appvoid commited on
Commit
85d2e2c
·
verified ·
1 Parent(s): 7115c90

Qwen90M step 110000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +48 -48
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +246 -246
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 105000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "2e7b636c49fbc9ff9f3020bb49601a599167b2c73e0d7cc4e1006453e0b1f4bb",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "7c54110ffbfd70520fd96fac4faa05d1b2f4692f233c1cb56199f384dddec7e5",
12
- "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "7e827680f2cb27108bc66b58bb4d6ede7b8c07fab9033e664cac4bed5acb1a98",
14
- "shard_scheduler.json": "5bd2f77add256b6e8a0557483b02a1356ccca65bb941a3fd80d4f8230fce91c9",
15
- "coverage.json": "e1d90370de539b285245087c6c1203bfa12344b8fb425a1eaa35cdbbf5795812",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 110000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "771a846dfcaa9ac2239f4c06cbb98a3687803f55e50cb2bc31108644214bc872",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "847a0836b0889eaaab0298c4b432da9e790a6c0dff6648e498d3e21db336447a",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "7e827680f2cb27108bc66b58bb4d6ede7b8c07fab9033e664cac4bed5acb1a98",
14
+ "shard_scheduler.json": "c064c98867fe8d56cbdc4767e22ae652e692137319af0d6e6bccc2225e28dd1b",
15
+ "coverage.json": "2ca6a281fc319220e745c11a5bb944816bba2680cf6eeb45cee83e9353622a10",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 105000,
3
  "stats": {
4
- "climbmix/blocks": 3780000,
5
- "climbmix/targets": 6315746226,
6
- "climbmix/input_tokens": 6316146929,
7
- "ultra_qa/blocks": 1417500,
8
- "ultra_qa/targets": 2312418151,
9
- "ultra_qa/input_tokens": 2312422626,
10
- "cortex/blocks": 1417500,
11
- "cortex/targets": 53929281,
12
- "cortex/input_tokens": 2253084355,
13
- "ultra_style/blocks": 1417500,
14
- "ultra_style/targets": 2242078536,
15
- "ultra_style/input_tokens": 2242195826,
16
- "openmath/blocks": 945000,
17
- "openmath/targets": 1460177997,
18
- "openmath/input_tokens": 1460664808,
19
- "rewrite6/blocks": 472500,
20
- "rewrite6/targets": 774617263,
21
- "rewrite6/input_tokens": 775466653
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 518,
26
- "row": 1264,
27
  "start": {
28
- "file": 25,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 0,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 2122992
38
  },
39
  "rewrite6": {
40
- "page": 44,
41
- "row": 2788,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 180224
47
  }
48
  },
49
- "epoch": 5,
50
  "offset": 0,
51
- "documents": 1210032
52
  },
53
  "openmath": {
54
- "page": 323,
55
- "row": 2616,
56
  "start": {
57
  "file": 3,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 13,
61
- "row": 58
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 1325624
67
  },
68
  "ultra_qa": {
69
- "page": 139,
70
- "row": 3775,
71
  "start": {
72
  "file": 1,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 0,
76
- "row": 49717
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 0,
81
- "documents": 573119
82
  },
83
  "ultra_style": {
84
- "page": 213,
85
- "row": 1835,
86
  "start": {
87
  "file": 1,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 2,
91
- "row": 16825
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 0,
96
- "documents": 874283
97
  },
98
  "cortex": {
99
- "page": 27486,
100
- "row": 47,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 3518255
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 110000,
3
  "stats": {
4
+ "climbmix/blocks": 3960000,
5
+ "climbmix/targets": 6616550128,
6
+ "climbmix/input_tokens": 6616969711,
7
+ "ultra_qa/blocks": 1485000,
8
+ "ultra_qa/targets": 2422404971,
9
+ "ultra_qa/input_tokens": 2422409661,
10
+ "cortex/blocks": 1485000,
11
+ "cortex/targets": 56496990,
12
+ "cortex/input_tokens": 2360431026,
13
+ "ultra_style/blocks": 1485000,
14
+ "ultra_style/targets": 2348057065,
15
+ "ultra_style/input_tokens": 2348180586,
16
+ "openmath/blocks": 990000,
17
+ "openmath/targets": 1529649447,
18
+ "openmath/input_tokens": 1530159661,
19
+ "rewrite6/blocks": 495000,
20
+ "rewrite6/targets": 811489243,
21
+ "rewrite6/input_tokens": 812378858
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 542,
26
+ "row": 3742,
27
  "start": {
28
+ "file": 26,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 13,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 0,
37
+ "documents": 2223774
38
  },
39
  "rewrite6": {
40
+ "page": 8,
41
+ "row": 2154,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 32768
47
  }
48
  },
49
+ "epoch": 6,
50
  "offset": 0,
51
+ "documents": 1267346
52
  },
53
  "openmath": {
54
+ "page": 339,
55
+ "row": 371,
56
  "start": {
57
  "file": 3,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 78,
61
+ "row": 594
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 1388915
67
  },
68
  "ultra_qa": {
69
+ "page": 146,
70
+ "row": 2430,
71
  "start": {
72
  "file": 1,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 1,
76
+ "row": 13902
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 2048,
81
+ "documents": 600446
82
  },
83
  "ultra_style": {
84
+ "page": 224,
85
+ "row": 65,
86
  "start": {
87
  "file": 1,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 2,
91
+ "row": 61881
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 0,
96
+ "documents": 917569
97
  },
98
  "cortex": {
99
+ "page": 28796,
100
+ "row": 61,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 3685949
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 105000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "2e7b636c49fbc9ff9f3020bb49601a599167b2c73e0d7cc4e1006453e0b1f4bb", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "7c54110ffbfd70520fd96fac4faa05d1b2f4692f233c1cb56199f384dddec7e5", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "7e827680f2cb27108bc66b58bb4d6ede7b8c07fab9033e664cac4bed5acb1a98", "shard_scheduler.json": "5bd2f77add256b6e8a0557483b02a1356ccca65bb941a3fd80d4f8230fce91c9", "coverage.json": "e1d90370de539b285245087c6c1203bfa12344b8fb425a1eaa35cdbbf5795812", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000105000"}
 
1
+ {"step": 110000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "771a846dfcaa9ac2239f4c06cbb98a3687803f55e50cb2bc31108644214bc872", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "847a0836b0889eaaab0298c4b432da9e790a6c0dff6648e498d3e21db336447a", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "7e827680f2cb27108bc66b58bb4d6ede7b8c07fab9033e664cac4bed5acb1a98", "shard_scheduler.json": "c064c98867fe8d56cbdc4767e22ae652e692137319af0d6e6bccc2225e28dd1b", "coverage.json": "2ca6a281fc319220e745c11a5bb944816bba2680cf6eeb45cee83e9353622a10", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000110000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2e7b636c49fbc9ff9f3020bb49601a599167b2c73e0d7cc4e1006453e0b1f4bb
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:771a846dfcaa9ac2239f4c06cbb98a3687803f55e50cb2bc31108644214bc872
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": 74828.03381926876,
26
- "required_pages_per_second_upper_estimate": 0.006485904321536365
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": 9353.504227408595,
40
- "required_pages_per_second_upper_estimate": 0.0037312526836638725
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": 18707.00845481719,
54
- "required_pages_per_second_upper_estimate": 0.004327537173917825
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": 28060.512682225788,
68
- "required_pages_per_second_upper_estimate": 0.0017087380002154327
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": 28060.512682225788,
82
- "required_pages_per_second_upper_estimate": 0.002160175942748361
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": 21,
95
- "required_token_positions_per_second": 28060.512682225788,
96
- "required_pages_per_second_upper_estimate": 0.3484176550185106
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": 61,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
- "producer_batch_seconds": 0.09439860396102517,
162
- "producer_p95_seconds": 0.32911561600121786,
163
- "gpu_update_seconds": 0.9852991751470356,
164
- "producer_batches_per_second": 10.593376999652188,
165
- "gpu_batches_per_second": 1.0149201635643008,
166
- "producer_headroom": 10.437645619778879,
167
  "sustainable": true,
168
- "buffer_seconds": 60.10324968396917,
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": 488633.5744452333,
174
- "upload_bytes_per_second": 4102653.020833448,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 104.84939026849507,
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": 1.0569672790006734,
184
- "bytes": 5933537,
185
- "pages": 27,
186
  "shards": 2
187
  },
188
  {
189
- "seconds": 0.8821923329996935,
190
- "bytes": 961115,
191
- "pages": 26,
192
- "shards": 1
193
- },
194
- {
195
- "seconds": 1.2227748900004372,
196
- "bytes": 12637313,
197
- "pages": 46,
198
- "shards": 4
199
  },
200
  {
201
- "seconds": 1.1576985259998764,
202
- "bytes": 6035027,
203
- "pages": 35,
204
- "shards": 2
205
  },
206
  {
207
- "seconds": 1.131166884999402,
208
- "bytes": 11754261,
209
- "pages": 32,
210
  "shards": 3
211
  },
212
  {
213
- "seconds": 1.3580378289989312,
214
- "bytes": 2581856,
215
- "pages": 36,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 1.6505394759988121,
220
- "bytes": 9887447,
221
- "pages": 28,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 0.7642709539977659,
226
- "bytes": 2553358,
227
- "pages": 36,
228
  "shards": 2
229
  },
230
  {
231
- "seconds": 1.2285513930000889,
232
- "bytes": 6249359,
233
  "pages": 35,
234
- "shards": 2
235
- },
236
- {
237
- "seconds": 0.8377171489992179,
238
- "bytes": 5823773,
239
- "pages": 30,
240
- "shards": 2
241
  },
242
  {
243
- "seconds": 1.6930653950003034,
244
- "bytes": 12802437,
245
- "pages": 34,
246
  "shards": 4
247
  },
248
  {
249
- "seconds": 1.2719050730011077,
250
- "bytes": 5991225,
251
- "pages": 33,
252
- "shards": 2
253
  },
254
  {
255
- "seconds": 0.9874920929978543,
256
- "bytes": 1116876,
257
- "pages": 34,
258
- "shards": 1
259
  },
260
  {
261
- "seconds": 1.0614115569987916,
262
- "bytes": 7614945,
263
- "pages": 37,
264
  "shards": 3
265
  },
266
  {
267
- "seconds": 1.0503759480016015,
268
- "bytes": 9758074,
269
- "pages": 29,
270
  "shards": 3
271
  },
272
  {
273
- "seconds": 1.035596928002633,
274
- "bytes": 2562080,
275
  "pages": 36,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.3296433830000751,
280
- "bytes": 11732086,
281
- "pages": 34,
282
- "shards": 3
283
  },
284
  {
285
- "seconds": 0.8983965400002489,
286
- "bytes": 906815,
287
- "pages": 27,
288
  "shards": 1
289
  },
290
  {
291
- "seconds": 1.218643701999099,
292
- "bytes": 11866314,
293
- "pages": 38,
294
- "shards": 5
295
  },
296
  {
297
- "seconds": 1.0883352769997146,
298
- "bytes": 6129558,
299
- "pages": 35,
300
- "shards": 2
301
  },
302
  {
303
- "seconds": 1.048354777998611,
304
- "bytes": 2655724,
305
  "pages": 36,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 0.8924994240005617,
310
- "bytes": 5699054,
311
- "pages": 19,
312
  "shards": 2
313
  },
314
  {
315
- "seconds": 1.2686124620013288,
316
- "bytes": 11642195,
317
- "pages": 44,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 0.9798610010002449,
322
- "bytes": 7526317,
323
- "pages": 37,
324
- "shards": 3
325
  },
326
  {
327
- "seconds": 0.7459925380026107,
328
- "bytes": 508402,
329
- "pages": 17,
330
- "shards": 1
331
  },
332
  {
333
- "seconds": 0.9750640929996734,
334
- "bytes": 6281942,
335
  "pages": 35,
336
  "shards": 2
337
  },
338
  {
339
- "seconds": 1.0568676230031997,
340
- "bytes": 6659990,
341
- "pages": 30,
342
- "shards": 3
343
  },
344
  {
345
- "seconds": 1.1445812689999002,
346
- "bytes": 6216290,
347
- "pages": 36,
348
- "shards": 2
349
  },
350
  {
351
- "seconds": 1.002836310002749,
352
- "bytes": 6201464,
353
- "pages": 35,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.0725969490013085,
358
- "bytes": 8132406,
359
- "pages": 29,
360
- "shards": 3
361
  },
362
  {
363
- "seconds": 1.011032721002266,
364
- "bytes": 10718239,
365
- "pages": 39,
366
- "shards": 3
367
  },
368
  {
369
- "seconds": 1.0137605279996933,
370
- "bytes": 6019285,
371
- "pages": 27,
372
  "shards": 2
373
  },
374
  {
375
- "seconds": 0.9841479640017496,
376
- "bytes": 2601592,
377
  "pages": 36,
 
 
 
 
 
 
378
  "shards": 2
379
  },
380
  {
381
- "seconds": 1.1784665140003199,
382
- "bytes": 6252756,
383
  "pages": 35,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 0.9329704880001373,
388
- "bytes": 2137944,
389
- "pages": 23,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 1.5199417449985049,
394
- "bytes": 16335206,
395
- "pages": 47,
396
- "shards": 4
397
  },
398
  {
399
- "seconds": 0.9530525390000548,
400
- "bytes": 948458,
401
- "pages": 26,
402
- "shards": 1
403
  },
404
  {
405
- "seconds": 1.2729225670009328,
406
- "bytes": 7447736,
407
  "pages": 37,
408
- "shards": 3
409
  },
410
  {
411
- "seconds": 1.1847233869993943,
412
- "bytes": 10716114,
413
  "pages": 33,
414
  "shards": 3
415
  },
416
  {
417
- "seconds": 1.124874531000387,
418
- "bytes": 2649444,
419
- "pages": 36,
420
- "shards": 2
421
  },
422
  {
423
- "seconds": 0.9738625030004187,
424
- "bytes": 6071559,
425
- "pages": 35,
426
  "shards": 2
427
  },
428
  {
429
- "seconds": 1.5427607269994041,
430
- "bytes": 11501779,
431
- "pages": 29,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 1.0622286870020616,
436
- "bytes": 2500050,
437
- "pages": 30,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 1.1363001400022767,
442
- "bytes": 9937212,
443
  "pages": 37,
444
  "shards": 3
445
  },
446
  {
447
- "seconds": 0.7940713860007236,
448
- "bytes": 1067006,
449
- "pages": 34,
450
- "shards": 1
451
  },
452
  {
453
- "seconds": 0.9383102729989332,
454
- "bytes": 7322897,
455
- "pages": 29,
456
  "shards": 3
457
  },
458
  {
459
- "seconds": 1.1287910210012342,
460
- "bytes": 10964395,
461
- "pages": 38,
462
- "shards": 3
463
  },
464
  {
465
- "seconds": 1.271768777998659,
466
- "bytes": 8367852,
467
- "pages": 34,
468
  "shards": 3
469
  },
470
  {
471
- "seconds": 1.080213831999572,
472
- "bytes": 6038268,
473
- "pages": 31,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 0.7658507999985886,
478
- "bytes": 1224241,
479
- "pages": 35,
480
- "shards": 1
481
  },
482
  {
483
- "seconds": 34.26547425300305,
484
- "bytes": 7641108,
485
- "pages": 29,
486
- "shards": 4
487
  },
488
  {
489
- "seconds": 1.169081484000344,
490
- "bytes": 10839288,
491
- "pages": 47,
492
  "shards": 3
493
  },
494
  {
495
- "seconds": 1.171269268001197,
496
- "bytes": 2613392,
497
- "pages": 36,
498
- "shards": 2
499
  },
500
  {
501
- "seconds": 1.150122408002062,
502
- "bytes": 11429829,
503
- "pages": 30,
504
  "shards": 3
505
  },
506
  {
507
- "seconds": 1.3652685859997291,
508
- "bytes": 10381816,
509
- "pages": 32,
510
- "shards": 3
511
  },
512
  {
513
- "seconds": 1.0450683730014134,
514
- "bytes": 2703655,
515
- "pages": 36,
516
- "shards": 2
517
  },
518
  {
519
- "seconds": 0.9104454030020861,
520
- "bytes": 6024195,
521
- "pages": 36,
522
- "shards": 2
523
  },
524
  {
525
- "seconds": 1.20260026500182,
526
- "bytes": 1126978,
527
- "pages": 33,
528
  "shards": 1
529
  },
530
  {
531
- "seconds": 1.128464265002549,
532
- "bytes": 12438546,
533
- "pages": 34,
534
- "shards": 4
535
  },
536
  {
537
- "seconds": 1.0890558419996523,
538
- "bytes": 11718926,
539
- "pages": 32,
540
  "shards": 3
541
  },
542
  {
543
- "seconds": 0.9750485680015117,
544
- "bytes": 2678604,
545
- "pages": 36,
546
  "shards": 2
547
  },
548
  {
549
- "seconds": 0.9471035349997692,
550
- "bytes": 5896771,
551
- "pages": 27,
552
  "shards": 2
553
  },
554
  {
555
- "seconds": 0.970585907001805,
556
- "bytes": 1087871,
557
- "pages": 35,
558
  "shards": 1
559
  },
560
  {
561
- "seconds": 1.300397458999214,
562
- "bytes": 11506344,
563
- "pages": 35,
564
- "shards": 4
 
 
 
 
 
 
 
 
 
 
 
 
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": 75187.87582822153,
26
+ "required_pages_per_second_upper_estimate": 0.006517094514861157
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": 9398.484478527691,
40
+ "required_pages_per_second_upper_estimate": 0.0037491959783499643
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": 18796.968957055382,
54
+ "required_pages_per_second_upper_estimate": 0.0043483479528598655
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": 28195.453435583073,
68
+ "required_pages_per_second_upper_estimate": 0.001716955184115474
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": 28195.453435583073,
82
+ "required_pages_per_second_upper_estimate": 0.0021705640554817176
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": 22,
95
+ "required_token_positions_per_second": 28195.453435583073,
96
+ "required_pages_per_second_upper_estimate": 0.3500931675575583
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": 688946,
161
+ "producer_batch_seconds": 0.08555045376556336,
162
+ "producer_p95_seconds": 0.3201156809955137,
163
+ "gpu_update_seconds": 0.9805836271853611,
164
+ "producer_batches_per_second": 11.689008719232886,
165
+ "gpu_batches_per_second": 1.0198008331735775,
166
+ "producer_headroom": 11.462050568106696,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.79618488549239,
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": 390327.88351984415,
174
+ "upload_bytes_per_second": 4067954.1604864304,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 83.75527743526929,
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": 1.0622286870020616,
184
+ "bytes": 2500050,
185
+ "pages": 30,
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 1.1363001400022767,
190
+ "bytes": 9937212,
191
+ "pages": 37,
192
+ "shards": 3
 
 
 
 
 
 
193
  },
194
  {
195
+ "seconds": 0.7940713860007236,
196
+ "bytes": 1067006,
197
+ "pages": 34,
198
+ "shards": 1
199
  },
200
  {
201
+ "seconds": 0.9383102729989332,
202
+ "bytes": 7322897,
203
+ "pages": 29,
204
  "shards": 3
205
  },
206
  {
207
+ "seconds": 1.1287910210012342,
208
+ "bytes": 10964395,
209
+ "pages": 38,
210
+ "shards": 3
211
  },
212
  {
213
+ "seconds": 1.271768777998659,
214
+ "bytes": 8367852,
215
+ "pages": 34,
216
  "shards": 3
217
  },
218
  {
219
+ "seconds": 1.080213831999572,
220
+ "bytes": 6038268,
221
+ "pages": 31,
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 0.7658507999985886,
226
+ "bytes": 1224241,
227
  "pages": 35,
228
+ "shards": 1
 
 
 
 
 
 
229
  },
230
  {
231
+ "seconds": 34.26547425300305,
232
+ "bytes": 7641108,
233
+ "pages": 29,
234
  "shards": 4
235
  },
236
  {
237
+ "seconds": 1.169081484000344,
238
+ "bytes": 10839288,
239
+ "pages": 47,
240
+ "shards": 3
241
  },
242
  {
243
+ "seconds": 1.171269268001197,
244
+ "bytes": 2613392,
245
+ "pages": 36,
246
+ "shards": 2
247
  },
248
  {
249
+ "seconds": 1.150122408002062,
250
+ "bytes": 11429829,
251
+ "pages": 30,
252
  "shards": 3
253
  },
254
  {
255
+ "seconds": 1.3652685859997291,
256
+ "bytes": 10381816,
257
+ "pages": 32,
258
  "shards": 3
259
  },
260
  {
261
+ "seconds": 1.0450683730014134,
262
+ "bytes": 2703655,
263
  "pages": 36,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 0.9104454030020861,
268
+ "bytes": 6024195,
269
+ "pages": 36,
270
+ "shards": 2
271
  },
272
  {
273
+ "seconds": 1.20260026500182,
274
+ "bytes": 1126978,
275
+ "pages": 33,
276
  "shards": 1
277
  },
278
  {
279
+ "seconds": 1.128464265002549,
280
+ "bytes": 12438546,
281
+ "pages": 34,
282
+ "shards": 4
283
  },
284
  {
285
+ "seconds": 1.0890558419996523,
286
+ "bytes": 11718926,
287
+ "pages": 32,
288
+ "shards": 3
289
  },
290
  {
291
+ "seconds": 0.9750485680015117,
292
+ "bytes": 2678604,
293
  "pages": 36,
294
  "shards": 2
295
  },
296
  {
297
+ "seconds": 0.9471035349997692,
298
+ "bytes": 5896771,
299
+ "pages": 27,
300
  "shards": 2
301
  },
302
  {
303
+ "seconds": 0.970585907001805,
304
+ "bytes": 1087871,
305
+ "pages": 35,
306
+ "shards": 1
 
 
 
 
 
 
307
  },
308
  {
309
+ "seconds": 1.300397458999214,
310
+ "bytes": 11506344,
311
+ "pages": 35,
312
+ "shards": 4
313
  },
314
  {
315
+ "seconds": 1.0356865150024532,
316
+ "bytes": 6075433,
317
  "pages": 35,
318
  "shards": 2
319
  },
320
  {
321
+ "seconds": 0.8513431389983452,
322
+ "bytes": 631745,
323
+ "pages": 18,
324
+ "shards": 1
325
  },
326
  {
327
+ "seconds": 1.1167246349978086,
328
+ "bytes": 13461458,
329
+ "pages": 33,
330
+ "shards": 4
331
  },
332
  {
333
+ "seconds": 1.146249319997878,
334
+ "bytes": 4617912,
335
+ "pages": 32,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 0.8623779709996597,
340
+ "bytes": 6286843,
341
+ "pages": 36,
342
+ "shards": 2
343
  },
344
  {
345
+ "seconds": 1.446577980001166,
346
+ "bytes": 2379048,
347
+ "pages": 30,
348
+ "shards": 2
349
  },
350
  {
351
+ "seconds": 1.0237194600013027,
352
+ "bytes": 6170294,
353
+ "pages": 34,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 1.2663461060001282,
358
+ "bytes": 11989707,
359
  "pages": 36,
360
+ "shards": 4
361
+ },
362
+ {
363
+ "seconds": 0.9998438940019696,
364
+ "bytes": 6675780,
365
+ "pages": 28,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.0791062770003919,
370
+ "bytes": 6176898,
371
  "pages": 35,
372
  "shards": 2
373
  },
374
  {
375
+ "seconds": 0.9380000849996577,
376
+ "bytes": 2661456,
377
+ "pages": 37,
378
  "shards": 2
379
  },
380
  {
381
+ "seconds": 1.1787855490001675,
382
+ "bytes": 9821385,
383
+ "pages": 34,
384
+ "shards": 3
385
  },
386
  {
387
+ "seconds": 1.015233548998367,
388
+ "bytes": 5854223,
389
+ "pages": 28,
390
+ "shards": 2
391
  },
392
  {
393
+ "seconds": 0.8780484679991787,
394
+ "bytes": 2637899,
395
  "pages": 37,
396
+ "shards": 2
397
  },
398
  {
399
+ "seconds": 1.1060435590006819,
400
+ "bytes": 11753109,
401
  "pages": 33,
402
  "shards": 3
403
  },
404
  {
405
+ "seconds": 1.0484200940009032,
406
+ "bytes": 6644677,
407
+ "pages": 34,
408
+ "shards": 3
409
  },
410
  {
411
+ "seconds": 0.9731116549992294,
412
+ "bytes": 6198048,
413
+ "pages": 36,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 1.2161200219998136,
418
+ "bytes": 5847296,
419
+ "pages": 28,
420
  "shards": 3
421
  },
422
  {
423
+ "seconds": 1.0819334130028437,
424
+ "bytes": 2586941,
425
+ "pages": 33,
426
  "shards": 2
427
  },
428
  {
429
+ "seconds": 1.1623649149987614,
430
+ "bytes": 11298342,
431
  "pages": 37,
432
  "shards": 3
433
  },
434
  {
435
+ "seconds": 0.995095141002821,
436
+ "bytes": 6689673,
437
+ "pages": 30,
438
+ "shards": 2
439
  },
440
  {
441
+ "seconds": 0.9397628990009252,
442
+ "bytes": 7665294,
443
+ "pages": 38,
444
  "shards": 3
445
  },
446
  {
447
+ "seconds": 1.2407093820002046,
448
+ "bytes": 5771889,
449
+ "pages": 29,
450
+ "shards": 2
451
  },
452
  {
453
+ "seconds": 1.0350117630005116,
454
+ "bytes": 6519406,
455
+ "pages": 35,
456
  "shards": 3
457
  },
458
  {
459
+ "seconds": 1.0306341840005189,
460
+ "bytes": 6113692,
461
+ "pages": 36,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 1.1684751539978606,
466
+ "bytes": 6202267,
467
+ "pages": 36,
468
+ "shards": 2
469
  },
470
  {
471
+ "seconds": 1.093655648001004,
472
+ "bytes": 8128170,
473
+ "pages": 33,
474
+ "shards": 3
475
  },
476
  {
477
+ "seconds": 1.1689348799991421,
478
+ "bytes": 10022604,
479
+ "pages": 37,
480
  "shards": 3
481
  },
482
  {
483
+ "seconds": 0.8061036489998514,
484
+ "bytes": 878150,
485
+ "pages": 26,
486
+ "shards": 1
487
  },
488
  {
489
+ "seconds": 1.0866164300023229,
490
+ "bytes": 7550162,
491
+ "pages": 38,
492
  "shards": 3
493
  },
494
  {
495
+ "seconds": 1.2529126609988452,
496
+ "bytes": 5669267,
497
+ "pages": 30,
498
+ "shards": 2
499
  },
500
  {
501
+ "seconds": 0.9512047030002577,
502
+ "bytes": 7497853,
503
+ "pages": 35,
504
+ "shards": 3
505
  },
506
  {
507
+ "seconds": 1.1582007809993229,
508
+ "bytes": 12016223,
509
+ "pages": 34,
510
+ "shards": 3
511
  },
512
  {
513
+ "seconds": 0.9243304510018788,
514
+ "bytes": 1103405,
515
+ "pages": 34,
516
  "shards": 1
517
  },
518
  {
519
+ "seconds": 1.4641582209987973,
520
+ "bytes": 7254109,
521
+ "pages": 30,
522
+ "shards": 3
523
  },
524
  {
525
+ "seconds": 1.1067068009979266,
526
+ "bytes": 10335265,
527
+ "pages": 39,
528
  "shards": 3
529
  },
530
  {
531
+ "seconds": 1.2287768099959067,
532
+ "bytes": 2576063,
533
+ "pages": 32,
534
  "shards": 2
535
  },
536
  {
537
+ "seconds": 1.1285306259960635,
538
+ "bytes": 5981374,
539
+ "pages": 32,
540
  "shards": 2
541
  },
542
  {
543
+ "seconds": 0.8470438740041573,
544
+ "bytes": 878239,
545
+ "pages": 27,
546
  "shards": 1
547
  },
548
  {
549
+ "seconds": 1.8046116949990392,
550
+ "bytes": 17147881,
551
+ "pages": 44,
552
+ "shards": 5
553
+ },
554
+ {
555
+ "seconds": 0.9106033309944905,
556
+ "bytes": 774779,
557
+ "pages": 26,
558
+ "shards": 1
559
+ },
560
+ {
561
+ "seconds": 1.1979517419968033,
562
+ "bytes": 6268819,
563
+ "pages": 37,
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 sha256:7c54110ffbfd70520fd96fac4faa05d1b2f4692f233c1cb56199f384dddec7e5
3
- size 721776575
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:847a0836b0889eaaab0298c4b432da9e790a6c0dff6648e498d3e21db336447a
3
+ size 721776063
training_config.json CHANGED
@@ -46,8 +46,8 @@
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,
 
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,