appvoid commited on
Commit
ec63cea
·
verified ·
1 Parent(s): fe9915a

Qwen90M step 675000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +46 -46
  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": 670000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "3e61d5e203d0bc4da303d0c73f37e2c56ee4a56703224613ed6b408956f545c6",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "baf181694b42eed96c2f274abdcd70b231be691131559b6fae917122d444a98d",
12
- "training_config.json": "99d98cec9c0814e63e7b7a5836277d5c336d3c2fb33fa69da403f3610c902c99",
13
  "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf",
14
- "shard_scheduler.json": "7d1f5ece6f823a90937aac50d59b702f0bdefd7399ebea8da1b48bbc8a86ccdf",
15
- "coverage.json": "7cfc71bc0efed1fc9ac9186a78aa9beee721c934d777b8a3ea87a912a1cbb4df",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"
 
1
  {
2
+ "step": 675000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "74dc22051f83b4c8984329a3666317ee194d1224a37d93e4e1a34b0174c4bbf8",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "3557acc152a7dfbc059cc3cb10ac7d0ee8716631de59030b922ca4ad23b76a00",
12
+ "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb",
13
  "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf",
14
+ "shard_scheduler.json": "5cf47e165a37561b98deded2063e10e26d280446908c77d7d22b1dd0b6ca3f1e",
15
+ "coverage.json": "2965b7663d35170c3297939a7eeaa8e936903ecb9f013cfad5133eed9394b125",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 670000,
3
  "stats": {
4
- "climbmix/blocks": 24120000,
5
- "climbmix/targets": 40322063590,
6
- "climbmix/input_tokens": 40324618718,
7
- "ultra_qa/blocks": 9045000,
8
- "ultra_qa/targets": 14756386919,
9
- "ultra_qa/input_tokens": 14756415423,
10
- "cortex/blocks": 9045000,
11
- "cortex/targets": 343924602,
12
- "cortex/input_tokens": 14377667142,
13
- "ultra_style/blocks": 9045000,
14
- "ultra_style/targets": 14306999204,
15
- "ultra_style/input_tokens": 14307749781,
16
- "openmath/blocks": 6030000,
17
- "openmath/targets": 9317442826,
18
- "openmath/input_tokens": 9320546731,
19
- "rewrite6/blocks": 3015000,
20
- "rewrite6/targets": 4942621070,
21
- "rewrite6/input_tokens": 4948040293
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3303,
26
- "row": 479,
27
  "start": {
28
- "file": 159,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 73,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 14336,
37
- "documents": 13529567
38
  },
39
  "rewrite6": {
40
- "page": 29,
41
- "row": 1718,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 118784
47
  }
48
  },
49
  "epoch": 37,
50
- "offset": 0,
51
- "documents": 7720450
52
  },
53
  "openmath": {
54
- "page": 2064,
55
- "row": 1413,
56
  "start": {
57
  "file": 19,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 157,
61
- "row": 794
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 8455557
67
  },
68
  "ultra_qa": {
69
- "page": 892,
70
- "row": 2846,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
- "row": 16248
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 3656478
82
  },
83
  "ultra_style": {
84
- "page": 1362,
85
- "row": 1579,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 1,
91
- "row": 14660
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 5580331
97
  },
98
  "cortex": {
99
- "page": 175421,
100
- "row": 25,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 22453913
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 675000,
3
  "stats": {
4
+ "climbmix/blocks": 24300000,
5
+ "climbmix/targets": 40622544143,
6
+ "climbmix/input_tokens": 40625118554,
7
+ "ultra_qa/blocks": 9112500,
8
+ "ultra_qa/targets": 14866485854,
9
+ "ultra_qa/input_tokens": 14866514573,
10
+ "cortex/blocks": 9112500,
11
+ "cortex/targets": 346482518,
12
+ "cortex/input_tokens": 14484924008,
13
+ "ultra_style/blocks": 9112500,
14
+ "ultra_style/targets": 14413351865,
15
+ "ultra_style/input_tokens": 14414108313,
16
+ "openmath/blocks": 6075000,
17
+ "openmath/targets": 9387040123,
18
+ "openmath/input_tokens": 9390167294,
19
+ "rewrite6/blocks": 3037500,
20
+ "rewrite6/targets": 4979579505,
21
+ "rewrite6/input_tokens": 4985038925
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3327,
26
+ "row": 3996,
27
  "start": {
28
+ "file": 161,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 2,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 13631388
38
  },
39
  "rewrite6": {
40
+ "page": 43,
41
+ "row": 1704,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 176128
47
  }
48
  },
49
  "epoch": 37,
50
+ "offset": 2048,
51
+ "documents": 7777780
52
  },
53
  "openmath": {
54
+ "page": 2079,
55
+ "row": 3284,
56
  "start": {
57
  "file": 19,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 219,
61
+ "row": 234
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 8518868
67
  },
68
  "ultra_qa": {
69
+ "page": 899,
70
+ "row": 1468,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
+ "row": 44920
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 3683772
82
  },
83
  "ultra_style": {
84
+ "page": 1372,
85
+ "row": 3093,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 1,
91
+ "row": 55620
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 5622805
97
  },
98
  "cortex": {
99
+ "page": 176729,
100
+ "row": 46,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 22621358
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 670000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "3e61d5e203d0bc4da303d0c73f37e2c56ee4a56703224613ed6b408956f545c6", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "baf181694b42eed96c2f274abdcd70b231be691131559b6fae917122d444a98d", "training_config.json": "99d98cec9c0814e63e7b7a5836277d5c336d3c2fb33fa69da403f3610c902c99", "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf", "shard_scheduler.json": "7d1f5ece6f823a90937aac50d59b702f0bdefd7399ebea8da1b48bbc8a86ccdf", "coverage.json": "7cfc71bc0efed1fc9ac9186a78aa9beee721c934d777b8a3ea87a912a1cbb4df", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000670000"}
 
1
+ {"step": 675000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "74dc22051f83b4c8984329a3666317ee194d1224a37d93e4e1a34b0174c4bbf8", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "3557acc152a7dfbc059cc3cb10ac7d0ee8716631de59030b922ca4ad23b76a00", "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb", "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf", "shard_scheduler.json": "5cf47e165a37561b98deded2063e10e26d280446908c77d7d22b1dd0b6ca3f1e", "coverage.json": "2965b7663d35170c3297939a7eeaa8e936903ecb9f013cfad5133eed9394b125", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000675000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3e61d5e203d0bc4da303d0c73f37e2c56ee4a56703224613ed6b408956f545c6
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74dc22051f83b4c8984329a3666317ee194d1224a37d93e4e1a34b0174c4bbf8
3
  size 360622744
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 74846.34535141522,
26
- "required_pages_per_second_upper_estimate": 0.0051780438388588255
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
36
  "tokens_per_second": 8533.957213384612,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
- "required_token_positions_per_second": 9355.793168926903,
40
- "required_pages_per_second_upper_estimate": 0.00391707522865193
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
50
  "tokens_per_second": 380144.35761751566,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
- "required_token_positions_per_second": 18711.586337853805,
54
- "required_pages_per_second_upper_estimate": 0.003897129976459624
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
64
  "tokens_per_second": 43742.16556775435,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
- "required_token_positions_per_second": 28067.379506780704,
68
- "required_pages_per_second_upper_estimate": 0.0016825965450244628
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
78
  "tokens_per_second": 22018.86276523721,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
- "required_token_positions_per_second": 28067.379506780704,
82
- "required_pages_per_second_upper_estimate": 0.0021821153012969137
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 22,
95
- "required_token_positions_per_second": 28067.379506780704,
96
- "required_pages_per_second_upper_estimate": 0.36001820790883526
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": 2.558590814491007,
157
  "selected": {
158
- "prefetch_batches": 61,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.10544663335164728,
162
- "producer_p95_seconds": 0.27246575100070913,
163
- "gpu_update_seconds": 0.9850581167835997,
164
- "producer_batches_per_second": 9.483470151817588,
165
- "gpu_batches_per_second": 1.0151685296144641,
166
- "producer_headroom": 9.341769248322912,
167
  "sustainable": true,
168
- "buffer_seconds": 60.08854512379958,
169
  "buffer_drain_seconds": null,
170
- "shard_target_mib": 32,
171
- "upload_pages_per_commit": 165,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 666047.1250430234,
174
- "upload_bytes_per_second": 5380735.178534467,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 71.45910412534727,
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": 0.9111081320006633,
184
- "bytes": 972942,
185
- "pages": 31,
186
- "shards": 1
187
- },
188
- {
189
- "seconds": 1.5192261520023749,
190
- "bytes": 17166243,
191
- "pages": 34,
192
- "shards": 5
193
  },
194
  {
195
- "seconds": 0.9617387159996724,
196
- "bytes": 1097341,
197
- "pages": 34,
198
- "shards": 1
199
  },
200
  {
201
- "seconds": 1.1497930719997385,
202
- "bytes": 6229541,
203
- "pages": 36,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 1.2241392690011708,
208
- "bytes": 7522866,
209
- "pages": 31,
210
  "shards": 3
211
  },
212
  {
213
- "seconds": 1.1316358820004098,
214
- "bytes": 5807643,
215
- "pages": 41,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 1.0283135649988253,
220
- "bytes": 5734251,
221
  "pages": 28,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 1.5221587859996362,
226
- "bytes": 13580138,
227
- "pages": 41,
228
- "shards": 4
229
- },
230
- {
231
- "seconds": 1.401579060999211,
232
- "bytes": 925271,
233
- "pages": 28,
234
- "shards": 1
235
  },
236
  {
237
- "seconds": 1.7116284690018801,
238
- "bytes": 11540303,
239
- "pages": 37,
240
- "shards": 4
241
  },
242
  {
243
- "seconds": 0.9012360309970973,
244
- "bytes": 1247702,
245
- "pages": 35,
246
- "shards": 1
247
  },
248
  {
249
- "seconds": 1.5333594760013511,
250
- "bytes": 7425995,
251
- "pages": 38,
252
- "shards": 3
253
  },
254
  {
255
- "seconds": 1.141444616998342,
256
- "bytes": 5373163,
257
- "pages": 23,
258
  "shards": 2
259
  },
260
  {
261
- "seconds": 1.302073504000873,
262
- "bytes": 12612142,
263
- "pages": 45,
264
- "shards": 3
265
  },
266
  {
267
- "seconds": 1.2979291070005274,
268
- "bytes": 7799504,
269
- "pages": 35,
270
- "shards": 3
271
  },
272
  {
273
- "seconds": 0.9238773860015499,
274
- "bytes": 1050494,
275
- "pages": 29,
276
- "shards": 1
277
  },
278
  {
279
- "seconds": 1.0488793139993504,
280
- "bytes": 6030424,
281
- "pages": 31,
282
  "shards": 2
283
  },
284
  {
285
- "seconds": 1.2485899509993033,
286
- "bytes": 11733880,
287
- "pages": 45,
288
  "shards": 4
289
  },
290
  {
291
- "seconds": 1.0325014549998741,
292
- "bytes": 941600,
293
- "pages": 26,
294
- "shards": 1
295
  },
296
  {
297
- "seconds": 1.327213976001076,
298
- "bytes": 13471794,
299
- "pages": 42,
300
- "shards": 4
301
  },
302
  {
303
- "seconds": 1.1652102200023364,
304
- "bytes": 5705697,
305
- "pages": 29,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 1.1611206180023146,
310
- "bytes": 6127791,
311
  "pages": 36,
312
- "shards": 3
313
  },
314
  {
315
- "seconds": 1.3078043249988696,
316
- "bytes": 7396085,
317
- "pages": 30,
318
  "shards": 3
319
  },
320
  {
321
- "seconds": 0.9337214369988942,
322
- "bytes": 1171128,
323
  "pages": 35,
 
 
 
 
 
 
324
  "shards": 1
325
  },
326
  {
327
- "seconds": 1.1392940650002856,
328
- "bytes": 6258107,
329
- "pages": 31,
 
 
 
 
 
 
330
  "shards": 2
331
  },
332
  {
333
- "seconds": 1.18578564299969,
334
- "bytes": 11925039,
335
- "pages": 24,
336
- "shards": 4
 
 
 
 
 
 
337
  },
338
  {
339
- "seconds": 1.0414635330016608,
340
- "bytes": 5963906,
 
 
 
 
 
 
341
  "pages": 28,
342
  "shards": 2
343
  },
344
  {
345
- "seconds": 1.0295740060028038,
346
- "bytes": 1098382,
347
- "pages": 35,
348
- "shards": 1
349
  },
350
  {
351
- "seconds": 1.3453053759985778,
352
- "bytes": 7563007,
353
  "pages": 37,
354
- "shards": 3
355
  },
356
  {
357
- "seconds": 1.2517445409976062,
358
- "bytes": 10940783,
359
- "pages": 32,
360
  "shards": 3
361
  },
362
  {
363
- "seconds": 1.2907915830001002,
364
- "bytes": 2619734,
365
- "pages": 37,
366
  "shards": 2
367
  },
368
  {
369
- "seconds": 1.1900975380012824,
370
- "bytes": 11578816,
371
- "pages": 34,
372
  "shards": 3
373
  },
374
  {
375
- "seconds": 1.276093910997588,
376
- "bytes": 903821,
377
- "pages": 26,
378
- "shards": 1
379
  },
380
  {
381
- "seconds": 1.6412474520002434,
382
- "bytes": 11839721,
383
- "pages": 42,
384
- "shards": 4
385
  },
386
  {
387
- "seconds": 1.4279512529974454,
388
- "bytes": 5596025,
389
- "pages": 31,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 1.1274023559999478,
394
- "bytes": 1023034,
395
- "pages": 31,
396
- "shards": 1
397
  },
398
  {
399
- "seconds": 1.0909878259990364,
400
- "bytes": 7709457,
401
- "pages": 39,
402
- "shards": 3
403
  },
404
  {
405
- "seconds": 2.9601359040025272,
406
- "bytes": 15395500,
407
- "pages": 35,
408
  "shards": 4
409
  },
410
  {
411
- "seconds": 1.3234173230011947,
412
- "bytes": 2631414,
413
  "pages": 37,
414
  "shards": 2
415
  },
416
  {
417
- "seconds": 1.286528674998408,
418
- "bytes": 6199355,
419
- "pages": 28,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.3863541229984548,
424
- "bytes": 10474994,
425
- "pages": 40,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 1.2548714610020397,
430
- "bytes": 2378266,
431
- "pages": 29,
432
- "shards": 2
433
  },
434
  {
435
- "seconds": 1.0437052470006165,
436
- "bytes": 6102982,
437
- "pages": 37,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 1.201682383001753,
442
- "bytes": 8258960,
443
  "pages": 32,
444
  "shards": 3
445
  },
446
  {
447
- "seconds": 1.4597207000006165,
448
- "bytes": 10914765,
449
- "pages": 43,
450
- "shards": 3
451
- },
452
- {
453
- "seconds": 1.293525116998353,
454
- "bytes": 5848497,
455
- "pages": 28,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 1.0607272609995562,
460
- "bytes": 2667610,
461
  "pages": 38,
462
- "shards": 2
463
  },
464
  {
465
- "seconds": 1.5769093039998552,
466
- "bytes": 5962126,
467
- "pages": 29,
468
- "shards": 2
469
  },
470
  {
471
- "seconds": 1.0608517170003324,
472
- "bytes": 4607080,
473
- "pages": 31,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.4511087869977928,
478
- "bytes": 13183448,
479
- "pages": 39,
480
- "shards": 4
481
- },
482
- {
483
- "seconds": 1.3023088439986168,
484
- "bytes": 6313993,
485
- "pages": 36,
486
  "shards": 2
487
  },
488
  {
489
- "seconds": 1.0238163379981415,
490
- "bytes": 2362328,
491
- "pages": 29,
492
- "shards": 2
493
  },
494
  {
495
- "seconds": 1.3659524169997894,
496
- "bytes": 10315345,
497
  "pages": 40,
498
- "shards": 4
499
  },
500
  {
501
- "seconds": 1.0596161510002275,
502
- "bytes": 5840926,
503
- "pages": 28,
504
- "shards": 2
505
  },
506
  {
507
- "seconds": 0.9319052459977684,
508
- "bytes": 2836860,
509
- "pages": 38,
510
  "shards": 2
511
  },
512
  {
513
- "seconds": 1.3332925520007848,
514
- "bytes": 16563668,
515
- "pages": 36,
516
  "shards": 4
517
  },
518
  {
519
- "seconds": 1.482665261002694,
520
- "bytes": 2527538,
521
  "pages": 33,
522
- "shards": 2
523
  },
524
  {
525
- "seconds": 1.213385584000207,
526
- "bytes": 6122336,
527
  "pages": 33,
528
- "shards": 2
529
- },
530
- {
531
- "seconds": 1.3804073209976195,
532
- "bytes": 6371391,
533
- "pages": 37,
534
- "shards": 2
535
  },
536
  {
537
- "seconds": 1.1654375459984294,
538
- "bytes": 2648981,
539
- "pages": 36,
540
  "shards": 2
541
  },
542
  {
543
- "seconds": 1.4223995709980954,
544
- "bytes": 9800135,
545
- "pages": 33,
546
  "shards": 3
547
  },
548
  {
549
- "seconds": 1.239405350999732,
550
- "bytes": 13306132,
551
- "pages": 35,
552
- "shards": 4
553
  },
554
  {
555
- "seconds": 1.4411270890013839,
556
- "bytes": 1188235,
557
  "pages": 36,
558
  "shards": 1
559
  },
560
  {
561
- "seconds": 1.200595816000714,
562
- "bytes": 10503607,
563
- "pages": 33,
564
  "shards": 3
 
 
 
 
 
 
565
  }
566
  ],
567
  "cache_key": {
 
22
  "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 76357.51005666674,
26
+ "required_pages_per_second_upper_estimate": 0.005282589719553327
27
  },
28
  {
29
  "source": "rewrite6",
 
36
  "tokens_per_second": 8533.957213384612,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
+ "required_token_positions_per_second": 9544.688757083342,
40
+ "required_pages_per_second_upper_estimate": 0.003996161866824596
41
  },
42
  {
43
  "source": "openmath",
 
50
  "tokens_per_second": 380144.35761751566,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
+ "required_token_positions_per_second": 19089.377514166685,
54
+ "required_pages_per_second_upper_estimate": 0.0039758139154622936
55
  },
56
  {
57
  "source": "ultra_qa",
 
64
  "tokens_per_second": 43742.16556775435,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
+ "required_token_positions_per_second": 28634.066271250027,
68
+ "required_pages_per_second_upper_estimate": 0.0017165685512738108
69
  },
70
  {
71
  "source": "ultra_style",
 
78
  "tokens_per_second": 22018.86276523721,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
+ "required_token_positions_per_second": 28634.066271250027,
82
+ "required_pages_per_second_upper_estimate": 0.002226172704642752
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 23,
95
+ "required_token_positions_per_second": 28634.066271250027,
96
+ "required_pages_per_second_upper_estimate": 0.367287057262606
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
+ "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.08469342544537994,
162
+ "producer_p95_seconds": 0.2953102920000674,
163
+ "gpu_update_seconds": 0.965563177024561,
164
+ "producer_batches_per_second": 11.80729194434242,
165
+ "gpu_batches_per_second": 1.0356650127043556,
166
+ "producer_headroom": 11.400686321835773,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.83048015254734,
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": 331883.15068931517,
174
+ "upload_bytes_per_second": 5613899.857854061,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 71.21439829358665,
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": 1.3863541229984548,
184
+ "bytes": 10474994,
185
+ "pages": 40,
186
+ "shards": 3
 
 
 
 
 
 
187
  },
188
  {
189
+ "seconds": 1.2548714610020397,
190
+ "bytes": 2378266,
191
+ "pages": 29,
192
+ "shards": 2
193
  },
194
  {
195
+ "seconds": 1.0437052470006165,
196
+ "bytes": 6102982,
197
+ "pages": 37,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.201682383001753,
202
+ "bytes": 8258960,
203
+ "pages": 32,
204
  "shards": 3
205
  },
206
  {
207
+ "seconds": 1.4597207000006165,
208
+ "bytes": 10914765,
209
+ "pages": 43,
210
+ "shards": 3
211
  },
212
  {
213
+ "seconds": 1.293525116998353,
214
+ "bytes": 5848497,
215
  "pages": 28,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.0607272609995562,
220
+ "bytes": 2667610,
221
+ "pages": 38,
222
+ "shards": 2
 
 
 
 
 
 
223
  },
224
  {
225
+ "seconds": 1.5769093039998552,
226
+ "bytes": 5962126,
227
+ "pages": 29,
228
+ "shards": 2
229
  },
230
  {
231
+ "seconds": 1.0608517170003324,
232
+ "bytes": 4607080,
233
+ "pages": 31,
234
+ "shards": 2
235
  },
236
  {
237
+ "seconds": 1.4511087869977928,
238
+ "bytes": 13183448,
239
+ "pages": 39,
240
+ "shards": 4
241
  },
242
  {
243
+ "seconds": 1.3023088439986168,
244
+ "bytes": 6313993,
245
+ "pages": 36,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.0238163379981415,
250
+ "bytes": 2362328,
251
+ "pages": 29,
252
+ "shards": 2
253
  },
254
  {
255
+ "seconds": 1.3659524169997894,
256
+ "bytes": 10315345,
257
+ "pages": 40,
258
+ "shards": 4
259
  },
260
  {
261
+ "seconds": 1.0596161510002275,
262
+ "bytes": 5840926,
263
+ "pages": 28,
264
+ "shards": 2
265
  },
266
  {
267
+ "seconds": 0.9319052459977684,
268
+ "bytes": 2836860,
269
+ "pages": 38,
270
  "shards": 2
271
  },
272
  {
273
+ "seconds": 1.3332925520007848,
274
+ "bytes": 16563668,
275
+ "pages": 36,
276
  "shards": 4
277
  },
278
  {
279
+ "seconds": 1.482665261002694,
280
+ "bytes": 2527538,
281
+ "pages": 33,
282
+ "shards": 2
283
  },
284
  {
285
+ "seconds": 1.213385584000207,
286
+ "bytes": 6122336,
287
+ "pages": 33,
288
+ "shards": 2
289
  },
290
  {
291
+ "seconds": 1.3804073209976195,
292
+ "bytes": 6371391,
293
+ "pages": 37,
294
  "shards": 2
295
  },
296
  {
297
+ "seconds": 1.1654375459984294,
298
+ "bytes": 2648981,
299
  "pages": 36,
300
+ "shards": 2
301
  },
302
  {
303
+ "seconds": 1.4223995709980954,
304
+ "bytes": 9800135,
305
+ "pages": 33,
306
  "shards": 3
307
  },
308
  {
309
+ "seconds": 1.239405350999732,
310
+ "bytes": 13306132,
311
  "pages": 35,
312
+ "shards": 4
313
+ },
314
+ {
315
+ "seconds": 1.4411270890013839,
316
+ "bytes": 1188235,
317
+ "pages": 36,
318
  "shards": 1
319
  },
320
  {
321
+ "seconds": 1.200595816000714,
322
+ "bytes": 10503607,
323
+ "pages": 33,
324
+ "shards": 3
325
+ },
326
+ {
327
+ "seconds": 1.0193097230003332,
328
+ "bytes": 2594227,
329
+ "pages": 37,
330
  "shards": 2
331
  },
332
  {
333
+ "seconds": 1.1054332999992766,
334
+ "bytes": 5601825,
335
+ "pages": 19,
336
+ "shards": 2
337
+ },
338
+ {
339
+ "seconds": 1.0161830090000876,
340
+ "bytes": 1229232,
341
+ "pages": 36,
342
+ "shards": 1
343
  },
344
  {
345
+ "seconds": 1.4130291219989886,
346
+ "bytes": 17178237,
347
+ "pages": 40,
348
+ "shards": 5
349
+ },
350
+ {
351
+ "seconds": 1.2973765460010327,
352
+ "bytes": 5824332,
353
  "pages": 28,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 1.3192006099998252,
358
+ "bytes": 2683646,
359
+ "pages": 38,
360
+ "shards": 2
361
  },
362
  {
363
+ "seconds": 1.3172165699979814,
364
+ "bytes": 5876805,
365
  "pages": 37,
366
+ "shards": 2
367
  },
368
  {
369
+ "seconds": 1.2478048169978138,
370
+ "bytes": 10304832,
371
+ "pages": 34,
372
  "shards": 3
373
  },
374
  {
375
+ "seconds": 0.9087454849977803,
376
+ "bytes": 2381627,
377
+ "pages": 29,
378
  "shards": 2
379
  },
380
  {
381
+ "seconds": 1.198183002001315,
382
+ "bytes": 11915570,
383
+ "pages": 35,
384
  "shards": 3
385
  },
386
  {
387
+ "seconds": 1.2237985769970692,
388
+ "bytes": 7820030,
389
+ "pages": 40,
390
+ "shards": 3
391
  },
392
  {
393
+ "seconds": 1.4434572989994194,
394
+ "bytes": 5751342,
395
+ "pages": 32,
396
+ "shards": 2
397
  },
398
  {
399
+ "seconds": 1.5010855370019272,
400
+ "bytes": 6022741,
401
+ "pages": 36,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 1.0192723729996942,
406
+ "bytes": 2390933,
407
+ "pages": 30,
408
+ "shards": 2
409
  },
410
  {
411
+ "seconds": 1.0420342719990003,
412
+ "bytes": 6189478,
413
+ "pages": 38,
414
+ "shards": 2
415
  },
416
  {
417
+ "seconds": 1.2739770709995355,
418
+ "bytes": 15491147,
419
+ "pages": 33,
420
  "shards": 4
421
  },
422
  {
423
+ "seconds": 1.0505857130010554,
424
+ "bytes": 2636972,
425
  "pages": 37,
426
  "shards": 2
427
  },
428
  {
429
+ "seconds": 1.07342461799999,
430
+ "bytes": 5972281,
431
+ "pages": 29,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.2648427659987647,
436
+ "bytes": 7742666,
437
+ "pages": 39,
438
  "shards": 3
439
  },
440
  {
441
+ "seconds": 1.1688129319991276,
442
+ "bytes": 5401350,
443
+ "pages": 34,
444
+ "shards": 3
445
  },
446
  {
447
+ "seconds": 1.2462927969972952,
448
+ "bytes": 6148998,
449
+ "pages": 33,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.1124823669997568,
454
+ "bytes": 8182028,
455
  "pages": 32,
456
  "shards": 3
457
  },
458
  {
459
+ "seconds": 1.1143527310014179,
460
+ "bytes": 6164100,
461
+ "pages": 37,
 
 
 
 
 
 
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 1.3281419349987118,
466
+ "bytes": 11547797,
467
  "pages": 38,
468
+ "shards": 4
469
  },
470
  {
471
+ "seconds": 0.8457415819975722,
472
+ "bytes": 905737,
473
+ "pages": 27,
474
+ "shards": 1
475
  },
476
  {
477
+ "seconds": 1.6407027059976826,
478
+ "bytes": 6090920,
479
+ "pages": 36,
480
  "shards": 2
481
  },
482
  {
483
+ "seconds": 1.0669106919995102,
484
+ "bytes": 2526136,
485
+ "pages": 33,
 
 
 
 
 
 
486
  "shards": 2
487
  },
488
  {
489
+ "seconds": 1.477937416999339,
490
+ "bytes": 16111480,
491
+ "pages": 34,
492
+ "shards": 4
493
  },
494
  {
495
+ "seconds": 1.2388989300015965,
496
+ "bytes": 7565588,
497
  "pages": 40,
498
+ "shards": 3
499
  },
500
  {
501
+ "seconds": 0.9064518010018219,
502
+ "bytes": 1123131,
503
+ "pages": 35,
504
+ "shards": 1
505
  },
506
  {
507
+ "seconds": 1.0333158499997808,
508
+ "bytes": 5705052,
509
+ "pages": 24,
510
  "shards": 2
511
  },
512
  {
513
+ "seconds": 1.4537645320015145,
514
+ "bytes": 12446484,
515
+ "pages": 45,
516
  "shards": 4
517
  },
518
  {
519
+ "seconds": 1.027312460999383,
520
+ "bytes": 1095320,
521
  "pages": 33,
522
+ "shards": 1
523
  },
524
  {
525
+ "seconds": 1.1021007999988797,
526
+ "bytes": 11824886,
527
  "pages": 33,
528
+ "shards": 3
 
 
 
 
 
 
529
  },
530
  {
531
+ "seconds": 1.0116616440027428,
532
+ "bytes": 2398524,
533
+ "pages": 30,
534
  "shards": 2
535
  },
536
  {
537
+ "seconds": 1.1044605290007894,
538
+ "bytes": 9770514,
539
+ "pages": 37,
540
  "shards": 3
541
  },
542
  {
543
+ "seconds": 1.1610946500004502,
544
+ "bytes": 7427357,
545
+ "pages": 38,
546
+ "shards": 3
547
  },
548
  {
549
+ "seconds": 1.3267378080017806,
550
+ "bytes": 1199966,
551
  "pages": 36,
552
  "shards": 1
553
  },
554
  {
555
+ "seconds": 1.69540837599925,
556
+ "bytes": 9973657,
557
+ "pages": 31,
558
  "shards": 3
559
+ },
560
+ {
561
+ "seconds": 1.1572326720015553,
562
+ "bytes": 13287961,
563
+ "pages": 34,
564
+ "shards": 4
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:baf181694b42eed96c2f274abdcd70b231be691131559b6fae917122d444a98d
3
- size 721763775
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3557acc152a7dfbc059cc3cb10ac7d0ee8716631de59030b922ca4ad23b76a00
3
+ size 721778687
training_config.json CHANGED
@@ -53,8 +53,8 @@
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
- "prefetch_batches": 61,
57
- "prefetch_low_water": 30,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
60
  "force_rebenchmark": false,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 63,
57
+ "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
60
  "force_rebenchmark": false,