appvoid commited on
Commit
e8b229d
·
verified ·
1 Parent(s): 73bf750

Qwen90M step 305000; 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 +247 -247
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 300000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "6492e2662f94faa6a2e8ee1d628fd698c35d02f34d6f8f370f9d4db12434c3dc",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "fbe139ff1c427ab9b5843d95ec3ecec658133f721c1f794977096107acee4acb",
12
- "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba",
14
- "shard_scheduler.json": "a2c401bec97cd3feb72edade9a3ae615ffe06fc28d4766333eb54ffc7af1ddd7",
15
- "coverage.json": "251945ddb781b0d2536a5c2056932a9ec606d4b451ee97c2b33e28f6ace29956",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 305000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "89b7e1e53c78bd469c045acaa45bf2613b245ad7c172ee689268c146802eaa57",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "124b35ad649617fdf90d8ce9db8b59458f1bc588be0bbe9ced8a83035f7877a9",
12
+ "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba",
14
+ "shard_scheduler.json": "9022b90c7229c4a1b45c4f0abd2854b2ee887f5b962134d4b9e98bb6007410cc",
15
+ "coverage.json": "156ed6483cb0396d79e81df2dfaa6ef016a16020c5c9fcc2a445e0bcff68144f",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 300000,
3
  "stats": {
4
- "climbmix/blocks": 10800000,
5
- "climbmix/targets": 18053965003,
6
- "climbmix/input_tokens": 18055109204,
7
- "ultra_qa/blocks": 4050000,
8
- "ultra_qa/targets": 6607020438,
9
- "ultra_qa/input_tokens": 6607033152,
10
- "cortex/blocks": 4050000,
11
- "cortex/targets": 154079984,
12
- "cortex/input_tokens": 6438099093,
13
- "ultra_style/blocks": 4050000,
14
- "ultra_style/targets": 6406628156,
15
- "ultra_style/input_tokens": 6406964117,
16
- "openmath/blocks": 2700000,
17
- "openmath/targets": 4172214215,
18
- "openmath/input_tokens": 4173605124,
19
- "rewrite6/blocks": 1350000,
20
- "rewrite6/targets": 2213124309,
21
- "rewrite6/input_tokens": 2215550812
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1479,
26
- "row": 738,
27
  "start": {
28
- "file": 71,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 49,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 6058722
38
  },
39
  "rewrite6": {
40
- "page": 41,
41
- "row": 2526,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 167936
47
  }
48
  },
49
- "epoch": 16,
50
- "offset": 0,
51
- "documents": 3456926
52
  },
53
  "openmath": {
54
- "page": 924,
55
- "row": 2465,
56
  "start": {
57
  "file": 8,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 291,
61
- "row": 504
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 3787169
67
  },
68
  "ultra_qa": {
69
- "page": 399,
70
- "row": 2747,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
- "row": 10712
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 1637051
82
  },
83
  "ultra_style": {
84
- "page": 610,
85
- "row": 401,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 5,
91
- "row": 18011
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 2498961
97
  },
98
  "cortex": {
99
- "page": 78549,
100
- "row": 103,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 10054375
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 305000,
3
  "stats": {
4
+ "climbmix/blocks": 10980000,
5
+ "climbmix/targets": 18354331663,
6
+ "climbmix/input_tokens": 18355495227,
7
+ "ultra_qa/blocks": 4117500,
8
+ "ultra_qa/targets": 6717340650,
9
+ "ultra_qa/input_tokens": 6717353602,
10
+ "cortex/blocks": 4117500,
11
+ "cortex/targets": 156645103,
12
+ "cortex/input_tokens": 6545322017,
13
+ "ultra_style/blocks": 4117500,
14
+ "ultra_style/targets": 6513296217,
15
+ "ultra_style/input_tokens": 6513637740,
16
+ "openmath/blocks": 2745000,
17
+ "openmath/targets": 4241678751,
18
+ "openmath/input_tokens": 4243092717,
19
+ "rewrite6/blocks": 1372500,
20
+ "rewrite6/targets": 2250039094,
21
+ "rewrite6/input_tokens": 2252505932
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1504,
26
+ "row": 337,
27
  "start": {
28
+ "file": 72,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 66,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 12288,
37
+ "documents": 6160721
38
  },
39
  "rewrite6": {
40
+ "page": 5,
41
+ "row": 1973,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 20480
47
  }
48
  },
49
+ "epoch": 17,
50
+ "offset": 2048,
51
+ "documents": 3514321
52
  },
53
  "openmath": {
54
+ "page": 939,
55
+ "row": 3990,
56
  "start": {
57
  "file": 8,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 352,
61
+ "row": 944
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 3850134
67
  },
68
  "ultra_qa": {
69
+ "page": 406,
70
+ "row": 1338,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
+ "row": 39384
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 1664314
82
  },
83
  "ultra_style": {
84
+ "page": 620,
85
+ "row": 994,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 5,
91
+ "row": 58971
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 2540514
97
  },
98
  "cortex": {
99
+ "page": 79857,
100
+ "row": 83,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 10221779
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 300000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "6492e2662f94faa6a2e8ee1d628fd698c35d02f34d6f8f370f9d4db12434c3dc", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "fbe139ff1c427ab9b5843d95ec3ecec658133f721c1f794977096107acee4acb", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba", "shard_scheduler.json": "a2c401bec97cd3feb72edade9a3ae615ffe06fc28d4766333eb54ffc7af1ddd7", "coverage.json": "251945ddb781b0d2536a5c2056932a9ec606d4b451ee97c2b33e28f6ace29956", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000300000"}
 
1
+ {"step": 305000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "89b7e1e53c78bd469c045acaa45bf2613b245ad7c172ee689268c146802eaa57", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "124b35ad649617fdf90d8ce9db8b59458f1bc588be0bbe9ced8a83035f7877a9", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "f89f82c3f4c44a52fcd532346ea71580902d757e59e1961e9b2bed0030259dba", "shard_scheduler.json": "9022b90c7229c4a1b45c4f0abd2854b2ee887f5b962134d4b9e98bb6007410cc", "coverage.json": "156ed6483cb0396d79e81df2dfaa6ef016a16020c5c9fcc2a445e0bcff68144f", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000305000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6492e2662f94faa6a2e8ee1d628fd698c35d02f34d6f8f370f9d4db12434c3dc
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89b7e1e53c78bd469c045acaa45bf2613b245ad7c172ee689268c146802eaa57
3
  size 360622744
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 75120.45760388167,
26
- "required_pages_per_second_upper_estimate": 0.006511250874045306
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
36
  "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
- "required_token_positions_per_second": 9390.05720048521,
40
- "required_pages_per_second_upper_estimate": 0.0037458342111397836
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
50
  "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
- "required_token_positions_per_second": 18780.11440097042,
54
- "required_pages_per_second_upper_estimate": 0.004344448947939665
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
64
  "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
- "required_token_positions_per_second": 28170.17160145563,
68
- "required_pages_per_second_upper_estimate": 0.0017154156530606451
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
78
  "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
- "required_token_positions_per_second": 28170.17160145563,
82
- "required_pages_per_second_upper_estimate": 0.0021686177899059907
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.042780299000014566,
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 22,
95
- "required_token_positions_per_second": 28170.17160145563,
96
- "required_pages_per_second_upper_estimate": 0.3497792517905513
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,412 +155,412 @@
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
157
  "selected": {
158
- "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 690670,
161
- "producer_batch_seconds": 0.20651756988294778,
162
- "producer_p95_seconds": 0.31602960099917254,
163
- "gpu_update_seconds": 0.9814636698404546,
164
- "producer_batches_per_second": 4.842203017238633,
165
- "gpu_batches_per_second": 1.0188864149832042,
166
- "producer_headroom": 4.752446343411551,
167
  "sustainable": true,
168
- "buffer_seconds": 60.85074753010819,
169
  "buffer_drain_seconds": null,
170
- "shard_target_mib": 32,
171
- "upload_pages_per_commit": 164,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 628791.9512074635,
174
- "upload_bytes_per_second": 2651314.957309086,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 67.4620576008221,
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": 1.3497758060002525,
184
- "bytes": 7313270,
185
- "pages": 37,
186
  "shards": 3
187
  },
188
  {
189
- "seconds": 1.170668664999539,
190
- "bytes": 6087923,
191
- "pages": 35,
192
  "shards": 2
193
  },
194
  {
195
- "seconds": 1.1090394350030692,
196
- "bytes": 4744304,
197
- "pages": 27,
198
  "shards": 2
199
  },
200
  {
201
- "seconds": 1.304362898998079,
202
- "bytes": 7978674,
203
- "pages": 39,
204
- "shards": 3
205
  },
206
  {
207
- "seconds": 1.0655079770003795,
208
- "bytes": 7055873,
209
- "pages": 37,
210
- "shards": 2
211
  },
212
  {
213
- "seconds": 1.268422623998049,
214
- "bytes": 5726191,
215
- "pages": 20,
216
  "shards": 2
217
  },
218
  {
219
- "seconds": 1.1066016949989717,
220
- "bytes": 7887477,
221
- "pages": 44,
222
- "shards": 3
223
  },
224
  {
225
- "seconds": 1.019014984001842,
226
- "bytes": 6005453,
227
- "pages": 36,
228
- "shards": 2
229
  },
230
  {
231
- "seconds": 1.00243121099993,
232
- "bytes": 7452907,
233
- "pages": 29,
234
  "shards": 3
235
  },
236
  {
237
- "seconds": 0.9669066969981941,
238
- "bytes": 1217353,
239
- "pages": 34,
240
  "shards": 1
241
  },
242
  {
243
- "seconds": 1.4986801279992505,
244
- "bytes": 15911648,
245
- "pages": 32,
246
- "shards": 4
247
  },
248
  {
249
- "seconds": 1.1255449119998957,
250
- "bytes": 2547314,
251
- "pages": 34,
252
- "shards": 2
 
 
 
 
 
 
253
  },
254
  {
255
- "seconds": 1.1576964999985648,
256
- "bytes": 6153703,
257
  "pages": 36,
258
- "shards": 2
259
  },
260
  {
261
- "seconds": 1.0455056530008733,
262
- "bytes": 6073870,
263
- "pages": 35,
264
  "shards": 2
265
  },
266
  {
267
- "seconds": 1.0066101270022045,
268
- "bytes": 7063346,
269
- "pages": 30,
270
  "shards": 3
271
  },
272
  {
273
- "seconds": 1.161024010998517,
274
- "bytes": 6075989,
275
- "pages": 36,
276
  "shards": 3
277
  },
278
  {
279
- "seconds": 0.8424355740025931,
280
- "bytes": 669079,
281
- "pages": 20,
282
  "shards": 1
283
  },
284
  {
285
- "seconds": 1.4896883629990043,
286
- "bytes": 13505965,
287
- "pages": 48,
288
- "shards": 4
289
  },
290
  {
291
- "seconds": 1.098302499998681,
292
- "bytes": 6017424,
293
- "pages": 23,
294
- "shards": 2
295
  },
296
  {
297
- "seconds": 1.0887724580024951,
298
- "bytes": 6603634,
299
- "pages": 44,
300
- "shards": 3
301
  },
302
  {
303
- "seconds": 1.2179049539990956,
304
- "bytes": 6068321,
305
  "pages": 36,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 2.3319496820004133,
310
- "bytes": 5855101,
311
- "pages": 27,
312
  "shards": 2
313
  },
314
  {
315
- "seconds": 1.5211132250005903,
316
- "bytes": 6886706,
317
- "pages": 29,
318
  "shards": 3
319
  },
320
  {
321
- "seconds": 1.7056977829997777,
322
- "bytes": 6292570,
323
- "pages": 16,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 10.767963721002161,
328
- "bytes": 5800923,
329
  "pages": 27,
330
  "shards": 2
331
  },
332
  {
333
- "seconds": 6.862793137002882,
334
- "bytes": 7606529,
335
- "pages": 38,
336
  "shards": 3
337
  },
338
  {
339
- "seconds": 3.3253811320028035,
340
- "bytes": 1169016,
341
- "pages": 34,
342
- "shards": 1
343
  },
344
  {
345
- "seconds": 12.285631268001453,
346
- "bytes": 12643645,
347
- "pages": 42,
348
- "shards": 4
349
  },
350
  {
351
- "seconds": 2.9626744119996147,
352
- "bytes": 6098299,
353
- "pages": 35,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.1153004340012558,
358
- "bytes": 6713918,
359
- "pages": 32,
360
- "shards": 2
361
  },
362
  {
363
- "seconds": 5.8015411799970025,
364
- "bytes": 7310989,
365
- "pages": 32,
366
  "shards": 3
367
  },
368
  {
369
- "seconds": 3.285385278999456,
370
- "bytes": 842975,
371
- "pages": 25,
372
- "shards": 1
373
- },
374
- {
375
- "seconds": 2.201649025999359,
376
- "bytes": 9943938,
377
  "pages": 29,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 1.3160915080006816,
382
- "bytes": 2528581,
383
- "pages": 36,
384
- "shards": 2
385
  },
386
  {
387
- "seconds": 8.146259255998302,
388
- "bytes": 6143251,
389
  "pages": 36,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 6.752744052999333,
394
- "bytes": 5959933,
395
  "pages": 27,
396
  "shards": 2
397
  },
398
  {
399
- "seconds": 8.626221913000336,
400
- "bytes": 11949436,
401
- "pages": 25,
402
- "shards": 4
403
  },
404
  {
405
- "seconds": 13.160608485999546,
406
- "bytes": 6360120,
407
- "pages": 35,
408
  "shards": 2
409
  },
410
  {
411
- "seconds": 7.225236703998235,
412
- "bytes": 1179968,
413
- "pages": 34,
414
- "shards": 1
415
  },
416
  {
417
- "seconds": 4.945576189998974,
418
- "bytes": 7777829,
419
- "pages": 37,
420
- "shards": 3
421
  },
422
  {
423
- "seconds": 1.2244086760001665,
424
- "bytes": 5469696,
425
- "pages": 28,
426
- "shards": 2
427
  },
428
  {
429
- "seconds": 1.178703661000327,
430
- "bytes": 7809750,
431
- "pages": 38,
432
- "shards": 3
433
  },
434
  {
435
- "seconds": 0.9307224249969295,
436
- "bytes": 6774165,
437
- "pages": 31,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 0.8456139500012796,
442
- "bytes": 6033435,
443
  "pages": 30,
444
- "shards": 2
445
  },
446
  {
447
- "seconds": 1.161550510001689,
448
- "bytes": 11437647,
449
- "pages": 35,
450
  "shards": 4
451
  },
452
  {
453
- "seconds": 0.8255374049986131,
454
- "bytes": 1098294,
455
- "pages": 34,
456
- "shards": 1
457
  },
458
  {
459
- "seconds": 0.9614813039988803,
460
- "bytes": 5875069,
461
- "pages": 29,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.3580094259996258,
466
- "bytes": 7460093,
467
- "pages": 38,
468
- "shards": 4
469
  },
470
  {
471
- "seconds": 1.418776655002148,
472
- "bytes": 10393041,
473
- "pages": 31,
474
  "shards": 3
475
  },
476
  {
477
- "seconds": 1.2120838849987194,
478
- "bytes": 7800195,
479
- "pages": 39,
480
- "shards": 3
481
  },
482
  {
483
- "seconds": 0.7076373689997126,
484
- "bytes": 1197583,
485
- "pages": 35,
486
- "shards": 1
487
  },
488
  {
489
- "seconds": 1.2053027080000902,
490
- "bytes": 10494285,
491
- "pages": 30,
492
- "shards": 3
493
  },
494
  {
495
- "seconds": 1.2219008450010733,
496
- "bytes": 7585902,
497
- "pages": 37,
498
- "shards": 3
499
  },
500
  {
501
- "seconds": 1.2029692989999603,
502
- "bytes": 1075789,
503
- "pages": 31,
504
- "shards": 1
505
  },
506
  {
507
- "seconds": 1.6663710430002538,
508
- "bytes": 11892665,
509
- "pages": 36,
510
- "shards": 3
511
  },
512
  {
513
- "seconds": 1.4550072420024662,
514
- "bytes": 2481838,
515
- "pages": 31,
516
  "shards": 2
517
  },
518
  {
519
- "seconds": 1.1862465560006967,
520
- "bytes": 10217288,
521
  "pages": 37,
522
  "shards": 3
523
  },
524
  {
525
- "seconds": 1.3918344490011805,
526
- "bytes": 7389964,
527
- "pages": 31,
528
  "shards": 3
529
  },
530
  {
531
- "seconds": 0.8301882180021494,
532
- "bytes": 1174116,
533
- "pages": 33,
534
- "shards": 1
535
  },
536
  {
537
- "seconds": 1.4423049569995783,
538
- "bytes": 10949179,
539
  "pages": 36,
540
- "shards": 3
541
  },
542
  {
543
- "seconds": 1.3191998570000578,
544
- "bytes": 13241805,
545
- "pages": 40,
546
- "shards": 4
547
  },
548
  {
549
- "seconds": 1.0903980459988816,
550
- "bytes": 954449,
551
- "pages": 26,
552
- "shards": 1
553
  },
554
  {
555
- "seconds": 1.1526671549981984,
556
- "bytes": 6176156,
557
- "pages": 36,
558
  "shards": 2
559
  },
560
  {
561
- "seconds": 1.0391054690007877,
562
- "bytes": 2626145,
563
- "pages": 36,
 
 
 
 
 
 
564
  "shards": 2
565
  }
566
  ],
 
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 74789.22219739005,
26
+ "required_pages_per_second_upper_estimate": 0.0064825402285190745
27
  },
28
  {
29
  "source": "rewrite6",
 
36
  "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
+ "required_token_positions_per_second": 9348.652774673756,
40
+ "required_pages_per_second_upper_estimate": 0.0037293173666322625
41
  },
42
  {
43
  "source": "openmath",
 
50
  "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
+ "required_token_positions_per_second": 18697.30554934751,
54
+ "required_pages_per_second_upper_estimate": 0.004325292577502718
55
  },
56
  {
57
  "source": "ultra_qa",
 
64
  "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
+ "required_token_positions_per_second": 28045.958324021267,
68
+ "required_pages_per_second_upper_estimate": 0.0017078517161615934
69
  },
70
  {
71
  "source": "ultra_style",
 
78
  "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
+ "required_token_positions_per_second": 28045.958324021267,
82
+ "required_pages_per_second_upper_estimate": 0.0021590555079647345
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.042780299000014566,
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 21,
95
+ "required_token_positions_per_second": 28045.958324021267,
96
+ "required_pages_per_second_upper_estimate": 0.3482369385999139
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
157
  "selected": {
158
+ "prefetch_batches": 61,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 690670,
161
+ "producer_batch_seconds": 0.42701781460147004,
162
+ "producer_p95_seconds": 0.3386782650013629,
163
+ "gpu_update_seconds": 0.9858104929265185,
164
+ "producer_batches_per_second": 2.341822672979783,
165
+ "gpu_batches_per_second": 1.0143937472519267,
166
+ "producer_headroom": 2.308593363596697,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.13444006851763,
169
  "buffer_drain_seconds": null,
170
+ "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 76,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 297301.97635776654,
174
+ "upload_bytes_per_second": 5711591.078716606,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 63.79408329057453,
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": 1.178703661000327,
184
+ "bytes": 7809750,
185
+ "pages": 38,
186
  "shards": 3
187
  },
188
  {
189
+ "seconds": 0.9307224249969295,
190
+ "bytes": 6774165,
191
+ "pages": 31,
192
  "shards": 2
193
  },
194
  {
195
+ "seconds": 0.8456139500012796,
196
+ "bytes": 6033435,
197
+ "pages": 30,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.161550510001689,
202
+ "bytes": 11437647,
203
+ "pages": 35,
204
+ "shards": 4
205
  },
206
  {
207
+ "seconds": 0.8255374049986131,
208
+ "bytes": 1098294,
209
+ "pages": 34,
210
+ "shards": 1
211
  },
212
  {
213
+ "seconds": 0.9614813039988803,
214
+ "bytes": 5875069,
215
+ "pages": 29,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.3580094259996258,
220
+ "bytes": 7460093,
221
+ "pages": 38,
222
+ "shards": 4
223
  },
224
  {
225
+ "seconds": 1.418776655002148,
226
+ "bytes": 10393041,
227
+ "pages": 31,
228
+ "shards": 3
229
  },
230
  {
231
+ "seconds": 1.2120838849987194,
232
+ "bytes": 7800195,
233
+ "pages": 39,
234
  "shards": 3
235
  },
236
  {
237
+ "seconds": 0.7076373689997126,
238
+ "bytes": 1197583,
239
+ "pages": 35,
240
  "shards": 1
241
  },
242
  {
243
+ "seconds": 1.2053027080000902,
244
+ "bytes": 10494285,
245
+ "pages": 30,
246
+ "shards": 3
247
  },
248
  {
249
+ "seconds": 1.2219008450010733,
250
+ "bytes": 7585902,
251
+ "pages": 37,
252
+ "shards": 3
253
+ },
254
+ {
255
+ "seconds": 1.2029692989999603,
256
+ "bytes": 1075789,
257
+ "pages": 31,
258
+ "shards": 1
259
  },
260
  {
261
+ "seconds": 1.6663710430002538,
262
+ "bytes": 11892665,
263
  "pages": 36,
264
+ "shards": 3
265
  },
266
  {
267
+ "seconds": 1.4550072420024662,
268
+ "bytes": 2481838,
269
+ "pages": 31,
270
  "shards": 2
271
  },
272
  {
273
+ "seconds": 1.1862465560006967,
274
+ "bytes": 10217288,
275
+ "pages": 37,
276
  "shards": 3
277
  },
278
  {
279
+ "seconds": 1.3918344490011805,
280
+ "bytes": 7389964,
281
+ "pages": 31,
282
  "shards": 3
283
  },
284
  {
285
+ "seconds": 0.8301882180021494,
286
+ "bytes": 1174116,
287
+ "pages": 33,
288
  "shards": 1
289
  },
290
  {
291
+ "seconds": 1.4423049569995783,
292
+ "bytes": 10949179,
293
+ "pages": 36,
294
+ "shards": 3
295
  },
296
  {
297
+ "seconds": 1.3191998570000578,
298
+ "bytes": 13241805,
299
+ "pages": 40,
300
+ "shards": 4
301
  },
302
  {
303
+ "seconds": 1.0903980459988816,
304
+ "bytes": 954449,
305
+ "pages": 26,
306
+ "shards": 1
307
  },
308
  {
309
+ "seconds": 1.1526671549981984,
310
+ "bytes": 6176156,
311
  "pages": 36,
312
  "shards": 2
313
  },
314
  {
315
+ "seconds": 1.0391054690007877,
316
+ "bytes": 2626145,
317
+ "pages": 36,
318
  "shards": 2
319
  },
320
  {
321
+ "seconds": 1.0629035870006192,
322
+ "bytes": 10213495,
323
+ "pages": 32,
324
  "shards": 3
325
  },
326
  {
327
+ "seconds": 0.9228753829993366,
328
+ "bytes": 2415467,
 
 
 
 
 
 
329
  "pages": 27,
330
  "shards": 2
331
  },
332
  {
333
+ "seconds": 1.343091220001952,
334
+ "bytes": 11741900,
335
+ "pages": 34,
336
  "shards": 3
337
  },
338
  {
339
+ "seconds": 1.2118571440005326,
340
+ "bytes": 6204328,
341
+ "pages": 36,
342
+ "shards": 2
343
  },
344
  {
345
+ "seconds": 1.1613783780012454,
346
+ "bytes": 6337641,
347
+ "pages": 34,
348
+ "shards": 3
349
  },
350
  {
351
+ "seconds": 1.4139873040003295,
352
+ "bytes": 5988516,
353
+ "pages": 36,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 1.5244087760002003,
358
+ "bytes": 914665,
359
+ "pages": 27,
360
+ "shards": 1
361
  },
362
  {
363
+ "seconds": 1.1577779409999494,
364
+ "bytes": 7413917,
365
+ "pages": 37,
366
  "shards": 3
367
  },
368
  {
369
+ "seconds": 1.1579117699984636,
370
+ "bytes": 10571255,
 
 
 
 
 
 
371
  "pages": 29,
372
  "shards": 3
373
  },
374
  {
375
+ "seconds": 1.0644437480004854,
376
+ "bytes": 8538245,
377
+ "pages": 40,
378
+ "shards": 3
379
  },
380
  {
381
+ "seconds": 1.3977851220006414,
382
+ "bytes": 6073631,
383
  "pages": 36,
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 1.1060321420009132,
388
+ "bytes": 6015087,
389
  "pages": 27,
390
  "shards": 2
391
  },
392
  {
393
+ "seconds": 1.2197453389999282,
394
+ "bytes": 6685693,
395
+ "pages": 36,
396
+ "shards": 3
397
  },
398
  {
399
+ "seconds": 1.2612095350013988,
400
+ "bytes": 6299046,
401
+ "pages": 36,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 0.8313243659977161,
406
+ "bytes": 944700,
407
+ "pages": 28,
408
+ "shards": 2
409
  },
410
  {
411
+ "seconds": 1.2853576430024987,
412
+ "bytes": 13551063,
413
+ "pages": 43,
414
+ "shards": 4
415
  },
416
  {
417
+ "seconds": 1.137746813998092,
418
+ "bytes": 10438351,
419
+ "pages": 27,
420
+ "shards": 3
421
  },
422
  {
423
+ "seconds": 0.9337895320022653,
424
+ "bytes": 2628632,
425
+ "pages": 37,
426
+ "shards": 2
427
  },
428
  {
429
+ "seconds": 0.9628497100020468,
430
+ "bytes": 5747199,
431
+ "pages": 36,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 0.9260915579980065,
436
+ "bytes": 968750,
437
  "pages": 30,
438
+ "shards": 1
439
  },
440
  {
441
+ "seconds": 1.5047391669977515,
442
+ "bytes": 11683531,
443
+ "pages": 33,
444
  "shards": 4
445
  },
446
  {
447
+ "seconds": 1.2526458680003998,
448
+ "bytes": 12133817,
449
+ "pages": 37,
450
+ "shards": 3
451
  },
452
  {
453
+ "seconds": 0.8809280429995852,
454
+ "bytes": 2767942,
455
+ "pages": 36,
456
  "shards": 2
457
  },
458
  {
459
+ "seconds": 1.0177543880017765,
460
+ "bytes": 5824791,
461
+ "pages": 28,
462
+ "shards": 2
463
  },
464
  {
465
+ "seconds": 1.107241077999788,
466
+ "bytes": 9747301,
467
+ "pages": 37,
468
  "shards": 3
469
  },
470
  {
471
+ "seconds": 0.9869726110009651,
472
+ "bytes": 2644359,
473
+ "pages": 36,
474
+ "shards": 2
475
  },
476
  {
477
+ "seconds": 1.3357893329994113,
478
+ "bytes": 5825933,
479
+ "pages": 29,
480
+ "shards": 2
481
  },
482
  {
483
+ "seconds": 1.0486931550003646,
484
+ "bytes": 6833428,
485
+ "pages": 34,
486
+ "shards": 2
487
  },
488
  {
489
+ "seconds": 1.242159831999743,
490
+ "bytes": 12337974,
491
+ "pages": 35,
492
+ "shards": 4
493
  },
494
  {
495
+ "seconds": 1.3698127219977323,
496
+ "bytes": 6046909,
497
+ "pages": 37,
498
+ "shards": 2
499
  },
500
  {
501
+ "seconds": 1.4521962930011796,
502
+ "bytes": 2626843,
503
+ "pages": 37,
504
+ "shards": 2
505
  },
506
  {
507
+ "seconds": 1.2158071900012146,
508
+ "bytes": 5739703,
509
+ "pages": 27,
510
  "shards": 2
511
  },
512
  {
513
+ "seconds": 1.3039640329989197,
514
+ "bytes": 9738932,
515
  "pages": 37,
516
  "shards": 3
517
  },
518
  {
519
+ "seconds": 1.1697420529999363,
520
+ "bytes": 8456264,
521
+ "pages": 36,
522
  "shards": 3
523
  },
524
  {
525
+ "seconds": 1.2002514119994885,
526
+ "bytes": 5878811,
527
+ "pages": 29,
528
+ "shards": 2
529
  },
530
  {
531
+ "seconds": 1.0716100809986528,
532
+ "bytes": 2567688,
533
  "pages": 36,
534
+ "shards": 2
535
  },
536
  {
537
+ "seconds": 1.52958795199811,
538
+ "bytes": 10493875,
539
+ "pages": 31,
540
+ "shards": 3
541
  },
542
  {
543
+ "seconds": 1.2153502640030638,
544
+ "bytes": 6178782,
545
+ "pages": 35,
546
+ "shards": 2
547
  },
548
  {
549
+ "seconds": 1.1377685129991733,
550
+ "bytes": 2613014,
551
+ "pages": 37,
552
  "shards": 2
553
  },
554
  {
555
+ "seconds": 1.2625757009991503,
556
+ "bytes": 11728601,
557
+ "pages": 32,
558
+ "shards": 3
559
+ },
560
+ {
561
+ "seconds": 0.9978417950005678,
562
+ "bytes": 5753379,
563
+ "pages": 30,
564
  "shards": 2
565
  }
566
  ],
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fbe139ff1c427ab9b5843d95ec3ecec658133f721c1f794977096107acee4acb
3
- size 721764543
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:124b35ad649617fdf90d8ce9db8b59458f1bc588be0bbe9ced8a83035f7877a9
3
+ size 721760127
training_config.json CHANGED
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 62,
50
- "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,
 
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
+ "prefetch_batches": 61,
50
+ "prefetch_low_water": 30,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,