appvoid commited on
Commit
b5ffa34
·
verified ·
1 Parent(s): c2aa893

Qwen90M step 865000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +49 -49
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +246 -246
  6. trainer.pt +2 -2
  7. training_config.json +1 -1
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 860000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "76d2d488ad46202033ce64137cf12ae9086c284d7e8060eb3f80363676d0e8d0",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "892eb6396b64cb4fd594382fdc135dcfdab02c54a38e9994701368b84bd73dcd",
12
- "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
  "performance.json": "172134497605e50805257548c05018674e515323c63930c6237cd46d43786367",
14
- "shard_scheduler.json": "9a29ea4bdc9cefd2e92ebd24e57d2572330e585a6e74a99d3c512c6e2de2f678",
15
- "coverage.json": "7fba50c91c135d7273f46b84ac1af03e10b40c0850608fe8d645d5160ea5fc41",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
 
1
  {
2
+ "step": 865000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "f24ee40d83e4992cf8ce8af69aca975058968fa55d27163a7767a086b92a8387",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "18178f6d7c97fbaf7be1e689601dba5e41d5acf35448a4af6e2587866720a8d9",
12
+ "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58",
13
  "performance.json": "172134497605e50805257548c05018674e515323c63930c6237cd46d43786367",
14
+ "shard_scheduler.json": "c95b4d546c5df7ee17ed03b0bd9cd6934d744393d9e9e9817ef0f64a430ddf94",
15
+ "coverage.json": "3109f65091eb186c8eaa80402ee04bc03abb10fd556167b2666aaef70ee57e60",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 860000,
3
  "stats": {
4
- "climbmix/blocks": 30960000,
5
- "climbmix/targets": 51754676559,
6
- "climbmix/input_tokens": 51757956002,
7
- "ultra_qa/blocks": 11610000,
8
- "ultra_qa/targets": 18941056864,
9
- "ultra_qa/input_tokens": 18941093290,
10
- "cortex/blocks": 11610000,
11
- "cortex/targets": 441483304,
12
- "cortex/input_tokens": 18454886279,
13
- "ultra_style/blocks": 11610000,
14
- "ultra_style/targets": 18363767524,
15
- "ultra_style/input_tokens": 18364730231,
16
- "openmath/blocks": 7740000,
17
- "openmath/targets": 11959461269,
18
- "openmath/input_tokens": 11963446369,
19
- "rewrite6/blocks": 3870000,
20
- "rewrite6/targets": 6344361516,
21
- "rewrite6/input_tokens": 6351317040
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 4240,
26
- "row": 396,
27
  "start": {
28
- "file": 205,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 20,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 0,
37
- "documents": 17367436
38
  },
39
  "rewrite6": {
40
- "page": 12,
41
- "row": 642,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 49152
47
  }
48
  },
49
  "epoch": 48,
50
  "offset": 2048,
51
- "documents": 9909186
52
  },
53
  "openmath": {
54
- "page": 2650,
55
- "row": 161,
56
  "start": {
57
  "file": 24,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 374,
61
- "row": 801
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 0,
66
- "documents": 10854561
67
  },
68
  "ultra_qa": {
69
- "page": 1145,
70
- "row": 3108,
71
  "start": {
72
  "file": 9,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
- "row": 13278
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 4096,
81
- "documents": 4693028
82
  },
83
  "ultra_style": {
84
- "page": 1748,
85
- "row": 2510,
86
  "start": {
87
  "file": 10,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 3,
91
- "row": 55900
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 2048,
96
- "documents": 7162318
97
  },
98
  "cortex": {
99
- "page": 225159,
100
- "row": 87,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 28820439
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 865000,
3
  "stats": {
4
+ "climbmix/blocks": 31140000,
5
+ "climbmix/targets": 52055877655,
6
+ "climbmix/input_tokens": 52059176107,
7
+ "ultra_qa/blocks": 11677500,
8
+ "ultra_qa/targets": 19051109408,
9
+ "ultra_qa/input_tokens": 19051146044,
10
+ "cortex/blocks": 11677500,
11
+ "cortex/targets": 444043754,
12
+ "cortex/input_tokens": 18562155884,
13
+ "ultra_style/blocks": 11677500,
14
+ "ultra_style/targets": 18470831928,
15
+ "ultra_style/input_tokens": 18471800000,
16
+ "openmath/blocks": 7785000,
17
+ "openmath/targets": 12029146241,
18
+ "openmath/input_tokens": 12033154511,
19
+ "rewrite6/blocks": 3892500,
20
+ "rewrite6/targets": 6381217228,
21
+ "rewrite6/input_tokens": 6388213504
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 4264,
26
+ "row": 2484,
27
  "start": {
28
+ "file": 206,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 33,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 17467828
38
  },
39
  "rewrite6": {
40
+ "page": 26,
41
+ "row": 1330,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 106496
47
  }
48
  },
49
  "epoch": 48,
50
  "offset": 2048,
51
+ "documents": 9967218
52
  },
53
  "openmath": {
54
+ "page": 2665,
55
+ "row": 1871,
56
  "start": {
57
  "file": 24,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 436,
61
+ "row": 241
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 2048,
66
+ "documents": 10917711
67
  },
68
  "ultra_qa": {
69
+ "page": 1152,
70
+ "row": 1712,
71
  "start": {
72
  "file": 9,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
+ "row": 41950
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 0,
81
+ "documents": 4720304
82
  },
83
  "ultra_style": {
84
+ "page": 1758,
85
+ "row": 2694,
86
  "start": {
87
  "file": 10,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 4,
91
+ "row": 12042
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 0,
96
+ "documents": 7203462
97
  },
98
  "cortex": {
99
+ "page": 226468,
100
+ "row": 78,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 28987982
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 860000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "76d2d488ad46202033ce64137cf12ae9086c284d7e8060eb3f80363676d0e8d0", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "892eb6396b64cb4fd594382fdc135dcfdab02c54a38e9994701368b84bd73dcd", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "172134497605e50805257548c05018674e515323c63930c6237cd46d43786367", "shard_scheduler.json": "9a29ea4bdc9cefd2e92ebd24e57d2572330e585a6e74a99d3c512c6e2de2f678", "coverage.json": "7fba50c91c135d7273f46b84ac1af03e10b40c0850608fe8d645d5160ea5fc41", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000860000"}
 
1
+ {"step": 865000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "f24ee40d83e4992cf8ce8af69aca975058968fa55d27163a7767a086b92a8387", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "18178f6d7c97fbaf7be1e689601dba5e41d5acf35448a4af6e2587866720a8d9", "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58", "performance.json": "172134497605e50805257548c05018674e515323c63930c6237cd46d43786367", "shard_scheduler.json": "c95b4d546c5df7ee17ed03b0bd9cd6934d744393d9e9e9817ef0f64a430ddf94", "coverage.json": "3109f65091eb186c8eaa80402ee04bc03abb10fd556167b2666aaef70ee57e60", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000865000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:76d2d488ad46202033ce64137cf12ae9086c284d7e8060eb3f80363676d0e8d0
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f24ee40d83e4992cf8ce8af69aca975058968fa55d27163a7767a086b92a8387
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": 76235.28940132579,
26
- "required_pages_per_second_upper_estimate": 0.0052741342110258485
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": 9529.411175165724,
40
- "required_pages_per_second_upper_estimate": 0.003989765462307878
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": 19058.822350331448,
54
- "required_pages_per_second_upper_estimate": 0.003969450080629271
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": 28588.23352549717,
68
- "required_pages_per_second_upper_estimate": 0.0017138209481484775
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": 28588.23352549717,
82
- "required_pages_per_second_upper_estimate": 0.00222260941025741
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 23,
95
- "required_token_positions_per_second": 28588.23352549717,
96
- "required_pages_per_second_upper_estimate": 0.36669916401145664
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": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.06858718393758068,
162
- "producer_p95_seconds": 0.22669146000043838,
163
- "gpu_update_seconds": 0.9671111709417584,
164
- "producer_batches_per_second": 14.579983352430284,
165
- "gpu_batches_per_second": 1.0340072889719751,
166
- "producer_headroom": 14.100464772280196,
167
  "sustainable": true,
168
- "buffer_seconds": 60.92800376933078,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 84,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 389341.2469068264,
174
- "upload_bytes_per_second": 3447999.215230209,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 83.54356818584054,
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": 1.182235641999796,
184
- "bytes": 6603515,
185
- "pages": 27,
186
  "shards": 2
187
  },
188
  {
189
- "seconds": 1.285518773000149,
190
- "bytes": 6002809,
191
- "pages": 36,
192
- "shards": 2
193
  },
194
  {
195
- "seconds": 1.7619018259993027,
196
- "bytes": 7674402,
197
- "pages": 38,
198
  "shards": 3
199
  },
200
  {
201
- "seconds": 1.6144984039983683,
202
- "bytes": 4728422,
203
- "pages": 30,
204
- "shards": 2
205
  },
206
  {
207
- "seconds": 1.6315553869990254,
208
- "bytes": 5703905,
209
- "pages": 24,
210
  "shards": 2
211
  },
212
  {
213
- "seconds": 1.11825536699871,
214
- "bytes": 2656311,
215
- "pages": 35,
216
  "shards": 2
217
  },
218
  {
219
- "seconds": 1.9071076060008636,
220
- "bytes": 12136915,
221
  "pages": 37,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 1.7720820229988021,
226
- "bytes": 11800483,
227
- "pages": 32,
228
- "shards": 4
229
- },
230
- {
231
- "seconds": 1.0689491579996684,
232
- "bytes": 1154476,
233
- "pages": 34,
234
- "shards": 1
235
  },
236
  {
237
- "seconds": 1.7027444539999124,
238
- "bytes": 6185388,
239
- "pages": 36,
240
- "shards": 2
241
  },
242
  {
243
- "seconds": 1.5859740610012523,
244
- "bytes": 2563401,
245
- "pages": 35,
246
  "shards": 2
247
  },
248
  {
249
- "seconds": 1.2778493570003775,
250
- "bytes": 10237665,
251
- "pages": 33,
252
  "shards": 3
253
  },
254
  {
255
- "seconds": 2.496101438000551,
256
- "bytes": 13297971,
257
- "pages": 33,
258
- "shards": 4
259
  },
260
  {
261
- "seconds": 1.2942262079995999,
262
- "bytes": 1101377,
263
- "pages": 34,
264
- "shards": 1
265
  },
266
  {
267
- "seconds": 1.815276637998977,
268
- "bytes": 10307802,
269
- "pages": 39,
270
  "shards": 3
271
  },
272
  {
273
- "seconds": 1.1417645319997973,
274
- "bytes": 2370286,
275
- "pages": 28,
276
- "shards": 2
277
  },
278
  {
279
- "seconds": 1.1171746460004215,
280
- "bytes": 6111968,
281
  "pages": 37,
282
  "shards": 2
283
  },
284
  {
285
- "seconds": 1.2683859379994828,
286
- "bytes": 7618531,
287
- "pages": 38,
288
- "shards": 3
289
  },
290
  {
291
- "seconds": 1.7501807870012271,
292
- "bytes": 11262011,
293
- "pages": 28,
294
  "shards": 3
295
  },
296
  {
297
- "seconds": 1.2981046400000196,
298
- "bytes": 6199716,
299
  "pages": 36,
300
  "shards": 2
301
  },
302
  {
303
- "seconds": 1.6966266469989932,
304
- "bytes": 7696834,
305
- "pages": 38,
306
- "shards": 3
307
  },
308
  {
309
- "seconds": 1.3060271180002019,
310
- "bytes": 1241981,
311
- "pages": 35,
312
- "shards": 1
313
  },
314
  {
315
- "seconds": 1.5251753980010108,
316
- "bytes": 9924170,
317
- "pages": 32,
 
 
 
 
 
 
318
  "shards": 3
319
  },
320
  {
321
- "seconds": 1.1737768090006284,
322
- "bytes": 7614180,
323
  "pages": 39,
324
  "shards": 3
325
  },
326
  {
327
- "seconds": 1.2979090160006308,
328
- "bytes": 6818365,
329
- "pages": 30,
330
- "shards": 2
331
  },
332
  {
333
- "seconds": 0.9478092970002763,
334
- "bytes": 253929,
335
- "pages": 8,
336
  "shards": 1
337
  },
338
  {
339
- "seconds": 1.6955216110000038,
340
- "bytes": 7314599,
341
- "pages": 30,
342
- "shards": 3
343
  },
344
  {
345
- "seconds": 1.2749757900000986,
346
- "bytes": 5237846,
347
- "pages": 31,
348
- "shards": 2
349
  },
350
  {
351
- "seconds": 1.5798416210000141,
352
- "bytes": 6237294,
353
- "pages": 36,
354
- "shards": 2
355
  },
356
  {
357
- "seconds": 1.5424033700001019,
358
- "bytes": 7705972,
359
- "pages": 39,
360
  "shards": 3
361
  },
362
  {
363
- "seconds": 0.9795733470000414,
364
- "bytes": 923741,
365
- "pages": 28,
366
- "shards": 2
367
  },
368
  {
369
- "seconds": 1.5211821389984834,
370
- "bytes": 18268501,
371
- "pages": 39,
372
- "shards": 5
373
  },
374
  {
375
- "seconds": 1.3536671069996373,
376
- "bytes": 6151717,
377
  "pages": 36,
378
- "shards": 2
379
  },
380
  {
381
- "seconds": 1.1601749330002349,
382
- "bytes": 883373,
383
- "pages": 25,
384
  "shards": 1
385
  },
386
  {
387
- "seconds": 1.1801181450009608,
388
- "bytes": 7664644,
389
  "pages": 38,
390
  "shards": 3
391
  },
392
  {
393
- "seconds": 1.0472080479994474,
394
- "bytes": 5473896,
395
- "pages": 33,
396
- "shards": 2
397
- },
398
- {
399
- "seconds": 1.548700262001148,
400
- "bytes": 13538856,
401
- "pages": 41,
402
- "shards": 4
403
  },
404
  {
405
- "seconds": 1.3191421729989088,
406
- "bytes": 5997715,
407
- "pages": 32,
408
- "shards": 2
409
  },
410
  {
411
- "seconds": 1.765130539000893,
412
- "bytes": 937778,
413
  "pages": 29,
414
- "shards": 1
415
  },
416
  {
417
- "seconds": 2.7900099470007262,
418
- "bytes": 11587181,
419
- "pages": 37,
420
  "shards": 4
421
  },
422
  {
423
- "seconds": 0.987754063000466,
424
- "bytes": 1220161,
425
  "pages": 35,
426
  "shards": 1
427
  },
428
  {
429
- "seconds": 1.716705255999841,
430
- "bytes": 6235687,
431
- "pages": 36,
432
- "shards": 2
433
- },
434
- {
435
- "seconds": 33.88602663399979,
436
- "bytes": 7355362,
437
- "pages": 31,
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 1.5573600560001069,
442
- "bytes": 11648033,
443
- "pages": 46,
444
- "shards": 3
445
  },
446
  {
447
- "seconds": 1.2928704839996499,
448
- "bytes": 7315946,
449
- "pages": 39,
450
- "shards": 3
451
  },
452
  {
453
- "seconds": 1.4081739580014982,
454
- "bytes": 5891309,
455
- "pages": 28,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 1.286470338000072,
460
- "bytes": 2778002,
461
- "pages": 37,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.1679330120005034,
466
- "bytes": 10820666,
467
- "pages": 37,
468
  "shards": 3
469
  },
470
  {
471
- "seconds": 1.4827527400011604,
472
- "bytes": 6647922,
473
- "pages": 30,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.248109399999521,
478
- "bytes": 7422482,
479
- "pages": 37,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 1.451942487999986,
484
- "bytes": 6077412,
485
- "pages": 30,
486
  "shards": 2
487
  },
488
  {
489
- "seconds": 1.4631003799986502,
490
- "bytes": 7075109,
491
- "pages": 43,
492
- "shards": 3
493
  },
494
  {
495
- "seconds": 2.129564337999909,
496
- "bytes": 5884538,
497
- "pages": 31,
498
- "shards": 2
499
  },
500
  {
501
- "seconds": 1.1728028199995606,
502
- "bytes": 6147166,
503
- "pages": 33,
 
 
 
 
 
 
504
  "shards": 2
505
  },
506
  {
507
- "seconds": 1.2317090710002958,
508
- "bytes": 8202394,
509
- "pages": 32,
510
- "shards": 3
511
  },
512
  {
513
- "seconds": 1.6838970150001842,
514
- "bytes": 9992623,
515
- "pages": 38,
516
  "shards": 3
517
  },
518
  {
519
- "seconds": 1.2029927740004496,
520
- "bytes": 6068424,
521
- "pages": 37,
522
- "shards": 2
523
  },
524
  {
525
- "seconds": 1.1870869580016006,
526
- "bytes": 2666964,
527
- "pages": 36,
528
  "shards": 2
529
  },
530
  {
531
- "seconds": 1.4462591800001974,
532
- "bytes": 10549433,
533
- "pages": 31,
534
  "shards": 3
535
  },
536
  {
537
- "seconds": 1.0944702749984572,
538
- "bytes": 2560030,
539
- "pages": 36,
540
  "shards": 2
541
  },
542
  {
543
- "seconds": 1.3531785280028998,
544
- "bytes": 11787584,
545
- "pages": 32,
546
- "shards": 4
547
  },
548
  {
549
- "seconds": 1.0967427190007584,
550
- "bytes": 6096101,
551
- "pages": 36,
552
  "shards": 2
553
  },
554
  {
555
- "seconds": 1.3933399300003657,
556
- "bytes": 2445888,
557
- "pages": 28,
558
  "shards": 2
559
  },
560
  {
561
- "seconds": 1.9226194889997714,
562
- "bytes": 9744014,
563
- "pages": 39,
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": 75711.298881415,
26
+ "required_pages_per_second_upper_estimate": 0.0052378833310329055
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": 9463.912360176875,
40
+ "required_pages_per_second_upper_estimate": 0.003962342476242838
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": 18927.82472035375,
54
+ "required_pages_per_second_upper_estimate": 0.003942166728944734
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": 28391.737080530624,
68
+ "required_pages_per_second_upper_estimate": 0.0017020412863054321
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": 28391.737080530624,
82
+ "required_pages_per_second_upper_estimate": 0.002207332675957781
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 22,
95
+ "required_token_positions_per_second": 28391.737080530624,
96
+ "required_pages_per_second_upper_estimate": 0.3641787185968705
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
+ "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.13520461964066044,
162
+ "producer_p95_seconds": 0.2855433209988405,
163
+ "gpu_update_seconds": 0.9738044530906622,
164
+ "producer_batches_per_second": 7.396196983932548,
165
+ "gpu_batches_per_second": 1.0269002126928033,
166
+ "producer_headroom": 7.202449558889239,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.37587609162106,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 77,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 358001.55127571075,
174
+ "upload_bytes_per_second": 3629836.3380909152,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 76.81879905417911,
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": 1.716705255999841,
184
+ "bytes": 6235687,
185
+ "pages": 36,
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 33.88602663399979,
190
+ "bytes": 7355362,
191
+ "pages": 31,
192
+ "shards": 3
193
  },
194
  {
195
+ "seconds": 1.5573600560001069,
196
+ "bytes": 11648033,
197
+ "pages": 46,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 1.2928704839996499,
202
+ "bytes": 7315946,
203
+ "pages": 39,
204
+ "shards": 3
205
  },
206
  {
207
+ "seconds": 1.4081739580014982,
208
+ "bytes": 5891309,
209
+ "pages": 28,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.286470338000072,
214
+ "bytes": 2778002,
215
+ "pages": 37,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.1679330120005034,
220
+ "bytes": 10820666,
221
  "pages": 37,
222
  "shards": 3
223
  },
224
  {
225
+ "seconds": 1.4827527400011604,
226
+ "bytes": 6647922,
227
+ "pages": 30,
228
+ "shards": 2
 
 
 
 
 
 
229
  },
230
  {
231
+ "seconds": 1.248109399999521,
232
+ "bytes": 7422482,
233
+ "pages": 37,
234
+ "shards": 3
235
  },
236
  {
237
+ "seconds": 1.451942487999986,
238
+ "bytes": 6077412,
239
+ "pages": 30,
240
  "shards": 2
241
  },
242
  {
243
+ "seconds": 1.4631003799986502,
244
+ "bytes": 7075109,
245
+ "pages": 43,
246
  "shards": 3
247
  },
248
  {
249
+ "seconds": 2.129564337999909,
250
+ "bytes": 5884538,
251
+ "pages": 31,
252
+ "shards": 2
253
  },
254
  {
255
+ "seconds": 1.1728028199995606,
256
+ "bytes": 6147166,
257
+ "pages": 33,
258
+ "shards": 2
259
  },
260
  {
261
+ "seconds": 1.2317090710002958,
262
+ "bytes": 8202394,
263
+ "pages": 32,
264
  "shards": 3
265
  },
266
  {
267
+ "seconds": 1.6838970150001842,
268
+ "bytes": 9992623,
269
+ "pages": 38,
270
+ "shards": 3
271
  },
272
  {
273
+ "seconds": 1.2029927740004496,
274
+ "bytes": 6068424,
275
  "pages": 37,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 1.1870869580016006,
280
+ "bytes": 2666964,
281
+ "pages": 36,
282
+ "shards": 2
283
  },
284
  {
285
+ "seconds": 1.4462591800001974,
286
+ "bytes": 10549433,
287
+ "pages": 31,
288
  "shards": 3
289
  },
290
  {
291
+ "seconds": 1.0944702749984572,
292
+ "bytes": 2560030,
293
  "pages": 36,
294
  "shards": 2
295
  },
296
  {
297
+ "seconds": 1.3531785280028998,
298
+ "bytes": 11787584,
299
+ "pages": 32,
300
+ "shards": 4
301
  },
302
  {
303
+ "seconds": 1.0967427190007584,
304
+ "bytes": 6096101,
305
+ "pages": 36,
306
+ "shards": 2
307
  },
308
  {
309
+ "seconds": 1.3933399300003657,
310
+ "bytes": 2445888,
311
+ "pages": 28,
312
+ "shards": 2
313
+ },
314
+ {
315
+ "seconds": 1.9226194889997714,
316
+ "bytes": 9744014,
317
+ "pages": 39,
318
  "shards": 3
319
  },
320
  {
321
+ "seconds": 1.1809931960015092,
322
+ "bytes": 7571529,
323
  "pages": 39,
324
  "shards": 3
325
  },
326
  {
327
+ "seconds": 0.7855844450023142,
328
+ "bytes": 302035,
329
+ "pages": 9,
330
+ "shards": 1
331
  },
332
  {
333
+ "seconds": 0.93341918599981,
334
+ "bytes": 1061744,
335
+ "pages": 32,
336
  "shards": 1
337
  },
338
  {
339
+ "seconds": 1.5622991370000818,
340
+ "bytes": 16646339,
341
+ "pages": 40,
342
+ "shards": 4
343
  },
344
  {
345
+ "seconds": 1.5330627459989046,
346
+ "bytes": 7305115,
347
+ "pages": 34,
348
+ "shards": 3
349
  },
350
  {
351
+ "seconds": 0.8943780709996645,
352
+ "bytes": 971661,
353
+ "pages": 29,
354
+ "shards": 1
355
  },
356
  {
357
+ "seconds": 1.4041120760011836,
358
+ "bytes": 7469317,
359
+ "pages": 38,
360
  "shards": 3
361
  },
362
  {
363
+ "seconds": 1.2255340310002794,
364
+ "bytes": 10181330,
365
+ "pages": 31,
366
+ "shards": 3
367
  },
368
  {
369
+ "seconds": 1.0713620990027266,
370
+ "bytes": 1066838,
371
+ "pages": 34,
372
+ "shards": 1
373
  },
374
  {
375
+ "seconds": 1.2431204950007668,
376
+ "bytes": 13191953,
377
  "pages": 36,
378
+ "shards": 4
379
  },
380
  {
381
+ "seconds": 1.2758062670000072,
382
+ "bytes": 1083353,
383
+ "pages": 30,
384
  "shards": 1
385
  },
386
  {
387
+ "seconds": 1.7504430440021679,
388
+ "bytes": 10417035,
389
  "pages": 38,
390
  "shards": 3
391
  },
392
  {
393
+ "seconds": 1.7296301900023536,
394
+ "bytes": 7708812,
395
+ "pages": 39,
396
+ "shards": 3
 
 
 
 
 
 
397
  },
398
  {
399
+ "seconds": 1.0786898769983964,
400
+ "bytes": 798109,
401
+ "pages": 26,
402
+ "shards": 1
403
  },
404
  {
405
+ "seconds": 1.6809108209999977,
406
+ "bytes": 7259646,
407
  "pages": 29,
408
+ "shards": 3
409
  },
410
  {
411
+ "seconds": 1.6123849089999567,
412
+ "bytes": 16470150,
413
+ "pages": 45,
414
  "shards": 4
415
  },
416
  {
417
+ "seconds": 1.350964933000796,
418
+ "bytes": 1228633,
419
  "pages": 35,
420
  "shards": 1
421
  },
422
  {
423
+ "seconds": 1.590518836001138,
424
+ "bytes": 7289840,
425
+ "pages": 30,
 
 
 
 
 
 
426
  "shards": 3
427
  },
428
  {
429
+ "seconds": 0.990031227000145,
430
+ "bytes": 1215408,
431
+ "pages": 35,
432
+ "shards": 1
433
  },
434
  {
435
+ "seconds": 1.4002689310000278,
436
+ "bytes": 11880783,
437
+ "pages": 32,
438
+ "shards": 4
439
  },
440
  {
441
+ "seconds": 1.1156020849994093,
442
+ "bytes": 6256874,
443
+ "pages": 38,
444
  "shards": 2
445
  },
446
  {
447
+ "seconds": 1.0478186670006835,
448
+ "bytes": 6856744,
449
+ "pages": 36,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.187728501001402,
454
+ "bytes": 7914573,
455
+ "pages": 30,
456
  "shards": 3
457
  },
458
  {
459
+ "seconds": 1.098511371998029,
460
+ "bytes": 5354193,
461
+ "pages": 34,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 2.7016539569995075,
466
+ "bytes": 7555720,
467
+ "pages": 39,
468
  "shards": 3
469
  },
470
  {
471
+ "seconds": 1.483174205000978,
472
+ "bytes": 5900678,
473
+ "pages": 28,
474
  "shards": 2
475
  },
476
  {
477
+ "seconds": 0.8595360259969311,
478
+ "bytes": 1182793,
479
+ "pages": 35,
480
+ "shards": 1
481
  },
482
  {
483
+ "seconds": 1.7856543759990018,
484
+ "bytes": 17299146,
485
+ "pages": 27,
486
+ "shards": 6
487
  },
488
  {
489
+ "seconds": 0.9336543670033279,
490
+ "bytes": 1145946,
491
+ "pages": 36,
492
+ "shards": 1
493
+ },
494
+ {
495
+ "seconds": 1.1468644119995588,
496
+ "bytes": 6284755,
497
+ "pages": 28,
498
  "shards": 2
499
  },
500
  {
501
+ "seconds": 1.0925358679996862,
502
+ "bytes": 2671177,
503
+ "pages": 37,
504
+ "shards": 2
505
  },
506
  {
507
+ "seconds": 1.42488349499763,
508
+ "bytes": 9837963,
509
+ "pages": 31,
510
  "shards": 3
511
  },
512
  {
513
+ "seconds": 1.2818151600004057,
514
+ "bytes": 7876217,
515
+ "pages": 40,
516
+ "shards": 3
517
  },
518
  {
519
+ "seconds": 1.1551925370004028,
520
+ "bytes": 6771836,
521
+ "pages": 30,
522
  "shards": 2
523
  },
524
  {
525
+ "seconds": 1.460754736999661,
526
+ "bytes": 10771768,
527
+ "pages": 35,
528
  "shards": 3
529
  },
530
  {
531
+ "seconds": 0.9620606520002184,
532
+ "bytes": 2617341,
533
+ "pages": 35,
534
  "shards": 2
535
  },
536
  {
537
+ "seconds": 1.1643879789990024,
538
+ "bytes": 6176643,
539
+ "pages": 37,
540
+ "shards": 2
541
  },
542
  {
543
+ "seconds": 1.7241964509994432,
544
+ "bytes": 5999571,
545
+ "pages": 29,
546
  "shards": 2
547
  },
548
  {
549
+ "seconds": 1.129534389001492,
550
+ "bytes": 2663790,
551
+ "pages": 37,
552
  "shards": 2
553
  },
554
  {
555
+ "seconds": 1.8240933249981026,
556
+ "bytes": 16162613,
557
+ "pages": 37,
558
+ "shards": 4
559
+ },
560
+ {
561
+ "seconds": 1.3452121949994762,
562
+ "bytes": 2460437,
563
+ "pages": 29,
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 sha256:892eb6396b64cb4fd594382fdc135dcfdab02c54a38e9994701368b84bd73dcd
3
- size 721770367
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:18178f6d7c97fbaf7be1e689601dba5e41d5acf35448a4af6e2587866720a8d9
3
+ size 721784575
training_config.json CHANGED
@@ -53,7 +53,7 @@
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
- "prefetch_batches": 63,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 62,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,