages": 34,
420
- "shards": 3
421
- },
422
- {
423
- "seconds": 0.8381233459949726,
424
- "bytes": 1148268,
425
- "pages": 32,
426
  "shards": 1
427
  },
428
  {
429
- "seconds": 0.9499172950017964,
430
- "bytes": 7609450,
431
- "pages": 39,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 0.9056497909987229,
436
- "bytes": 4987206,
437
- "pages": 25,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 1.4441368190018693,
442
- "bytes": 11962393,
443
- "pages": 36,
444
- "shards": 3
445
- },
446
- {
447
- "seconds": 1.1221481789980317,
448
- "bytes": 7553276,
449
- "pages": 38,
450
- "shards": 3
451
- },
452
- {
453
- "seconds": 0.8207161469981656,
454
- "bytes": 1164563,
455
- "pages": 36,
456
- "shards": 1
457
- },
458
- {
459
- "seconds": 1.0845683700026711,
460
- "bytes": 10426904,
461
- "pages": 28,
462
- "shards": 3
463
  },
464
  {
465
- "seconds": 1.0798400059939013,
466
- "bytes": 7684665,
467
  "pages": 39,
468
- "shards": 3
469
  },
470
  {
471
- "seconds": 1.0031564969976898,
472
- "bytes": 997756,
473
- "pages": 28,
474
  "shards": 1
475
  },
476
  {
477
- "seconds": 1.3090939850007999,
478
- "bytes": 13205063,
479
- "pages": 36,
480
- "shards": 4
481
- },
482
- {
483
- "seconds": 1.0676715709996643,
484
- "bytes": 4977396,
485
- "pages": 37,
486
  "shards": 3
487
  },
488
  {
489
- "seconds": 0.9372590880011558,
490
- "bytes": 6217543,
491
- "pages": 37,
492
  "shards": 2
493
  },
494
  {
495
- "seconds": 1.0400478770025074,
496
- "bytes": 7313545,
497
- "pages": 31,
498
- "shards": 3
499
- },
500
- {
501
- "seconds": 0.8696056030021282,
502
- "bytes": 1079259,
503
- "pages": 34,
504
- "shards": 1
505
  },
506
  {
507
- "seconds": 1.1580835760032642,
508
- "bytes": 10392878,
509
- "pages": 30,
510
  "shards": 3
511
  },
512
  {
513
- "seconds": 1.3803016649981146,
514
- "bytes": 13473573,
515
- "pages": 39,
516
- "shards": 4
517
- },
518
- {
519
- "seconds": 0.9864481259937747,
520
- "bytes": 1047459,
521
- "pages": 35,
522
- "shards": 1
523
  },
524
  {
525
- "seconds": 1.009063662000699,
526
- "bytes": 7464078,
527
- "pages": 31,
528
- "shards": 3
529
  },
530
  {
531
- "seconds": 1.1531412479962455,
532
- "bytes": 5169545,
533
- "pages": 35,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 1.040382787999988,
538
- "bytes": 5992429,
539
- "pages": 33,
540
  "shards": 2
541
  },
542
  {
543
- "seconds": 1.0942238300049212,
544
- "bytes": 7544195,
545
- "pages": 32,
546
  "shards": 3
547
  },
548
  {
549
- "seconds": 1.183839971999987,
550
- "bytes": 6786466,
551
- "pages": 32,
552
  "shards": 2
553
  },
554
  {
555
- "seconds": 1.5759701440038043,
556
- "bytes": 11667943,
557
- "pages": 39,
558
- "shards": 4
559
- },
560
- {
561
- "seconds": 1.0282209380020504,
562
- "bytes": 1021880,
563
  "pages": 32,
564
- "shards": 1
565
  }
566
  ],
567
  "cache_key": {
@@ -606,7 +462,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 1.1575575899996693,
610
- 0.014994120000665134
611
  ]
612
  }
 
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": 75458.65267281262,
26
+ "required_pages_per_second_upper_estimate": 0.006540564765472674
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": 9432.331584101577,
40
+ "required_pages_per_second_upper_estimate": 0.003762698094822713
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": 18864.663168203155,
54
+ "required_pages_per_second_upper_estimate": 0.004364007817231477
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": 28296.994752304734,
68
+ "required_pages_per_second_upper_estimate": 0.0017231385175577071
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": 28296.994752304734,
82
+ "required_pages_per_second_upper_estimate": 0.002178380987127306
83
  },
84
  {
85
  "source": "cortex",
 
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": 28296.994752304734,
96
+ "required_pages_per_second_upper_estimate": 0.35135397087431536
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
157
  "selected": {
158
+ "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 690670,
161
+ "producer_batch_seconds": 0.0511901649687303,
162
+ "producer_p95_seconds": 0.18669305799994618,
163
+ "gpu_update_seconds": 0.9770648877032472,
164
+ "producer_batches_per_second": 19.535002487506215,
165
+ "gpu_batches_per_second": 1.0234734791776885,
166
+ "producer_headroom": 19.086965011737917,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.57802303760133,
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": 395163.83161593793,
174
+ "upload_bytes_per_second": 5372382.916989888,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 84.79295932158092,
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": 1.0731404800000064,
184
+ "bytes": 44464,
185
+ "pages": 1,
186
+ "shards": 1
187
  },
188
  {
189
+ "seconds": 1.131805757000052,
190
+ "bytes": 882364,
191
+ "pages": 25,
192
+ "shards": 1
193
  },
194
  {
195
+ "seconds": 1.1303505340000584,
196
+ "bytes": 11613053,
197
+ "pages": 29,
198
+ "shards": 3
199
  },
200
  {
201
+ "seconds": 0.9671371199999612,
202
+ "bytes": 2609325,
203
+ "pages": 34,
204
  "shards": 2
205
  },
206
  {
207
+ "seconds": 0.9158913300000222,
208
+ "bytes": 6129878,
209
+ "pages": 32,
210
+ "shards": 2
 
 
 
 
 
 
211
  },
212
  {
213
+ "seconds": 1.5951903079999283,
214
+ "bytes": 6553135,
215
+ "pages": 34,
216
  "shards": 3
217
  },
218
  {
219
+ "seconds": 0.9309580040001038,
220
+ "bytes": 6008731,
221
+ "pages": 35,
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 1.3192515309999635,
226
+ "bytes": 6183122,
227
+ "pages": 35,
228
+ "shards": 2
229
  },
230
  {
231
+ "seconds": 1.2959069240000645,
232
+ "bytes": 8294236,
233
+ "pages": 30,
234
  "shards": 3
235
  },
236
  {
237
+ "seconds": 1.3743943120000495,
238
+ "bytes": 10663078,
239
+ "pages": 35,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 0.8003151779998916,
244
+ "bytes": 1078242,
245
+ "pages": 33,
246
  "shards": 1
247
  },
248
  {
249
+ "seconds": 1.0970220180001888,
250
+ "bytes": 7456296,
251
+ "pages": 36,
252
+ "shards": 3
253
  },
254
  {
255
+ "seconds": 1.0635986559996127,
256
+ "bytes": 6186061,
257
  "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 1.1765199270002995,
262
+ "bytes": 6893561,
263
+ "pages": 28,
 
 
 
 
 
 
264
  "shards": 3
265
  },
266
  {
267
+ "seconds": 3.6869929439999396,
268
+ "bytes": 11547756,
269
+ "pages": 29,
 
 
 
 
 
 
 
 
 
 
 
 
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 0.8728801329998532,
274
+ "bytes": 1038671,
275
+ "pages": 33,
276
  "shards": 1
277
  },
278
  {
279
+ "seconds": 1.359243317000164,
280
+ "bytes": 7794752,
281
+ "pages": 37,
282
  "shards": 4
283
  },
284
  {
285
+ "seconds": 0.911286476999976,
286
+ "bytes": 856023,
287
+ "pages": 24,
 
 
 
 
 
 
288
  "shards": 1
289
  },
290
  {
291
+ "seconds": 1.8453115359998264,
292
+ "bytes": 9879354,
293
+ "pages": 35,
 
 
 
 
 
 
294
  "shards": 3
295
  },
296
  {
297
+ "seconds": 1.0517817119998654,
298
+ "bytes": 7714712,
299
+ "pages": 37,
 
 
 
 
 
 
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 0.9509194960000968,
304
+ "bytes": 6981450,
 
 
 
 
 
 
 
 
 
 
 
 
305
  "pages": 37,
 
 
 
 
 
 
 
 
 
 
 
 
306
  "shards": 2
307
  },
308
  {
309
+ "seconds": 1.187183233000269,
310
+ "bytes": 9995831,
311
  "pages": 33,
312
  "shards": 3
313
  },
314
  {
315
+ "seconds": 1.3044787690000703,
316
+ "bytes": 7311631,
317
+ "pages": 28,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
318
  "shards": 3
319
  },
320
  {
321
+ "seconds": 0.7353289520001454,
322
+ "bytes": 1217005,
323
  "pages": 34,
 
 
 
 
 
 
324
  "shards": 1
325
  },
326
  {
327
+ "seconds": 1.1706305770003382,
328
+ "bytes": 7901550,
329
+ "pages": 37,
330
  "shards": 3
331
  },
332
  {
333
+ "seconds": 1.067005848000008,
334
+ "bytes": 5080800,
335
+ "pages": 32,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 0.9038640889998533,
340
+ "bytes": 6072537,
341
+ "pages": 27,
342
+ "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
343
  },
344
  {
345
+ "seconds": 1.318457908000255,
346
+ "bytes": 13323724,
347
  "pages": 39,
348
+ "shards": 4
349
  },
350
  {
351
+ "seconds": 0.8108393100001194,
352
+ "bytes": 1203088,
353
+ "pages": 33,
354
  "shards": 1
355
  },
356
  {
357
+ "seconds": 1.1118799440000657,
358
+ "bytes": 10497270,
359
+ "pages": 30,
 
 
 
 
 
 
360
  "shards": 3
361
  },
362
  {
363
+ "seconds": 0.8437159310005882,
364
+ "bytes": 2537590,
365
+ "pages": 35,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.248541894000482,
370
+ "bytes": 6395911,
371
+ "pages": 35,
372
+ "shards": 2
 
 
 
 
 
 
373
  },
374
  {
375
+ "seconds": 1.171668709000187,
376
+ "bytes": 7290583,
377
+ "pages": 33,
378
  "shards": 3
379
  },
380
  {
381
+ "seconds": 1.2945431420002933,
382
+ "bytes": 10577919,
383
+ "pages": 37,
384
+ "shards": 3
 
 
 
 
 
 
385
  },
386
  {
387
+ "seconds": 1.2328179739997722,
388
+ "bytes": 5713232,
389
+ "pages": 27,
390
+ "shards": 2
391
  },
392
  {
393
+ "seconds": 0.9817969460000313,
394
+ "bytes": 2633705,
395
+ "pages": 36,
396
  "shards": 2
397
  },
398
  {
399
+ "seconds": 1.1478959790001682,
400
+ "bytes": 6105703,
401
+ "pages": 35,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 1.3559997279999152,
406
+ "bytes": 10442134,
407
+ "pages": 30,
408
  "shards": 3
409
  },
410
  {
411
+ "seconds": 0.8416516239994962,
412
+ "bytes": 2569575,
413
+ "pages": 36,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 1.165861663999749,
418
+ "bytes": 11609605,
 
 
 
 
 
 
419
  "pages": 32,
420
+ "shards": 3
421
  }
422
  ],
423
  "cache_key": {
 
462
  },
463
  "cache_reused": true,
464
  "resume_probe_seconds": [
465
+ 2.358020696999972,
466
+ 0.013266629999975521
467
  ]
468
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0becabd5ddaf3503de95d1c46e0a2cab7a3ad153cc954b1a5c27591cd4e7fba4
3
- size 721773887
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38c2f886ccf49b55cd34e215a6a6c1aa8127d46138bf70445d4ca70032231d6c
3
+ size 721783871
training_config.json CHANGED
@@ -46,7 +46,7 @@
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,
 
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,