appvoid commited on
Commit
5320b3c
·
verified ·
1 Parent(s): 682bb06

Qwen90M step 700000; 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 +270 -162
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 695000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "7bc36582f93f2989aadd91232699a7af1bc7686d271b31d30c2381bb46514fed",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "b91f72a9cbd032ac4aee83653c42086a0ab52034712e3ebf828258d87653ea7c",
12
- "training_config.json": "78af77f846fb3dcba08002ed692d24a6e4b3900ba35e5592c606aca3466b41f8",
13
  "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29",
14
- "shard_scheduler.json": "2790a0c930904e09253ffeb8a2fa8e78d0efbaea59ac655453a067775f8e70cb",
15
- "coverage.json": "829a2aa07677fbb65f65c1542e4754afdcac1f9751729ae84cf87c3de9db2b47",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac"
 
1
  {
2
+ "step": 700000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea",
12
+ "training_config.json": "6b17d8d96cc52df4441b50c79edf3e942bfe52f3c28b64d97f8a592a94b8549e",
13
  "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29",
14
+ "shard_scheduler.json": "cc4225a4f4c10a3ebe7ae762beddc0b1028f2619039eed4fbdfc67099a29a929",
15
+ "coverage.json": "fbe12c0005986d94f81200adeea5b3161fe8c1a4817c2774a13be90699b44102",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 695000,
3
  "stats": {
4
- "climbmix/blocks": 25020000,
5
- "climbmix/targets": 41825829199,
6
- "climbmix/input_tokens": 41828479893,
7
- "ultra_qa/blocks": 9382500,
8
- "ultra_qa/targets": 15306790749,
9
- "ultra_qa/input_tokens": 15306820344,
10
- "cortex/blocks": 9382500,
11
- "cortex/targets": 356748789,
12
- "cortex/input_tokens": 14914158761,
13
- "ultra_style/blocks": 9382500,
14
- "ultra_style/targets": 14840323358,
15
- "ultra_style/input_tokens": 14841102239,
16
- "openmath/blocks": 6255000,
17
- "openmath/targets": 9665356984,
18
- "openmath/input_tokens": 9668577218,
19
- "rewrite6/blocks": 3127500,
20
- "rewrite6/targets": 5127151612,
21
- "rewrite6/input_tokens": 5132772712
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3426,
26
- "row": 2946,
27
  "start": {
28
- "file": 165,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 67,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 4096,
37
- "documents": 14035842
38
  },
39
  "rewrite6": {
40
- "page": 49,
41
- "row": 2054,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 200704
47
  }
48
  },
49
- "epoch": 38,
50
  "offset": 0,
51
- "documents": 8008110
52
  },
53
  "openmath": {
54
- "page": 2141,
55
- "row": 2196,
56
  "start": {
57
  "file": 20,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 36,
61
- "row": 536
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 0,
66
- "documents": 8771732
67
  },
68
  "ultra_qa": {
69
- "page": 925,
70
- "row": 4086,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 2,
76
- "row": 22133
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 3792886
82
  },
83
  "ultra_style": {
84
- "page": 1413,
85
- "row": 1853,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 3,
91
- "row": 52009
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 5789501
97
  },
98
  "cortex": {
99
- "page": 181966,
100
- "row": 121,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 23291769
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 700000,
3
  "stats": {
4
+ "climbmix/blocks": 25200000,
5
+ "climbmix/targets": 42126492444,
6
+ "climbmix/input_tokens": 42129162281,
7
+ "ultra_qa/blocks": 9450000,
8
+ "ultra_qa/targets": 15416907813,
9
+ "ultra_qa/input_tokens": 15416937583,
10
+ "cortex/blocks": 9450000,
11
+ "cortex/targets": 359315602,
12
+ "cortex/input_tokens": 15021391599,
13
+ "ultra_style/blocks": 9450000,
14
+ "ultra_style/targets": 14946958997,
15
+ "ultra_style/input_tokens": 14947743302,
16
+ "openmath/blocks": 6300000,
17
+ "openmath/targets": 9734892069,
18
+ "openmath/input_tokens": 9738135363,
19
+ "rewrite6/blocks": 3150000,
20
+ "rewrite6/targets": 5163995804,
21
+ "rewrite6/input_tokens": 5169657315
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3451,
26
+ "row": 1860,
27
  "start": {
28
+ "file": 167,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 2,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 14137156
38
  },
39
  "rewrite6": {
40
+ "page": 13,
41
+ "row": 1638,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 53248
47
  }
48
  },
49
+ "epoch": 39,
50
  "offset": 0,
51
+ "documents": 8065642
52
  },
53
  "openmath": {
54
+ "page": 2156,
55
+ "row": 3724,
56
  "start": {
57
  "file": 20,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 97,
61
+ "row": 976
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 2048,
66
+ "documents": 8834700
67
  },
68
  "ultra_qa": {
69
+ "page": 932,
70
+ "row": 2614,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 2,
76
+ "row": 50805
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 3820086
82
  },
83
  "ultra_style": {
84
+ "page": 1423,
85
+ "row": 2123,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 4,
91
+ "row": 8311
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 5830731
97
  },
98
  "cortex": {
99
+ "page": 183275,
100
+ "row": 1,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 23459201
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 695000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "7bc36582f93f2989aadd91232699a7af1bc7686d271b31d30c2381bb46514fed", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "b91f72a9cbd032ac4aee83653c42086a0ab52034712e3ebf828258d87653ea7c", "training_config.json": "78af77f846fb3dcba08002ed692d24a6e4b3900ba35e5592c606aca3466b41f8", "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29", "shard_scheduler.json": "2790a0c930904e09253ffeb8a2fa8e78d0efbaea59ac655453a067775f8e70cb", "coverage.json": "829a2aa07677fbb65f65c1542e4754afdcac1f9751729ae84cf87c3de9db2b47", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac", "folder": "step-000000695000"}
 
1
+ {"step": 700000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea", "training_config.json": "6b17d8d96cc52df4441b50c79edf3e942bfe52f3c28b64d97f8a592a94b8549e", "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29", "shard_scheduler.json": "cc4225a4f4c10a3ebe7ae762beddc0b1028f2619039eed4fbdfc67099a29a929", "coverage.json": "fbe12c0005986d94f81200adeea5b3161fe8c1a4817c2774a13be90699b44102", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac", "folder": "step-000000700000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7bc36582f93f2989aadd91232699a7af1bc7686d271b31d30c2381bb46514fed
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd
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": 77286.3711192274,
26
- "required_pages_per_second_upper_estimate": 0.005346850483115876
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": 9660.796389903426,
40
- "required_pages_per_second_upper_estimate": 0.004044773708083281
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": 19321.59277980685,
54
- "required_pages_per_second_upper_estimate": 0.004024178231366769
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": 28982.389169710274,
68
- "required_pages_per_second_upper_estimate": 0.0017374499771782323
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": 28982.389169710274,
82
- "required_pages_per_second_upper_estimate": 0.002253253277887524
83
  },
84
  {
85
  "source": "cortex",
@@ -92,8 +92,8 @@
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": 28982.389169710274,
96
- "required_pages_per_second_upper_estimate": 0.3717549694040645
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,168 +155,30 @@
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": 0.106444388218776,
162
- "producer_p95_seconds": 0.20390691600005084,
163
- "gpu_update_seconds": 0.953958620806015,
164
- "producer_batches_per_second": 9.394576987418933,
165
- "gpu_batches_per_second": 1.0482634971683404,
166
- "producer_headroom": 8.962037705974092,
167
  "sustainable": true,
168
- "buffer_seconds": 60.09939311077895,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 83,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 375549.2298314112,
174
- "upload_bytes_per_second": 3808720.5720877824,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 80.58412238318208,
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": 0.9768797280003128,
184
- "bytes": 23252,
185
- "pages": 1,
186
- "shards": 1
187
- },
188
- {
189
- "seconds": 1.0272072289999414,
190
- "bytes": 7383107,
191
- "pages": 38,
192
- "shards": 3
193
- },
194
- {
195
- "seconds": 1.1484548910002559,
196
- "bytes": 5982767,
197
- "pages": 37,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.7043199660001846,
202
- "bytes": 963598,
203
- "pages": 26,
204
- "shards": 1
205
- },
206
- {
207
- "seconds": 1.1770538490000035,
208
- "bytes": 8003790,
209
- "pages": 38,
210
- "shards": 3
211
- },
212
- {
213
- "seconds": 1.2935991909998847,
214
- "bytes": 11779732,
215
- "pages": 36,
216
- "shards": 3
217
- },
218
- {
219
- "seconds": 1.0504046130008646,
220
- "bytes": 6464700,
221
- "pages": 39,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 0.9827144040000348,
226
- "bytes": 5877076,
227
- "pages": 26,
228
- "shards": 2
229
- },
230
- {
231
- "seconds": 1.1320399030000772,
232
- "bytes": 865700,
233
- "pages": 26,
234
- "shards": 1
235
- },
236
- {
237
- "seconds": 1.2121052059992508,
238
- "bytes": 12463391,
239
- "pages": 43,
240
- "shards": 4
241
- },
242
- {
243
- "seconds": 1.107387490000292,
244
- "bytes": 5957600,
245
- "pages": 27,
246
- "shards": 2
247
- },
248
- {
249
- "seconds": 25.62068473599993,
250
- "bytes": 7109407,
251
- "pages": 37,
252
- "shards": 2
253
- },
254
- {
255
- "seconds": 1.223384894999981,
256
- "bytes": 7577595,
257
- "pages": 39,
258
- "shards": 3
259
- },
260
- {
261
- "seconds": 0.9591294869997,
262
- "bytes": 6044241,
263
- "pages": 36,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 1.1359251849999055,
268
- "bytes": 7093211,
269
- "pages": 33,
270
- "shards": 3
271
- },
272
- {
273
- "seconds": 1.1475876189997507,
274
- "bytes": 6414179,
275
- "pages": 37,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 1.1066383229999701,
280
- "bytes": 6242311,
281
- "pages": 36,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 1.1400584189996152,
286
- "bytes": 8258283,
287
- "pages": 32,
288
- "shards": 4
289
- },
290
- {
291
- "seconds": 1.5345939340004406,
292
- "bytes": 10306218,
293
- "pages": 37,
294
- "shards": 3
295
- },
296
- {
297
- "seconds": 1.8383978550000393,
298
- "bytes": 2628306,
299
- "pages": 33,
300
- "shards": 2
301
- },
302
- {
303
- "seconds": 1.650117919000877,
304
- "bytes": 6106251,
305
- "pages": 36,
306
- "shards": 2
307
- },
308
- {
309
- "seconds": 1.0884742970001753,
310
- "bytes": 6215302,
311
- "pages": 37,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.2488605079997797,
316
- "bytes": 6396111,
317
- "pages": 35,
318
- "shards": 3
319
- },
320
  {
321
  "seconds": 1.8616032419995463,
322
  "bytes": 11914182,
@@ -454,6 +316,252 @@
454
  "bytes": 11238485,
455
  "pages": 32,
456
  "shards": 4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
457
  }
458
  ],
459
  "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": 77507.70753695644,
26
+ "required_pages_per_second_upper_estimate": 0.005362163050065615
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": 9688.463442119555,
40
+ "required_pages_per_second_upper_estimate": 0.004056357325092425
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": 19376.92688423911,
54
+ "required_pages_per_second_upper_estimate": 0.004035702866061544
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": 29065.39032635867,
68
+ "required_pages_per_second_upper_estimate": 0.0017424257697838755
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": 29065.39032635867,
82
+ "required_pages_per_second_upper_estimate": 0.002259706252733432
83
  },
84
  {
85
  "source": "cortex",
 
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": 29065.39032635867,
96
+ "required_pages_per_second_upper_estimate": 0.3728196191218516
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": 64,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.08243451003123425,
162
+ "producer_p95_seconds": 0.21277156500036654,
163
+ "gpu_update_seconds": 0.9512344300061482,
164
+ "producer_batches_per_second": 12.130841799400544,
165
+ "gpu_batches_per_second": 1.0512655644660978,
166
+ "producer_headroom": 11.539274384547534,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.87900352039348,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 80,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 409692.50155643857,
174
+ "upload_bytes_per_second": 4323983.225698083,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 87.91047368068568,
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": 1.8616032419995463,
184
  "bytes": 11914182,
 
316
  "bytes": 11238485,
317
  "pages": 32,
318
  "shards": 4
319
+ },
320
+ {
321
+ "seconds": 0.9324971470014134,
322
+ "bytes": 1088704,
323
+ "pages": 34,
324
+ "shards": 1
325
+ },
326
+ {
327
+ "seconds": 1.2869174759998714,
328
+ "bytes": 6108290,
329
+ "pages": 9,
330
+ "shards": 2
331
+ },
332
+ {
333
+ "seconds": 1.1298799779997353,
334
+ "bytes": 7635963,
335
+ "pages": 39,
336
+ "shards": 4
337
+ },
338
+ {
339
+ "seconds": 1.3202205550005601,
340
+ "bytes": 6021371,
341
+ "pages": 27,
342
+ "shards": 2
343
+ },
344
+ {
345
+ "seconds": 1.0867800260002696,
346
+ "bytes": 6181304,
347
+ "pages": 41,
348
+ "shards": 2
349
+ },
350
+ {
351
+ "seconds": 25.14463806799904,
352
+ "bytes": 7664241,
353
+ "pages": 38,
354
+ "shards": 3
355
+ },
356
+ {
357
+ "seconds": 1.7939357480008766,
358
+ "bytes": 6243705,
359
+ "pages": 36,
360
+ "shards": 2
361
+ },
362
+ {
363
+ "seconds": 1.0855320710015803,
364
+ "bytes": 8387056,
365
+ "pages": 37,
366
+ "shards": 3
367
+ },
368
+ {
369
+ "seconds": 1.502407841000604,
370
+ "bytes": 9864729,
371
+ "pages": 33,
372
+ "shards": 3
373
+ },
374
+ {
375
+ "seconds": 1.077067022999472,
376
+ "bytes": 1093074,
377
+ "pages": 34,
378
+ "shards": 1
379
+ },
380
+ {
381
+ "seconds": 1.176304970000274,
382
+ "bytes": 7496411,
383
+ "pages": 39,
384
+ "shards": 3
385
+ },
386
+ {
387
+ "seconds": 1.0410123799993016,
388
+ "bytes": 5644158,
389
+ "pages": 19,
390
+ "shards": 2
391
+ },
392
+ {
393
+ "seconds": 1.434721014000388,
394
+ "bytes": 6921105,
395
+ "pages": 46,
396
+ "shards": 3
397
+ },
398
+ {
399
+ "seconds": 1.2365994310002861,
400
+ "bytes": 11662593,
401
+ "pages": 31,
402
+ "shards": 3
403
+ },
404
+ {
405
+ "seconds": 1.2687110730003042,
406
+ "bytes": 6191110,
407
+ "pages": 36,
408
+ "shards": 2
409
+ },
410
+ {
411
+ "seconds": 1.3943383940004423,
412
+ "bytes": 6935244,
413
+ "pages": 34,
414
+ "shards": 3
415
+ },
416
+ {
417
+ "seconds": 1.2232918650006468,
418
+ "bytes": 6107930,
419
+ "pages": 37,
420
+ "shards": 2
421
+ },
422
+ {
423
+ "seconds": 1.1886329499993735,
424
+ "bytes": 6263136,
425
+ "pages": 35,
426
+ "shards": 2
427
+ },
428
+ {
429
+ "seconds": 0.9854559350005729,
430
+ "bytes": 2421861,
431
+ "pages": 30,
432
+ "shards": 2
433
+ },
434
+ {
435
+ "seconds": 1.444283180999264,
436
+ "bytes": 6519870,
437
+ "pages": 22,
438
+ "shards": 2
439
+ },
440
+ {
441
+ "seconds": 1.324441224000111,
442
+ "bytes": 10689415,
443
+ "pages": 37,
444
+ "shards": 3
445
+ },
446
+ {
447
+ "seconds": 1.1898289029995794,
448
+ "bytes": 7719194,
449
+ "pages": 39,
450
+ "shards": 3
451
+ },
452
+ {
453
+ "seconds": 0.9053547809999145,
454
+ "bytes": 813842,
455
+ "pages": 25,
456
+ "shards": 1
457
+ },
458
+ {
459
+ "seconds": 1.1792037220002385,
460
+ "bytes": 7645453,
461
+ "pages": 39,
462
+ "shards": 3
463
+ },
464
+ {
465
+ "seconds": 1.0877346289998968,
466
+ "bytes": 10017417,
467
+ "pages": 33,
468
+ "shards": 3
469
+ },
470
+ {
471
+ "seconds": 1.1130377059998864,
472
+ "bytes": 986874,
473
+ "pages": 29,
474
+ "shards": 1
475
+ },
476
+ {
477
+ "seconds": 1.09502404000159,
478
+ "bytes": 13537871,
479
+ "pages": 40,
480
+ "shards": 4
481
+ },
482
+ {
483
+ "seconds": 0.918569364999712,
484
+ "bytes": 999510,
485
+ "pages": 30,
486
+ "shards": 1
487
+ },
488
+ {
489
+ "seconds": 1.372887202000129,
490
+ "bytes": 11719387,
491
+ "pages": 39,
492
+ "shards": 4
493
+ },
494
+ {
495
+ "seconds": 1.0358149310013687,
496
+ "bytes": 6338461,
497
+ "pages": 36,
498
+ "shards": 2
499
+ },
500
+ {
501
+ "seconds": 0.8834309400008351,
502
+ "bytes": 840516,
503
+ "pages": 27,
504
+ "shards": 1
505
+ },
506
+ {
507
+ "seconds": 1.1130535650008824,
508
+ "bytes": 11588700,
509
+ "pages": 34,
510
+ "shards": 4
511
+ },
512
+ {
513
+ "seconds": 1.0404952170010802,
514
+ "bytes": 6993608,
515
+ "pages": 35,
516
+ "shards": 2
517
+ },
518
+ {
519
+ "seconds": 1.010654464000254,
520
+ "bytes": 6096042,
521
+ "pages": 37,
522
+ "shards": 3
523
+ },
524
+ {
525
+ "seconds": 1.1592712370002118,
526
+ "bytes": 7445713,
527
+ "pages": 36,
528
+ "shards": 3
529
+ },
530
+ {
531
+ "seconds": 1.0245760489997338,
532
+ "bytes": 5205900,
533
+ "pages": 31,
534
+ "shards": 2
535
+ },
536
+ {
537
+ "seconds": 1.034200502999738,
538
+ "bytes": 7563521,
539
+ "pages": 38,
540
+ "shards": 3
541
+ },
542
+ {
543
+ "seconds": 1.312004507999518,
544
+ "bytes": 5834009,
545
+ "pages": 26,
546
+ "shards": 2
547
+ },
548
+ {
549
+ "seconds": 1.3078732329995546,
550
+ "bytes": 7188440,
551
+ "pages": 42,
552
+ "shards": 2
553
+ },
554
+ {
555
+ "seconds": 1.37187312900096,
556
+ "bytes": 11914910,
557
+ "pages": 34,
558
+ "shards": 4
559
+ },
560
+ {
561
+ "seconds": 1.1714240479996079,
562
+ "bytes": 6274938,
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:b91f72a9cbd032ac4aee83653c42086a0ab52034712e3ebf828258d87653ea7c
3
- size 721759359
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea
3
+ size 721768191
training_config.json CHANGED
@@ -53,8 +53,8 @@
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,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 64,
57
+ "prefetch_low_water": 32,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
60
  "force_rebenchmark": false,