appvoid commited on
Commit
2ca9bc4
·
verified ·
1 Parent(s): 3d2e136

Qwen90M step 705000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +47 -47
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +239 -239
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
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"
 
1
  {
2
+ "step": 705000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "c24a63cd2e1cb9edf30d6409a48273ab6c01c49122624fe51ba44135f928c526",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "7d0f9cff5ba25bc9ee4e96e42582ae75644b70f80f581eeb59cc1552ebe7c806",
12
+ "training_config.json": "78af77f846fb3dcba08002ed692d24a6e4b3900ba35e5592c606aca3466b41f8",
13
  "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29",
14
+ "shard_scheduler.json": "92f5af91d86da30a207fc2ede8db0f46e2f61efa3eee22a9c1c2a58df9954b09",
15
+ "coverage.json": "8a4c9c5871c780362c346a7c8a31c7e57337e0f11d467c16d825aefec83f9598",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac"
coverage.json CHANGED
@@ -1,107 +1,107 @@
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
  }
 
1
  {
2
+ "step": 705000,
3
  "stats": {
4
+ "climbmix/blocks": 25380000,
5
+ "climbmix/targets": 42427427006,
6
+ "climbmix/input_tokens": 42430116101,
7
+ "ultra_qa/blocks": 9517500,
8
+ "ultra_qa/targets": 15526937657,
9
+ "ultra_qa/input_tokens": 15526967636,
10
+ "cortex/blocks": 9517500,
11
+ "cortex/targets": 361880143,
12
+ "cortex/input_tokens": 15128656803,
13
+ "ultra_style/blocks": 9517500,
14
+ "ultra_style/targets": 15053668730,
15
+ "ultra_style/input_tokens": 15054458776,
16
+ "openmath/blocks": 6345000,
17
+ "openmath/targets": 9804346129,
18
+ "openmath/input_tokens": 9807612496,
19
+ "rewrite6/blocks": 3172500,
20
+ "rewrite6/targets": 5200843675,
21
+ "rewrite6/input_tokens": 5206545918
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3476,
26
+ "row": 737,
27
  "start": {
28
+ "file": 168,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 19,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 0,
37
+ "documents": 14238433
38
  },
39
  "rewrite6": {
40
+ "page": 27,
41
+ "row": 2327,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 110592
47
  }
48
  },
49
  "epoch": 39,
50
  "offset": 0,
51
+ "documents": 8123675
52
  },
53
  "openmath": {
54
+ "page": 2172,
55
+ "row": 1176,
56
  "start": {
57
  "file": 20,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 163,
61
+ "row": 512
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 0,
66
+ "documents": 8897688
67
  },
68
  "ultra_qa": {
69
+ "page": 939,
70
+ "row": 1127,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 3,
76
+ "row": 15078
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 3847271
82
  },
83
  "ultra_style": {
84
+ "page": 1433,
85
+ "row": 3105,
86
  "start": {
87
  "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
+ "row": 49271
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 5872673
97
  },
98
  "cortex": {
99
+ "page": 184583,
100
+ "row": 79,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 23626703
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
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"}
 
1
+ {"step": 705000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "c24a63cd2e1cb9edf30d6409a48273ab6c01c49122624fe51ba44135f928c526", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "7d0f9cff5ba25bc9ee4e96e42582ae75644b70f80f581eeb59cc1552ebe7c806", "training_config.json": "78af77f846fb3dcba08002ed692d24a6e4b3900ba35e5592c606aca3466b41f8", "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29", "shard_scheduler.json": "92f5af91d86da30a207fc2ede8db0f46e2f61efa3eee22a9c1c2a58df9954b09", "coverage.json": "8a4c9c5871c780362c346a7c8a31c7e57337e0f11d467c16d825aefec83f9598", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac", "folder": "step-000000705000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c24a63cd2e1cb9edf30d6409a48273ab6c01c49122624fe51ba44135f928c526
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": 77507.70753695644,
26
- "required_pages_per_second_upper_estimate": 0.005362163050065615
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": 9688.463442119555,
40
- "required_pages_per_second_upper_estimate": 0.004056357325092425
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": 19376.92688423911,
54
- "required_pages_per_second_upper_estimate": 0.004035702866061544
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": 29065.39032635867,
68
- "required_pages_per_second_upper_estimate": 0.0017424257697838755
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": 29065.39032635867,
82
- "required_pages_per_second_upper_estimate": 0.002259706252733432
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": 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.",
@@ -155,413 +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": 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,
185
- "pages": 32,
186
- "shards": 3
187
- },
188
- {
189
- "seconds": 1.137752229000398,
190
- "bytes": 7686898,
191
- "pages": 38,
192
- "shards": 3
193
- },
194
- {
195
- "seconds": 1.062077575999865,
196
- "bytes": 5281467,
197
- "pages": 34,
198
  "shards": 2
199
  },
200
  {
201
- "seconds": 0.9683684410001661,
202
- "bytes": 6284844,
203
- "pages": 36,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 0.9990272300001379,
208
- "bytes": 2457257,
209
- "pages": 27,
210
  "shards": 2
211
  },
212
  {
213
- "seconds": 1.5623461810000663,
214
- "bytes": 6316246,
215
  "pages": 37,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 1.5784334970003329,
220
- "bytes": 16058487,
221
- "pages": 41,
222
- "shards": 4
223
  },
224
  {
225
- "seconds": 1.192334227000174,
226
- "bytes": 2321455,
227
- "pages": 29,
228
- "shards": 2
229
  },
230
  {
231
- "seconds": 1.0339759600001344,
232
- "bytes": 6174564,
233
- "pages": 37,
234
- "shards": 2
235
  },
236
  {
237
- "seconds": 1.00537013399844,
238
- "bytes": 7625189,
239
- "pages": 38,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 1.1632388540001557,
244
- "bytes": 4899738,
245
- "pages": 24,
246
- "shards": 2
247
  },
248
  {
249
- "seconds": 1.1982349889985926,
250
- "bytes": 6456251,
251
- "pages": 39,
252
- "shards": 2
253
  },
254
  {
255
- "seconds": 1.1799431950003054,
256
- "bytes": 8208196,
257
- "pages": 35,
258
- "shards": 3
259
  },
260
  {
261
- "seconds": 1.0678595420013153,
262
- "bytes": 6041574,
 
 
 
 
 
 
263
  "pages": 36,
264
  "shards": 2
265
  },
266
  {
267
- "seconds": 1.7238812670002517,
268
- "bytes": 11592714,
269
- "pages": 34,
270
- "shards": 4
271
  },
272
  {
273
- "seconds": 1.1851188379987434,
274
- "bytes": 1113462,
275
- "pages": 32,
276
- "shards": 1
277
  },
278
  {
279
- "seconds": 1.064678984999773,
280
- "bytes": 6185154,
281
- "pages": 37,
282
  "shards": 2
283
  },
284
  {
285
- "seconds": 1.4897821249996923,
286
- "bytes": 7479884,
287
  "pages": 37,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 1.613565400999505,
292
- "bytes": 11743615,
293
- "pages": 37,
294
  "shards": 3
295
  },
296
  {
297
- "seconds": 1.0746530969991,
298
- "bytes": 5830095,
299
- "pages": 27,
300
  "shards": 2
301
  },
302
  {
303
- "seconds": 1.0229502059992228,
304
- "bytes": 2562647,
305
- "pages": 37,
306
- "shards": 2
307
  },
308
  {
309
- "seconds": 1.0365147789998446,
310
- "bytes": 6395327,
311
- "pages": 37,
312
  "shards": 2
313
  },
314
  {
315
- "seconds": 1.1455854699997872,
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": {
 
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": 77341.28166839482,
26
+ "required_pages_per_second_upper_estimate": 0.005350649322317305
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": 9667.660208549352,
40
+ "required_pages_per_second_upper_estimate": 0.004047647445617498
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": 19335.320417098705,
54
+ "required_pages_per_second_upper_estimate": 0.0040270373361925215
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": 29002.98062564806,
68
+ "required_pages_per_second_upper_estimate": 0.0017386844035203698
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": 29002.98062564806,
82
+ "required_pages_per_second_upper_estimate": 0.002254854173014444
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": 29002.98062564806,
96
+ "required_pages_per_second_upper_estimate": 0.3720190944914516
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": 0.09527713949219674,
162
+ "producer_p95_seconds": 0.2191641060016991,
163
+ "gpu_update_seconds": 0.9532813319038729,
164
+ "producer_batches_per_second": 10.495697135007928,
165
+ "gpu_batches_per_second": 1.0490082691568308,
166
+ "producer_headroom": 10.00535214412002,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.05672390994399,
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": 402564.9410295652,
174
+ "upload_bytes_per_second": 5735685.118812759,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 86.38106511273591,
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": 1.1886329499993735,
184
+ "bytes": 6263136,
185
+ "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 0.9854559350005729,
190
+ "bytes": 2421861,
191
+ "pages": 30,
192
  "shards": 2
193
  },
194
  {
195
+ "seconds": 1.444283180999264,
196
+ "bytes": 6519870,
197
+ "pages": 22,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.324441224000111,
202
+ "bytes": 10689415,
203
  "pages": 37,
204
+ "shards": 3
205
  },
206
  {
207
+ "seconds": 1.1898289029995794,
208
+ "bytes": 7719194,
209
+ "pages": 39,
210
+ "shards": 3
211
  },
212
  {
213
+ "seconds": 0.9053547809999145,
214
+ "bytes": 813842,
215
+ "pages": 25,
216
+ "shards": 1
217
  },
218
  {
219
+ "seconds": 1.1792037220002385,
220
+ "bytes": 7645453,
221
+ "pages": 39,
222
+ "shards": 3
223
  },
224
  {
225
+ "seconds": 1.0877346289998968,
226
+ "bytes": 10017417,
227
+ "pages": 33,
228
  "shards": 3
229
  },
230
  {
231
+ "seconds": 1.1130377059998864,
232
+ "bytes": 986874,
233
+ "pages": 29,
234
+ "shards": 1
235
  },
236
  {
237
+ "seconds": 1.09502404000159,
238
+ "bytes": 13537871,
239
+ "pages": 40,
240
+ "shards": 4
241
  },
242
  {
243
+ "seconds": 0.918569364999712,
244
+ "bytes": 999510,
245
+ "pages": 30,
246
+ "shards": 1
247
  },
248
  {
249
+ "seconds": 1.372887202000129,
250
+ "bytes": 11719387,
251
+ "pages": 39,
252
+ "shards": 4
253
+ },
254
+ {
255
+ "seconds": 1.0358149310013687,
256
+ "bytes": 6338461,
257
  "pages": 36,
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 0.8834309400008351,
262
+ "bytes": 840516,
263
+ "pages": 27,
264
+ "shards": 1
265
  },
266
  {
267
+ "seconds": 1.1130535650008824,
268
+ "bytes": 11588700,
269
+ "pages": 34,
270
+ "shards": 4
271
  },
272
  {
273
+ "seconds": 1.0404952170010802,
274
+ "bytes": 6993608,
275
+ "pages": 35,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 1.010654464000254,
280
+ "bytes": 6096042,
281
  "pages": 37,
282
  "shards": 3
283
  },
284
  {
285
+ "seconds": 1.1592712370002118,
286
+ "bytes": 7445713,
287
+ "pages": 36,
288
  "shards": 3
289
  },
290
  {
291
+ "seconds": 1.0245760489997338,
292
+ "bytes": 5205900,
293
+ "pages": 31,
294
  "shards": 2
295
  },
296
  {
297
+ "seconds": 1.034200502999738,
298
+ "bytes": 7563521,
299
+ "pages": 38,
300
+ "shards": 3
301
  },
302
  {
303
+ "seconds": 1.312004507999518,
304
+ "bytes": 5834009,
305
+ "pages": 26,
306
  "shards": 2
307
  },
308
  {
309
+ "seconds": 1.3078732329995546,
310
+ "bytes": 7188440,
311
+ "pages": 42,
312
+ "shards": 2
313
  },
314
  {
315
+ "seconds": 1.37187312900096,
316
+ "bytes": 11914910,
317
  "pages": 34,
318
+ "shards": 4
319
  },
320
  {
321
+ "seconds": 1.1714240479996079,
322
+ "bytes": 6274938,
323
+ "pages": 37,
324
  "shards": 2
325
  },
326
  {
327
+ "seconds": 1.1267517289998068,
328
+ "bytes": 2389323,
329
+ "pages": 28,
330
+ "shards": 2
331
  },
332
  {
333
+ "seconds": 1.253374073001396,
334
+ "bytes": 6015964,
335
+ "pages": 37,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 1.0633229350005422,
340
+ "bytes": 4861789,
341
+ "pages": 27,
342
  "shards": 2
343
  },
344
  {
345
+ "seconds": 1.667441460000191,
346
+ "bytes": 13590477,
347
+ "pages": 39,
348
+ "shards": 4
349
  },
350
  {
351
+ "seconds": 1.1102066319999722,
352
+ "bytes": 6253510,
353
  "pages": 36,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 1.0856502950009599,
358
+ "bytes": 2667766,
359
+ "pages": 38,
360
+ "shards": 2
361
  },
362
  {
363
+ "seconds": 1.1392225029994734,
364
+ "bytes": 10365382,
365
  "pages": 33,
366
  "shards": 3
367
  },
368
  {
369
+ "seconds": 0.9051334899995709,
370
+ "bytes": 1149324,
371
  "pages": 34,
372
  "shards": 1
373
  },
374
  {
375
+ "seconds": 1.495498950000183,
376
+ "bytes": 7289548,
377
+ "pages": 26,
378
  "shards": 3
379
  },
380
  {
381
+ "seconds": 1.2270069759997568,
382
+ "bytes": 11800938,
383
+ "pages": 38,
384
+ "shards": 3
385
  },
386
  {
387
+ "seconds": 1.0093337090002024,
388
+ "bytes": 5575178,
389
+ "pages": 32,
390
+ "shards": 2
391
  },
392
  {
393
+ "seconds": 1.1225266829987959,
394
+ "bytes": 7591094,
395
+ "pages": 38,
396
  "shards": 3
397
  },
398
  {
399
+ "seconds": 1.4399703239996597,
400
+ "bytes": 6008739,
401
  "pages": 36,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 1.3038473750020785,
406
+ "bytes": 2307657,
407
+ "pages": 29,
408
+ "shards": 2
409
+ },
410
+ {
411
+ "seconds": 1.5242876569973305,
412
+ "bytes": 9910776,
413
  "pages": 34,
414
  "shards": 3
415
  },
416
  {
417
+ "seconds": 0.9998362250007631,
418
+ "bytes": 6697585,
419
+ "pages": 27,
420
  "shards": 2
421
  },
422
  {
423
+ "seconds": 1.1621864580010879,
424
+ "bytes": 7532043,
425
+ "pages": 39,
426
+ "shards": 3
427
  },
428
  {
429
+ "seconds": 1.2297442769995541,
430
+ "bytes": 6251173,
431
+ "pages": 35,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 0.9681137329971534,
436
+ "bytes": 5823551,
437
+ "pages": 34,
438
  "shards": 2
439
  },
440
  {
441
+ "seconds": 1.3083030150010018,
442
+ "bytes": 7346473,
443
+ "pages": 31,
 
 
 
 
 
 
444
  "shards": 3
445
  },
446
  {
447
+ "seconds": 0.8391288720013108,
448
+ "bytes": 1133832,
449
+ "pages": 34,
450
  "shards": 1
451
  },
452
  {
453
+ "seconds": 1.679156835001777,
454
+ "bytes": 13329088,
455
+ "pages": 34,
456
+ "shards": 4
457
  },
458
  {
459
+ "seconds": 1.1372946299998148,
460
+ "bytes": 9925526,
461
+ "pages": 35,
462
  "shards": 3
463
  },
464
  {
465
+ "seconds": 1.0260383269996964,
466
+ "bytes": 1162438,
467
+ "pages": 35,
468
+ "shards": 2
469
  },
470
  {
471
+ "seconds": 1.0241040929977316,
472
+ "bytes": 7727150,
473
+ "pages": 39,
474
+ "shards": 3
475
  },
476
  {
477
+ "seconds": 1.1232291689993872,
478
+ "bytes": 903046,
479
+ "pages": 26,
480
  "shards": 1
481
  },
482
  {
483
+ "seconds": 1.245589108999411,
484
+ "bytes": 10344850,
485
+ "pages": 37,
486
+ "shards": 3
487
  },
488
  {
489
+ "seconds": 1.3122096120023343,
490
+ "bytes": 13394029,
491
+ "pages": 40,
492
+ "shards": 4
493
  },
494
  {
495
+ "seconds": 1.2143651210026292,
496
+ "bytes": 938191,
497
+ "pages": 29,
498
  "shards": 1
499
  },
500
  {
501
+ "seconds": 1.3705294199971831,
502
+ "bytes": 12626897,
503
+ "pages": 40,
504
  "shards": 4
505
  },
506
  {
507
+ "seconds": 1.047090729996853,
508
+ "bytes": 5981571,
509
  "pages": 35,
510
  "shards": 2
511
  },
512
  {
513
+ "seconds": 0.8534065610001562,
514
+ "bytes": 1138564,
515
+ "pages": 34,
516
+ "shards": 1
517
  },
518
  {
519
+ "seconds": 1.0472652070020558,
520
+ "bytes": 7166131,
521
+ "pages": 30,
522
  "shards": 3
523
  },
524
  {
525
+ "seconds": 1.1100580310012447,
526
+ "bytes": 6769711,
527
+ "pages": 33,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.1388595060016087,
532
+ "bytes": 11609898,
533
  "pages": 38,
534
+ "shards": 4
535
  },
536
  {
537
+ "seconds": 1.240830022998125,
538
+ "bytes": 6131822,
539
+ "pages": 37,
540
  "shards": 2
541
  },
542
  {
543
+ "seconds": 1.2179747789996327,
544
+ "bytes": 1206466,
545
+ "pages": 34,
546
+ "shards": 1
547
  },
548
  {
549
+ "seconds": 1.7537132789984753,
550
+ "bytes": 11388396,
551
  "pages": 34,
552
  "shards": 4
553
  },
554
  {
555
+ "seconds": 1.1776387320023787,
556
+ "bytes": 5920541,
557
+ "pages": 31,
558
  "shards": 2
559
+ },
560
+ {
561
+ "seconds": 1.2547966029997042,
562
+ "bytes": 8617085,
563
+ "pages": 41,
564
+ "shards": 3
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea
3
- size 721768191
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d0f9cff5ba25bc9ee4e96e42582ae75644b70f80f581eeb59cc1552ebe7c806
3
+ size 721782079
training_config.json CHANGED
@@ -53,8 +53,8 @@
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,
 
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,