appvoid commited on
Commit
7628b88
·
verified ·
1 Parent(s): fe18c57

Qwen90M step 295000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +45 -45
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +250 -250
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 290000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "0a0aa696d2def99e7d22de3de4aeff668d43be50c970daf9230586d60febe595",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "00b52798ce4bbc979b3cc619c785cf5f2e3b8c46472773cb746b19a780d676ff",
12
- "training_config.json": "65390fe09f12d0188f44175607f7a9d1665082a136656274cb5d054eb998e80b",
13
  "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba",
14
- "shard_scheduler.json": "6b224639b01f5768d2f11ee38fffbf310fdb30edc5736dacb2cee65fc42a8162",
15
- "coverage.json": "244fa16ffdbcd8c9dc4f95fdb90318ed6782c1c785fc2b0b5b817be4e931ecb6",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 295000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "59f58f5c4bdb2a0576a3b78dd8f8666a1dd1d6108b3210d2d71cfcc0ff980702",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "9182b42958ff73eae94225e7d961ee0aec9363f3f5f387fc8e3072607f9dc6fc",
12
+ "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba",
14
+ "shard_scheduler.json": "aeba209266c0f57fa8c60a462539562d488605c9131398cabd5bb4cec88efffe",
15
+ "coverage.json": "517abe535f2de9e8ba9121e0877123b28aeea27b9303155337162e9d0163f11e",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 290000,
3
  "stats": {
4
- "climbmix/blocks": 10440000,
5
- "climbmix/targets": 17452321234,
6
- "climbmix/input_tokens": 17453427448,
7
- "ultra_qa/blocks": 3915000,
8
- "ultra_qa/targets": 6386778194,
9
- "ultra_qa/input_tokens": 6386790465,
10
- "cortex/blocks": 3915000,
11
- "cortex/targets": 148927427,
12
- "cortex/input_tokens": 6223329188,
13
- "ultra_style/blocks": 3915000,
14
- "ultra_style/targets": 6192313215,
15
- "ultra_style/input_tokens": 6192638355,
16
- "openmath/blocks": 2610000,
17
- "openmath/targets": 4033044745,
18
- "openmath/input_tokens": 4034389394,
19
- "rewrite6/blocks": 1305000,
20
- "rewrite6/targets": 2139357551,
21
- "rewrite6/input_tokens": 2141703031
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1429,
26
- "row": 3549,
27
  "start": {
28
- "file": 69,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 14,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 5856733
38
  },
39
  "rewrite6": {
40
- "page": 13,
41
- "row": 1729,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 53248
47
  }
48
  },
49
  "epoch": 16,
50
  "offset": 0,
51
- "documents": 3341441
52
  },
53
  "openmath": {
54
- "page": 893,
55
- "row": 3429,
56
  "start": {
57
  "file": 8,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 164,
61
- "row": 528
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 3661157
67
  },
68
  "ultra_qa": {
69
- "page": 386,
70
- "row": 1389,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
- "row": 22181
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 1582445
82
  },
83
  "ultra_style": {
84
- "page": 590,
85
- "row": 113,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
- "row": 20398
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 2048,
96
- "documents": 2416753
97
  },
98
  "cortex": {
99
- "page": 75929,
100
- "row": 52,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 9718964
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 295000,
3
  "stats": {
4
+ "climbmix/blocks": 10620000,
5
+ "climbmix/targets": 17753092470,
6
+ "climbmix/input_tokens": 17754217730,
7
+ "ultra_qa/blocks": 3982500,
8
+ "ultra_qa/targets": 6496902054,
9
+ "ultra_qa/input_tokens": 6496914542,
10
+ "cortex/blocks": 3982500,
11
+ "cortex/targets": 151499631,
12
+ "cortex/input_tokens": 6330619515,
13
+ "ultra_style/blocks": 3982500,
14
+ "ultra_style/targets": 6299348778,
15
+ "ultra_style/input_tokens": 6299679365,
16
+ "openmath/blocks": 2655000,
17
+ "openmath/targets": 4102640002,
18
+ "openmath/input_tokens": 4104007922,
19
+ "rewrite6/blocks": 1327500,
20
+ "rewrite6/targets": 2176202421,
21
+ "rewrite6/input_tokens": 2178588636
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1454,
26
+ "row": 2314,
27
  "start": {
28
+ "file": 70,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 31,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 5957898
38
  },
39
  "rewrite6": {
40
+ "page": 27,
41
+ "row": 2421,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 110592
47
  }
48
  },
49
  "epoch": 16,
50
  "offset": 0,
51
+ "documents": 3399477
52
  },
53
  "openmath": {
54
+ "page": 909,
55
+ "row": 1156,
56
  "start": {
57
  "file": 8,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 230,
61
+ "row": 64
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 3724420
67
  },
68
  "ultra_qa": {
69
+ "page": 393,
70
+ "row": 0,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
+ "row": 50853
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 1609728
82
  },
83
  "ultra_style": {
84
+ "page": 600,
85
+ "row": 410,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
+ "row": 61358
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 4096,
96
+ "documents": 2458010
97
  },
98
  "cortex": {
99
+ "page": 77238,
100
+ "row": 62,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 9886526
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 290000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "0a0aa696d2def99e7d22de3de4aeff668d43be50c970daf9230586d60febe595", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "00b52798ce4bbc979b3cc619c785cf5f2e3b8c46472773cb746b19a780d676ff", "training_config.json": "65390fe09f12d0188f44175607f7a9d1665082a136656274cb5d054eb998e80b", "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba", "shard_scheduler.json": "6b224639b01f5768d2f11ee38fffbf310fdb30edc5736dacb2cee65fc42a8162", "coverage.json": "244fa16ffdbcd8c9dc4f95fdb90318ed6782c1c785fc2b0b5b817be4e931ecb6", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000290000"}
 
1
+ {"step": 295000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "59f58f5c4bdb2a0576a3b78dd8f8666a1dd1d6108b3210d2d71cfcc0ff980702", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "9182b42958ff73eae94225e7d961ee0aec9363f3f5f387fc8e3072607f9dc6fc", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba", "shard_scheduler.json": "aeba209266c0f57fa8c60a462539562d488605c9131398cabd5bb4cec88efffe", "coverage.json": "517abe535f2de9e8ba9121e0877123b28aeea27b9303155337162e9d0163f11e", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000295000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0a0aa696d2def99e7d22de3de4aeff668d43be50c970daf9230586d60febe595
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:59f58f5c4bdb2a0576a3b78dd8f8666a1dd1d6108b3210d2d71cfcc0ff980702
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": 76378.33191482387,
26
- "required_pages_per_second_upper_estimate": 0.006620280231264481
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": 9547.291489352983,
40
- "required_pages_per_second_upper_estimate": 0.0038085573198312524
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": 19094.582978705967,
54
- "required_pages_per_second_upper_estimate": 0.004417195718940841
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": 28641.87446805895,
68
- "required_pages_per_second_upper_estimate": 0.0017441398828031117
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": 28641.87446805895,
82
- "required_pages_per_second_upper_estimate": 0.0022049307823342216
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": 28641.87446805895,
96
- "required_pages_per_second_upper_estimate": 0.35563622270582407
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": 690670,
161
- "producer_batch_seconds": 0.0629032884064884,
162
- "producer_p95_seconds": 0.20650791799926083,
163
- "gpu_update_seconds": 0.9652999502819793,
164
- "producer_batches_per_second": 15.897420076640241,
165
- "gpu_batches_per_second": 1.0359474272301414,
166
- "producer_headroom": 15.345778809592565,
167
  "sustainable": true,
168
- "buffer_seconds": 60.8138968677647,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 82,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 543062.8026689846,
174
- "upload_bytes_per_second": 5443648.407667226,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 116.52863559772638,
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": 1.2733040839993919,
184
- "bytes": 13322526,
185
- "pages": 35,
186
  "shards": 4
187
  },
188
  {
189
- "seconds": 0.8405508110008668,
190
- "bytes": 1090490,
191
- "pages": 33,
 
 
 
 
 
 
192
  "shards": 1
193
  },
194
  {
195
- "seconds": 1.1666756240010727,
196
- "bytes": 10146533,
197
- "pages": 27,
 
 
 
 
 
 
198
  "shards": 3
199
  },
200
  {
201
- "seconds": 1.1112560749988916,
202
- "bytes": 2663601,
203
- "pages": 36,
 
 
 
 
 
 
204
  "shards": 3
205
  },
206
  {
207
- "seconds": 1.182102421998934,
208
- "bytes": 5962190,
209
- "pages": 36,
 
 
 
 
 
 
210
  "shards": 2
211
  },
212
  {
213
- "seconds": 0.9607173089989374,
214
- "bytes": 7750938,
215
  "pages": 37,
216
  "shards": 3
217
  },
218
  {
219
- "seconds": 1.1050905230003991,
220
- "bytes": 6187776,
221
- "pages": 15,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 1.2981188600006135,
226
- "bytes": 10611352,
227
- "pages": 45,
228
- "shards": 3
229
  },
230
  {
231
- "seconds": 1.1970796210007393,
232
- "bytes": 2312155,
233
- "pages": 27,
234
  "shards": 2
235
  },
236
  {
237
- "seconds": 1.6818293220003397,
238
- "bytes": 5841398,
239
- "pages": 35,
240
- "shards": 2
 
 
 
 
 
 
241
  },
242
  {
243
- "seconds": 1.1767334710002615,
244
- "bytes": 1036234,
245
  "pages": 30,
246
  "shards": 1
247
  },
248
  {
249
- "seconds": 1.7388112589997036,
250
- "bytes": 12464285,
251
- "pages": 43,
252
- "shards": 4
253
  },
254
  {
255
- "seconds": 0.8487006530012877,
256
- "bytes": 592565,
257
- "pages": 19,
258
- "shards": 1
259
  },
260
  {
261
- "seconds": 1.3723258429999987,
262
- "bytes": 11921886,
263
- "pages": 40,
264
- "shards": 3
265
  },
266
  {
267
- "seconds": 1.34920726099881,
268
- "bytes": 7816018,
269
- "pages": 35,
270
  "shards": 3
271
  },
272
  {
273
- "seconds": 1.4124847809998755,
274
- "bytes": 4874448,
275
- "pages": 37,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.2419382850002876,
280
- "bytes": 7186580,
281
- "pages": 29,
282
  "shards": 3
283
  },
284
  {
285
- "seconds": 1.1265525479993812,
286
- "bytes": 6298017,
287
- "pages": 35,
288
- "shards": 2
289
  },
290
  {
291
- "seconds": 0.9624441889991431,
292
- "bytes": 1027701,
293
- "pages": 30,
294
  "shards": 1
295
  },
296
  {
297
- "seconds": 1.7364580140001635,
298
- "bytes": 17104134,
299
- "pages": 27,
300
- "shards": 5
301
  },
302
  {
303
- "seconds": 0.9077872060006484,
304
- "bytes": 916446,
305
- "pages": 29,
306
  "shards": 1
307
  },
308
  {
309
- "seconds": 1.0004914799992548,
310
- "bytes": 6063816,
311
  "pages": 36,
312
- "shards": 2
313
  },
314
  {
315
- "seconds": 0.9483974380000291,
316
- "bytes": 2643403,
317
- "pages": 36,
318
  "shards": 2
319
  },
320
  {
321
- "seconds": 1.2182073419990047,
322
- "bytes": 10399159,
323
- "pages": 23,
324
- "shards": 3
325
  },
326
  {
327
- "seconds": 0.9434353479991842,
328
- "bytes": 5975783,
329
- "pages": 35,
330
  "shards": 2
331
  },
332
  {
333
- "seconds": 0.9842061780000222,
334
- "bytes": 8165489,
335
- "pages": 31,
336
- "shards": 3
337
- },
338
- {
339
- "seconds": 0.9350508080005966,
340
- "bytes": 6046815,
341
  "pages": 36,
342
  "shards": 2
343
  },
344
  {
345
- "seconds": 1.3128581109995139,
346
- "bytes": 2374553,
347
  "pages": 27,
348
  "shards": 2
349
  },
350
  {
351
- "seconds": 1.2559754029989563,
352
- "bytes": 9926404,
353
- "pages": 38,
354
- "shards": 3
355
  },
356
  {
357
- "seconds": 0.9813094319997617,
358
- "bytes": 1049873,
359
- "pages": 31,
360
- "shards": 1
361
  },
362
  {
363
- "seconds": 1.218386226999428,
364
- "bytes": 7220161,
365
- "pages": 32,
366
  "shards": 3
367
  },
368
  {
369
- "seconds": 0.9670737249998638,
370
- "bytes": 5661886,
371
- "pages": 23,
372
- "shards": 2
373
  },
374
  {
375
- "seconds": 1.3326636759993562,
376
- "bytes": 12247274,
377
- "pages": 47,
378
- "shards": 3
379
  },
380
  {
381
- "seconds": 1.6376108280001063,
382
- "bytes": 7445606,
383
- "pages": 29,
384
  "shards": 3
385
  },
386
  {
387
- "seconds": 0.7973876159994688,
388
- "bytes": 1188820,
389
  "pages": 34,
390
  "shards": 1
391
  },
392
  {
393
- "seconds": 1.1932972649992735,
394
- "bytes": 7607279,
395
- "pages": 37,
396
- "shards": 4
397
- },
398
- {
399
- "seconds": 1.4406277779999073,
400
- "bytes": 9581963,
401
- "pages": 33,
402
  "shards": 3
403
  },
404
  {
405
- "seconds": 1.2597373989992775,
406
- "bytes": 7034037,
407
- "pages": 38,
408
- "shards": 2
409
- },
410
- {
411
- "seconds": 1.1449642560000939,
412
- "bytes": 7450954,
413
- "pages": 28,
414
- "shards": 3
415
  },
416
  {
417
- "seconds": 0.8681581979999464,
418
- "bytes": 818733,
419
- "pages": 23,
420
  "shards": 1
421
  },
422
  {
423
- "seconds": 1.2214710799999011,
424
- "bytes": 12703928,
425
- "pages": 49,
426
- "shards": 4
427
  },
428
  {
429
- "seconds": 1.1181347120000282,
430
- "bytes": 6085437,
431
  "pages": 35,
432
  "shards": 2
433
  },
434
  {
435
- "seconds": 1.1846620260002965,
436
- "bytes": 927364,
437
- "pages": 26,
438
- "shards": 1
439
- },
440
- {
441
- "seconds": 1.263599650001197,
442
- "bytes": 13782793,
443
- "pages": 42,
444
- "shards": 4
445
  },
446
  {
447
- "seconds": 1.1351800550000917,
448
- "bytes": 9521447,
449
- "pages": 29,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 1.0451152109999384,
454
- "bytes": 1099454,
455
- "pages": 31,
456
- "shards": 1
457
  },
458
  {
459
- "seconds": 1.1156518739990133,
460
- "bytes": 7370564,
461
- "pages": 33,
462
- "shards": 3
463
  },
464
  {
465
- "seconds": 1.0604308100009803,
466
- "bytes": 5273416,
467
- "pages": 37,
468
- "shards": 2
469
  },
470
  {
471
- "seconds": 1.123748488000274,
472
- "bytes": 5660713,
473
- "pages": 27,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.1639190339992638,
478
- "bytes": 7647995,
479
- "pages": 37,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 1.0442986829984875,
484
- "bytes": 6780113,
485
- "pages": 31,
486
- "shards": 2
487
  },
488
  {
489
- "seconds": 1.2669327539988444,
490
- "bytes": 12406942,
491
- "pages": 41,
492
  "shards": 4
493
  },
494
  {
495
- "seconds": 1.4563389440008905,
496
- "bytes": 6496961,
497
- "pages": 33,
498
  "shards": 2
499
  },
500
  {
501
- "seconds": 1.1376306590027525,
502
- "bytes": 968236,
503
- "pages": 28,
504
- "shards": 1
505
  },
506
  {
507
- "seconds": 1.3645541389996652,
508
- "bytes": 7442538,
509
- "pages": 39,
510
- "shards": 3
511
  },
512
  {
513
- "seconds": 0.78800499800127,
514
- "bytes": 976266,
515
  "pages": 30,
516
- "shards": 1
517
- },
518
- {
519
- "seconds": 1.7201119790006487,
520
- "bytes": 17653647,
521
- "pages": 35,
522
- "shards": 5
523
  },
524
  {
525
- "seconds": 1.0727579819977109,
526
- "bytes": 6128146,
527
  "pages": 36,
528
- "shards": 2
529
  },
530
  {
531
- "seconds": 1.0307726470018679,
532
- "bytes": 1163007,
533
- "pages": 34,
534
  "shards": 1
535
  },
536
  {
537
- "seconds": 1.1981934089999413,
538
- "bytes": 7536231,
539
- "pages": 38,
540
- "shards": 3
541
  },
542
  {
543
- "seconds": 1.27145491400006,
544
- "bytes": 5090460,
545
- "pages": 31,
546
  "shards": 2
547
  },
548
  {
549
- "seconds": 1.1308283350008423,
550
- "bytes": 7872807,
551
- "pages": 35,
552
  "shards": 3
553
  },
554
  {
555
- "seconds": 2.8297132219995547,
556
- "bytes": 11437747,
557
- "pages": 30,
558
- "shards": 3
559
  },
560
  {
561
- "seconds": 1.1368421070001204,
562
- "bytes": 1161800,
563
- "pages": 34,
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": 73779.41610322442,
26
+ "required_pages_per_second_upper_estimate": 0.006395012795606944
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": 9222.427012903052,
40
+ "required_pages_per_second_upper_estimate": 0.0036789640230186103
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": 18444.854025806104,
54
+ "required_pages_per_second_upper_estimate": 0.004266892360526481
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": 27667.281038709156,
68
+ "required_pages_per_second_upper_estimate": 0.0016847922562522565
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": 27667.281038709156,
82
+ "required_pages_per_second_upper_estimate": 0.002129903882295592
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": 27667.281038709156,
96
+ "required_pages_per_second_upper_estimate": 0.3435350340676851
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": 0.4320460435155269,
162
+ "producer_p95_seconds": 0.3225658029987244,
163
+ "gpu_update_seconds": 0.9993031104616974,
164
+ "producer_batches_per_second": 2.314568123024744,
165
+ "gpu_batches_per_second": 1.0006973755320152,
166
+ "producer_headroom": 2.3129551247141196,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.95748973816354,
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": 296585.41787994764,
174
+ "upload_bytes_per_second": 3972390.366839625,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 63.640326521862235,
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": 1.2214710799999011,
184
+ "bytes": 12703928,
185
+ "pages": 49,
186
  "shards": 4
187
  },
188
  {
189
+ "seconds": 1.1181347120000282,
190
+ "bytes": 6085437,
191
+ "pages": 35,
192
+ "shards": 2
193
+ },
194
+ {
195
+ "seconds": 1.1846620260002965,
196
+ "bytes": 927364,
197
+ "pages": 26,
198
  "shards": 1
199
  },
200
  {
201
+ "seconds": 1.263599650001197,
202
+ "bytes": 13782793,
203
+ "pages": 42,
204
+ "shards": 4
205
+ },
206
+ {
207
+ "seconds": 1.1351800550000917,
208
+ "bytes": 9521447,
209
+ "pages": 29,
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.0451152109999384,
214
+ "bytes": 1099454,
215
+ "pages": 31,
216
+ "shards": 1
217
+ },
218
+ {
219
+ "seconds": 1.1156518739990133,
220
+ "bytes": 7370564,
221
+ "pages": 33,
222
  "shards": 3
223
  },
224
  {
225
+ "seconds": 1.0604308100009803,
226
+ "bytes": 5273416,
227
+ "pages": 37,
228
+ "shards": 2
229
+ },
230
+ {
231
+ "seconds": 1.123748488000274,
232
+ "bytes": 5660713,
233
+ "pages": 27,
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.1639190339992638,
238
+ "bytes": 7647995,
239
  "pages": 37,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 1.0442986829984875,
244
+ "bytes": 6780113,
245
+ "pages": 31,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.2669327539988444,
250
+ "bytes": 12406942,
251
+ "pages": 41,
252
+ "shards": 4
253
  },
254
  {
255
+ "seconds": 1.4563389440008905,
256
+ "bytes": 6496961,
257
+ "pages": 33,
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 1.1376306590027525,
262
+ "bytes": 968236,
263
+ "pages": 28,
264
+ "shards": 1
265
+ },
266
+ {
267
+ "seconds": 1.3645541389996652,
268
+ "bytes": 7442538,
269
+ "pages": 39,
270
+ "shards": 3
271
  },
272
  {
273
+ "seconds": 0.78800499800127,
274
+ "bytes": 976266,
275
  "pages": 30,
276
  "shards": 1
277
  },
278
  {
279
+ "seconds": 1.7201119790006487,
280
+ "bytes": 17653647,
281
+ "pages": 35,
282
+ "shards": 5
283
  },
284
  {
285
+ "seconds": 1.0727579819977109,
286
+ "bytes": 6128146,
287
+ "pages": 36,
288
+ "shards": 2
289
  },
290
  {
291
+ "seconds": 1.0307726470018679,
292
+ "bytes": 1163007,
293
+ "pages": 34,
294
+ "shards": 1
295
  },
296
  {
297
+ "seconds": 1.1981934089999413,
298
+ "bytes": 7536231,
299
+ "pages": 38,
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 1.27145491400006,
304
+ "bytes": 5090460,
305
+ "pages": 31,
306
  "shards": 2
307
  },
308
  {
309
+ "seconds": 1.1308283350008423,
310
+ "bytes": 7872807,
311
+ "pages": 35,
312
  "shards": 3
313
  },
314
  {
315
+ "seconds": 2.8297132219995547,
316
+ "bytes": 11437747,
317
+ "pages": 30,
318
+ "shards": 3
319
  },
320
  {
321
+ "seconds": 1.1368421070001204,
322
+ "bytes": 1161800,
323
+ "pages": 34,
324
  "shards": 1
325
  },
326
  {
327
+ "seconds": 1.1932739869989746,
328
+ "bytes": 11243706,
329
+ "pages": 26,
330
+ "shards": 4
331
  },
332
  {
333
+ "seconds": 1.4697227199976624,
334
+ "bytes": 921486,
335
+ "pages": 26,
336
  "shards": 1
337
  },
338
  {
339
+ "seconds": 1.586299536000297,
340
+ "bytes": 6350565,
341
  "pages": 36,
342
+ "shards": 3
343
  },
344
  {
345
+ "seconds": 0.9621634939976502,
346
+ "bytes": 2554887,
347
+ "pages": 35,
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 1.2608247060015856,
352
+ "bytes": 16000072,
353
+ "pages": 34,
354
+ "shards": 4
355
  },
356
  {
357
+ "seconds": 1.1701329820025421,
358
+ "bytes": 6128160,
359
+ "pages": 36,
360
  "shards": 2
361
  },
362
  {
363
+ "seconds": 0.9900568350021786,
364
+ "bytes": 2575859,
 
 
 
 
 
 
365
  "pages": 36,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.209377144001337,
370
+ "bytes": 5838608,
371
  "pages": 27,
372
  "shards": 2
373
  },
374
  {
375
+ "seconds": 1.249938580000162,
376
+ "bytes": 4790943,
377
+ "pages": 33,
378
+ "shards": 2
379
  },
380
  {
381
+ "seconds": 1.2871860160012147,
382
+ "bytes": 7434728,
383
+ "pages": 38,
384
+ "shards": 3
385
  },
386
  {
387
+ "seconds": 31.810438357002567,
388
+ "bytes": 11940488,
389
+ "pages": 36,
390
  "shards": 3
391
  },
392
  {
393
+ "seconds": 1.3719915590008895,
394
+ "bytes": 12062246,
395
+ "pages": 44,
396
+ "shards": 4
397
  },
398
  {
399
+ "seconds": 0.804579094001383,
400
+ "bytes": 1179275,
401
+ "pages": 34,
402
+ "shards": 1
403
  },
404
  {
405
+ "seconds": 1.2755006959996535,
406
+ "bytes": 7429804,
407
+ "pages": 31,
408
  "shards": 3
409
  },
410
  {
411
+ "seconds": 1.0133618440013379,
412
+ "bytes": 1167322,
413
  "pages": 34,
414
  "shards": 1
415
  },
416
  {
417
+ "seconds": 1.291220240000257,
418
+ "bytes": 10903036,
419
+ "pages": 35,
 
 
 
 
 
 
420
  "shards": 3
421
  },
422
  {
423
+ "seconds": 1.2618899920016702,
424
+ "bytes": 13332712,
425
+ "pages": 32,
426
+ "shards": 4
 
 
 
 
 
 
427
  },
428
  {
429
+ "seconds": 0.7849141779988713,
430
+ "bytes": 1153413,
431
+ "pages": 34,
432
  "shards": 1
433
  },
434
  {
435
+ "seconds": 1.3497758060002525,
436
+ "bytes": 7313270,
437
+ "pages": 37,
438
+ "shards": 3
439
  },
440
  {
441
+ "seconds": 1.170668664999539,
442
+ "bytes": 6087923,
443
  "pages": 35,
444
  "shards": 2
445
  },
446
  {
447
+ "seconds": 1.1090394350030692,
448
+ "bytes": 4744304,
449
+ "pages": 27,
450
+ "shards": 2
 
 
 
 
 
 
451
  },
452
  {
453
+ "seconds": 1.304362898998079,
454
+ "bytes": 7978674,
455
+ "pages": 39,
456
  "shards": 3
457
  },
458
  {
459
+ "seconds": 1.0655079770003795,
460
+ "bytes": 7055873,
461
+ "pages": 37,
462
+ "shards": 2
463
  },
464
  {
465
+ "seconds": 1.268422623998049,
466
+ "bytes": 5726191,
467
+ "pages": 20,
468
+ "shards": 2
469
  },
470
  {
471
+ "seconds": 1.1066016949989717,
472
+ "bytes": 7887477,
473
+ "pages": 44,
474
+ "shards": 3
475
  },
476
  {
477
+ "seconds": 1.019014984001842,
478
+ "bytes": 6005453,
479
+ "pages": 36,
480
  "shards": 2
481
  },
482
  {
483
+ "seconds": 1.00243121099993,
484
+ "bytes": 7452907,
485
+ "pages": 29,
486
  "shards": 3
487
  },
488
  {
489
+ "seconds": 0.9669066969981941,
490
+ "bytes": 1217353,
491
+ "pages": 34,
492
+ "shards": 1
493
  },
494
  {
495
+ "seconds": 1.4986801279992505,
496
+ "bytes": 15911648,
497
+ "pages": 32,
498
  "shards": 4
499
  },
500
  {
501
+ "seconds": 1.1255449119998957,
502
+ "bytes": 2547314,
503
+ "pages": 34,
504
  "shards": 2
505
  },
506
  {
507
+ "seconds": 1.1576964999985648,
508
+ "bytes": 6153703,
509
+ "pages": 36,
510
+ "shards": 2
511
  },
512
  {
513
+ "seconds": 1.0455056530008733,
514
+ "bytes": 6073870,
515
+ "pages": 35,
516
+ "shards": 2
517
  },
518
  {
519
+ "seconds": 1.0066101270022045,
520
+ "bytes": 7063346,
521
  "pages": 30,
522
+ "shards": 3
 
 
 
 
 
 
523
  },
524
  {
525
+ "seconds": 1.161024010998517,
526
+ "bytes": 6075989,
527
  "pages": 36,
528
+ "shards": 3
529
  },
530
  {
531
+ "seconds": 0.8424355740025931,
532
+ "bytes": 669079,
533
+ "pages": 20,
534
  "shards": 1
535
  },
536
  {
537
+ "seconds": 1.4896883629990043,
538
+ "bytes": 13505965,
539
+ "pages": 48,
540
+ "shards": 4
541
  },
542
  {
543
+ "seconds": 1.098302499998681,
544
+ "bytes": 6017424,
545
+ "pages": 23,
546
  "shards": 2
547
  },
548
  {
549
+ "seconds": 1.0887724580024951,
550
+ "bytes": 6603634,
551
+ "pages": 44,
552
  "shards": 3
553
  },
554
  {
555
+ "seconds": 1.2179049539990956,
556
+ "bytes": 6068321,
557
+ "pages": 36,
558
+ "shards": 2
559
  },
560
  {
561
+ "seconds": 2.3319496820004133,
562
+ "bytes": 5855101,
563
+ "pages": 27,
564
+ "shards": 2
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:00b52798ce4bbc979b3cc619c785cf5f2e3b8c46472773cb746b19a780d676ff
3
- size 721780479
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9182b42958ff73eae94225e7d961ee0aec9363f3f5f387fc8e3072607f9dc6fc
3
+ size 721764223
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,