appvoid commited on
Commit
9c3f218
·
verified ·
1 Parent(s): 55519d1

Qwen90M step 360000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 355000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "e07e1e6496a6b48cf11640c7168b06232c8e131cc65408e66ebbe90be5fcabf0",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "27deb3f82952932a7dca09467002556e1f8621d42020878ad6f833309e2c4d60",
12
- "training_config.json": "a9be8fd9bf1c8d33aae0c56259fc899809df0d3ff60285ef0a4c2ea73471632c",
13
- "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a",
14
- "shard_scheduler.json": "a586d4c8276b86561726c239519a16e3d56185f24feafbd8f72991b1ac951402",
15
- "coverage.json": "c48347a79685e76cda2c7abe60994cb73ba2a2da3ecf8adff831efd7fc00c98b",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 360000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "aefca4aceeb21f90da9c36c76bd5a05ccb0e95d3e2344dbae626f2f7f0fcc916",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "ef5519c0d2203b921c794e8e2cec4418be2e30e6f0ee6e640ab5ca3fc6c954f5",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
+ "performance.json": "e3923fa86352f6d878fd8d37b6b9fc374577f52be522203b0b865badf630da15",
14
+ "shard_scheduler.json": "83f1c90fc05480fdbcecae8e2db98bf4bb3dc03eaa8272164f0d6fe631c29fab",
15
+ "coverage.json": "f2b2198c10a4d312de6292dca73b81effe885572010ec59de71866b8e4bd577d",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 355000,
3
  "stats": {
4
- "climbmix/blocks": 12780000,
5
- "climbmix/targets": 21363361651,
6
- "climbmix/input_tokens": 21364715661,
7
- "ultra_qa/blocks": 4792500,
8
- "ultra_qa/targets": 7818219856,
9
- "ultra_qa/input_tokens": 7818234963,
10
- "cortex/blocks": 4792500,
11
- "cortex/targets": 182318776,
12
- "cortex/input_tokens": 7618501449,
13
- "ultra_style/blocks": 4792500,
14
- "ultra_style/targets": 7580344733,
15
- "ultra_style/input_tokens": 7580742835,
16
- "openmath/blocks": 3195000,
17
- "openmath/targets": 4937114048,
18
- "openmath/input_tokens": 4938759767,
19
- "rewrite6/blocks": 1597500,
20
- "rewrite6/targets": 2618935783,
21
- "rewrite6/input_tokens": 2621807196
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1750,
26
- "row": 2459,
27
  "start": {
28
- "file": 84,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 58,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 7170459
38
  },
39
  "rewrite6": {
40
- "page": 45,
41
- "row": 3686,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 184320
47
  }
48
  },
49
- "epoch": 19,
50
  "offset": 0,
51
- "documents": 4090682
52
  },
53
  "openmath": {
54
- "page": 1094,
55
- "row": 149,
56
  "start": {
57
  "file": 10,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 114,
61
- "row": 524
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 4481173
67
  },
68
  "ultra_qa": {
69
- "page": 472,
70
- "row": 3902,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 5,
76
- "row": 51451
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 1937214
82
  },
83
  "ultra_style": {
84
- "page": 722,
85
- "row": 802,
86
  "start": {
87
  "file": 4,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 2,
91
- "row": 47149
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 2958114
97
  },
98
  "cortex": {
99
- "page": 92951,
100
- "row": 78,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 11897806
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 360000,
3
  "stats": {
4
+ "climbmix/blocks": 12960000,
5
+ "climbmix/targets": 21664115367,
6
+ "climbmix/input_tokens": 21665488277,
7
+ "ultra_qa/blocks": 4860000,
8
+ "ultra_qa/targets": 7928409470,
9
+ "ultra_qa/input_tokens": 7928424783,
10
+ "cortex/blocks": 4860000,
11
+ "cortex/targets": 184879230,
12
+ "cortex/input_tokens": 7725738038,
13
+ "ultra_style/blocks": 4860000,
14
+ "ultra_style/targets": 7687434969,
15
+ "ultra_style/input_tokens": 7687838419,
16
+ "openmath/blocks": 3240000,
17
+ "openmath/targets": 5006598820,
18
+ "openmath/input_tokens": 5008267424,
19
+ "rewrite6/blocks": 1620000,
20
+ "rewrite6/targets": 2655769484,
21
+ "rewrite6/input_tokens": 2658681060
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1775,
26
+ "row": 1037,
27
  "start": {
28
+ "file": 85,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 75,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 7271437
38
  },
39
  "rewrite6": {
40
+ "page": 9,
41
+ "row": 3028,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 36864
47
  }
48
  },
49
+ "epoch": 20,
50
  "offset": 0,
51
+ "documents": 4147972
52
  },
53
  "openmath": {
54
+ "page": 1109,
55
+ "row": 1368,
56
  "start": {
57
  "file": 10,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 175,
61
+ "row": 964
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 4543832
67
  },
68
  "ultra_qa": {
69
+ "page": 479,
70
+ "row": 2483,
71
  "start": {
72
  "file": 3,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 6,
76
+ "row": 15507
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 1964467
82
  },
83
  "ultra_style": {
84
+ "page": 732,
85
+ "row": 901,
86
  "start": {
87
  "file": 4,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 3,
91
+ "row": 3439
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 2999173
97
  },
98
  "cortex": {
99
+ "page": 94259,
100
+ "row": 79,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 12065231
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 355000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "e07e1e6496a6b48cf11640c7168b06232c8e131cc65408e66ebbe90be5fcabf0", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "27deb3f82952932a7dca09467002556e1f8621d42020878ad6f833309e2c4d60", "training_config.json": "a9be8fd9bf1c8d33aae0c56259fc899809df0d3ff60285ef0a4c2ea73471632c", "performance.json": "90ad39d684ac792f235cbfbeef498d4c81f09977d8f89fb8c3bdd41e515f347a", "shard_scheduler.json": "a586d4c8276b86561726c239519a16e3d56185f24feafbd8f72991b1ac951402", "coverage.json": "c48347a79685e76cda2c7abe60994cb73ba2a2da3ecf8adff831efd7fc00c98b", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000355000"}
 
1
+ {"step": 360000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "aefca4aceeb21f90da9c36c76bd5a05ccb0e95d3e2344dbae626f2f7f0fcc916", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "ef5519c0d2203b921c794e8e2cec4418be2e30e6f0ee6e640ab5ca3fc6c954f5", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "e3923fa86352f6d878fd8d37b6b9fc374577f52be522203b0b865badf630da15", "shard_scheduler.json": "83f1c90fc05480fdbcecae8e2db98bf4bb3dc03eaa8272164f0d6fe631c29fab", "coverage.json": "f2b2198c10a4d312de6292dca73b81effe885572010ec59de71866b8e4bd577d", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000360000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e07e1e6496a6b48cf11640c7168b06232c8e131cc65408e66ebbe90be5fcabf0
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aefca4aceeb21f90da9c36c76bd5a05ccb0e95d3e2344dbae626f2f7f0fcc916
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9004576890000067,
53
  "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9568634569999972,
56
- "input_tokens_per_second": 195598.30756245417
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9009280120000085,
53
  "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9565700770000376,
56
+ "input_tokens_per_second": 195496.19687038695
57
  },
58
  "measurements": [
59
  {
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": 73436.5976740253,
26
- "required_pages_per_second_upper_estimate": 0.006365298163029331
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": 9179.574709253162,
40
- "required_pages_per_second_upper_estimate": 0.003661869598393634
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": 18359.149418506324,
54
- "required_pages_per_second_upper_estimate": 0.0042470661079772485
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": 27538.724127759484,
68
- "required_pages_per_second_upper_estimate": 0.0016769638148614063
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": 27538.724127759484,
82
- "required_pages_per_second_upper_estimate": 0.00212000721542238
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": 21,
95
- "required_token_positions_per_second": 27538.724127759484,
96
- "required_pages_per_second_upper_estimate": 0.3419387874859938
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,413 +155,149 @@
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": 60,
159
  "prefetch_capacity_limit": 512,
160
- "batch_memory_estimate_bytes": 690670,
161
- "producer_batch_seconds": 0.41315269255511566,
162
- "producer_p95_seconds": 0.4113726069990662,
163
- "gpu_update_seconds": 1.0039680804286195,
164
- "producer_batches_per_second": 2.4204126416690297,
165
- "gpu_batches_per_second": 0.9960476030005601,
166
- "producer_headroom": 2.43001703370162,
167
  "sustainable": true,
168
- "buffer_seconds": 60.23808482571717,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 84,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 324828.29635209485,
174
- "upload_bytes_per_second": 5287585.875586291,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 69.70059078142295,
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": 0.9463060489979398,
184
- "bytes": 743440,
185
- "pages": 25,
186
  "shards": 1
187
  },
188
  {
189
- "seconds": 1.0638403249977273,
190
- "bytes": 11964464,
191
- "pages": 46,
192
- "shards": 4
193
- },
194
- {
195
- "seconds": 1.8820245999959297,
196
- "bytes": 865291,
197
- "pages": 27,
198
- "shards": 1
199
- },
200
- {
201
- "seconds": 1.3777121140010422,
202
- "bytes": 7459243,
203
- "pages": 37,
204
- "shards": 3
205
- },
206
- {
207
- "seconds": 2.28389226100262,
208
- "bytes": 15733663,
209
- "pages": 34,
210
- "shards": 5
211
- },
212
- {
213
- "seconds": 0.7980627049983013,
214
- "bytes": 1094775,
215
- "pages": 34,
216
- "shards": 1
217
- },
218
- {
219
- "seconds": 1.0217303430035827,
220
- "bytes": 7986291,
221
- "pages": 39,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 0.8759991519982577,
226
- "bytes": 805335,
227
- "pages": 26,
228
- "shards": 1
229
- },
230
- {
231
- "seconds": 1.0561498970055254,
232
- "bytes": 10650758,
233
- "pages": 34,
234
- "shards": 3
235
- },
236
- {
237
- "seconds": 1.2883731410038308,
238
- "bytes": 7659801,
239
- "pages": 38,
240
- "shards": 3
241
- },
242
- {
243
- "seconds": 1.1231011929994565,
244
- "bytes": 6897324,
245
  "pages": 35,
246
- "shards": 2
247
- },
248
- {
249
- "seconds": 1.0080412499955855,
250
- "bytes": 7466769,
251
- "pages": 29,
252
  "shards": 3
253
  },
254
  {
255
- "seconds": 1.1737449919965002,
256
- "bytes": 4963940,
257
- "pages": 26,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.0047582629995304,
262
- "bytes": 6204635,
263
- "pages": 42,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 1.2032458940011566,
268
- "bytes": 7462591,
269
  "pages": 34,
270
- "shards": 3
271
- },
272
- {
273
- "seconds": 1.0979009180009598,
274
- "bytes": 1059989,
275
- "pages": 33,
276
  "shards": 1
277
  },
278
  {
279
- "seconds": 1.474424235006154,
280
- "bytes": 14927789,
281
- "pages": 21,
282
- "shards": 4
283
- },
284
- {
285
- "seconds": 1.0885418239995488,
286
- "bytes": 8199005,
287
- "pages": 47,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 0.9340340310009196,
292
- "bytes": 958542,
293
  "pages": 30,
294
- "shards": 1
295
- },
296
- {
297
- "seconds": 1.0266037770052208,
298
- "bytes": 7314912,
299
- "pages": 35,
300
  "shards": 3
301
  },
302
  {
303
- "seconds": 1.1214634670031955,
304
- "bytes": 5616787,
305
- "pages": 29,
306
- "shards": 2
307
- },
308
- {
309
- "seconds": 1.0272974420004175,
310
- "bytes": 6127988,
311
- "pages": 37,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.1608434069930809,
316
- "bytes": 13413721,
317
- "pages": 38,
318
- "shards": 4
319
- },
320
- {
321
- "seconds": 0.8657637560027069,
322
- "bytes": 989817,
323
  "pages": 27,
324
- "shards": 1
325
- },
326
- {
327
- "seconds": 0.8787106930030859,
328
- "bytes": 615085,
329
- "pages": 17,
330
- "shards": 1
331
- },
332
- {
333
- "seconds": 1.1383813439970254,
334
- "bytes": 11307353,
335
- "pages": 35,
336
- "shards": 4
337
- },
338
- {
339
- "seconds": 0.9193243289992097,
340
- "bytes": 5971902,
341
- "pages": 36,
342
- "shards": 2
343
- },
344
- {
345
- "seconds": 0.7975082029952318,
346
- "bytes": 1112881,
347
- "pages": 34,
348
- "shards": 1
349
- },
350
- {
351
- "seconds": 1.220746667997446,
352
- "bytes": 11991108,
353
- "pages": 31,
354
- "shards": 4
355
- },
356
- {
357
- "seconds": 7.638744584997767,
358
- "bytes": 11951922,
359
- "pages": 34,
360
- "shards": 3
361
- },
362
- {
363
- "seconds": 0.8686913920028019,
364
- "bytes": 2634908,
365
- "pages": 37,
366
  "shards": 2
367
  },
368
  {
369
- "seconds": 1.1666307280029287,
370
- "bytes": 5700254,
371
- "pages": 28,
372
- "shards": 2
373
- },
374
- {
375
- "seconds": 0.8491491380045773,
376
- "bytes": 1087747,
377
- "pages": 34,
378
- "shards": 1
379
- },
380
- {
381
- "seconds": 1.1218249969970202,
382
- "bytes": 11930213,
383
- "pages": 36,
384
- "shards": 4
385
- },
386
- {
387
- "seconds": 1.2040022809960647,
388
- "bytes": 6093346,
389
  "pages": 35,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 0.9920474090031348,
394
- "bytes": 6801926,
395
- "pages": 31,
396
- "shards": 3
397
- },
398
- {
399
- "seconds": 1.4719860830009566,
400
- "bytes": 7846458,
401
- "pages": 39,
402
- "shards": 3
403
- },
404
- {
405
- "seconds": 0.9881938749967958,
406
- "bytes": 4661656,
407
- "pages": 32,
408
- "shards": 2
409
- },
410
- {
411
- "seconds": 1.2476324340022984,
412
- "bytes": 7151046,
413
- "pages": 30,
414
- "shards": 3
415
- },
416
- {
417
- "seconds": 0.9890799559943844,
418
- "bytes": 6390118,
419
  "pages": 36,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.1548123519969522,
424
- "bytes": 1183539,
425
- "pages": 34,
426
- "shards": 1
427
- },
428
- {
429
- "seconds": 1.2286420309974346,
430
- "bytes": 17809213,
431
  "pages": 32,
432
- "shards": 5
433
  },
434
  {
435
- "seconds": 0.9948159990017302,
436
- "bytes": 1198699,
437
  "pages": 36,
438
- "shards": 1
439
- },
440
- {
441
- "seconds": 1.0069344939984148,
442
- "bytes": 6086180,
443
- "pages": 31,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 1.1158472850002,
448
- "bytes": 7287693,
449
- "pages": 36,
450
- "shards": 3
451
- },
452
- {
453
- "seconds": 1.026877347001573,
454
- "bytes": 5561907,
455
- "pages": 34,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 1.067893715997343,
460
- "bytes": 7222726,
461
- "pages": 30,
462
- "shards": 3
463
- },
464
- {
465
- "seconds": 1.3662753429962322,
466
- "bytes": 11997986,
467
- "pages": 39,
468
  "shards": 3
469
  },
470
  {
471
- "seconds": 0.77281211099762,
472
- "bytes": 859716,
473
- "pages": 26,
474
- "shards": 1
475
- },
476
- {
477
- "seconds": 1.1721114309984841,
478
- "bytes": 11272182,
479
- "pages": 38,
480
  "shards": 4
481
  },
482
  {
483
- "seconds": 0.8626750620023813,
484
- "bytes": 1117608,
485
- "pages": 35,
486
- "shards": 1
487
- },
488
- {
489
- "seconds": 0.9938163240003632,
490
- "bytes": 5819862,
491
- "pages": 27,
492
  "shards": 2
493
  },
494
  {
495
- "seconds": 1.1086062760005007,
496
- "bytes": 7618244,
497
  "pages": 35,
498
- "shards": 3
499
- },
500
- {
501
- "seconds": 1.1348141839989694,
502
- "bytes": 11320691,
503
- "pages": 34,
504
- "shards": 3
505
- },
506
- {
507
- "seconds": 0.981941227997595,
508
- "bytes": 7228716,
509
- "pages": 30,
510
- "shards": 3
511
  },
512
  {
513
- "seconds": 0.914243400999112,
514
- "bytes": 1165448,
515
- "pages": 36,
516
- "shards": 1
517
  },
518
  {
519
- "seconds": 1.0446478919984656,
520
- "bytes": 5999613,
521
- "pages": 36,
522
  "shards": 2
523
  },
524
  {
525
- "seconds": 1.4630355209956178,
526
- "bytes": 11583159,
527
- "pages": 30,
528
  "shards": 4
529
  },
530
  {
531
- "seconds": 1.1508738010015804,
532
- "bytes": 1166291,
533
  "pages": 35,
534
- "shards": 1
535
- },
536
- {
537
- "seconds": 1.4934369809998316,
538
- "bytes": 13370474,
539
- "pages": 36,
540
- "shards": 4
541
- },
542
- {
543
- "seconds": 0.8853696480000508,
544
- "bytes": 1220951,
545
- "pages": 34,
546
- "shards": 1
547
- },
548
- {
549
- "seconds": 2.1309150849992875,
550
- "bytes": 9524825,
551
- "pages": 30,
552
- "shards": 3
553
  },
554
  {
555
- "seconds": 1.1543053290006355,
556
- "bytes": 7653463,
557
- "pages": 39,
558
  "shards": 3
559
- },
560
- {
561
- "seconds": 1.0381377000012435,
562
- "bytes": 839375,
563
- "pages": 26,
564
- "shards": 1
565
  }
566
  ],
567
  "cache_key": {
@@ -606,7 +342,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 2.358020696999972,
610
- 0.013266629999975521
611
  ]
612
  }
 
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": 75665.69836686787,
26
+ "required_pages_per_second_upper_estimate": 0.00655851096148087
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": 9458.212295858484,
40
+ "required_pages_per_second_upper_estimate": 0.003773022297693667
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": 18916.42459171697,
54
+ "required_pages_per_second_upper_estimate": 0.004375981911591458
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": 28374.636887575452,
68
+ "required_pages_per_second_upper_estimate": 0.0017278665162388923
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": 28374.636887575452,
82
+ "required_pages_per_second_upper_estimate": 0.002184358093627632
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": 22,
95
+ "required_token_positions_per_second": 28374.636887575452,
96
+ "required_pages_per_second_upper_estimate": 0.3523180263428667
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": 62,
159
  "prefetch_capacity_limit": 512,
160
+ "batch_memory_estimate_bytes": 717798,
161
+ "producer_batch_seconds": 0.13129274778125932,
162
+ "producer_p95_seconds": 0.1682472989996313,
163
+ "gpu_update_seconds": 0.9743913238271736,
164
+ "producer_batches_per_second": 7.61656692314836,
165
+ "gpu_batches_per_second": 1.026281716130478,
166
+ "producer_headroom": 7.421516727264794,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.412262077284765,
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": 351443.47014520573,
174
+ "upload_bytes_per_second": 5528607.3444076935,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 75.41158750788811,
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": 0.8837165979998645,
184
+ "bytes": 52118,
185
+ "pages": 1,
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 1.4180460719999246,
190
+ "bytes": 7624068,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
191
  "pages": 35,
 
 
 
 
 
 
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 1.001895477999824,
196
+ "bytes": 1118907,
 
 
 
 
 
 
 
 
 
 
 
 
197
  "pages": 34,
 
 
 
 
 
 
198
  "shards": 1
199
  },
200
  {
201
+ "seconds": 1.186168517999704,
202
+ "bytes": 7790528,
203
+ "pages": 37,
 
 
 
 
 
 
204
  "shards": 3
205
  },
206
  {
207
+ "seconds": 1.008963309000137,
208
+ "bytes": 10351006,
209
  "pages": 30,
 
 
 
 
 
 
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 0.8917956289997164,
214
+ "bytes": 6042604,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
215
  "pages": 27,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.600639215999763,
220
+ "bytes": 2577832,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
221
  "pages": 35,
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 0.8447169340001892,
226
+ "bytes": 6262278,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
227
  "pages": 36,
228
  "shards": 2
229
  },
230
  {
231
+ "seconds": 1.2238314549999814,
232
+ "bytes": 9984108,
 
 
 
 
 
 
233
  "pages": 32,
234
+ "shards": 3
235
  },
236
  {
237
+ "seconds": 0.8533821090004494,
238
+ "bytes": 2621863,
239
  "pages": 36,
 
 
 
 
 
 
240
  "shards": 2
241
  },
242
  {
243
+ "seconds": 1.0488426049996633,
244
+ "bytes": 5758950,
245
+ "pages": 26,
 
 
 
 
 
 
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.0318703240000104,
250
+ "bytes": 8301935,
251
+ "pages": 33,
 
 
 
 
 
 
252
  "shards": 3
253
  },
254
  {
255
+ "seconds": 1.1579725150004379,
256
+ "bytes": 10569380,
257
+ "pages": 37,
 
 
 
 
 
 
258
  "shards": 4
259
  },
260
  {
261
+ "seconds": 1.112220934999641,
262
+ "bytes": 6262678,
263
+ "pages": 34,
 
 
 
 
 
 
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 0.9345845090001603,
268
+ "bytes": 2621511,
269
  "pages": 35,
270
+ "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
271
  },
272
  {
273
+ "seconds": 0.8686625299997104,
274
+ "bytes": 5945555,
275
+ "pages": 33,
276
+ "shards": 2
277
  },
278
  {
279
+ "seconds": 1.1098429370003942,
280
+ "bytes": 4826131,
281
+ "pages": 32,
282
  "shards": 2
283
  },
284
  {
285
+ "seconds": 1.7090945019999708,
286
+ "bytes": 13118107,
287
+ "pages": 37,
288
  "shards": 4
289
  },
290
  {
291
+ "seconds": 1.577916964999531,
292
+ "bytes": 6095231,
293
  "pages": 35,
294
+ "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
295
  },
296
  {
297
+ "seconds": 1.0922540970004775,
298
+ "bytes": 6780784,
299
+ "pages": 29,
300
  "shards": 3
 
 
 
 
 
 
301
  }
302
  ],
303
  "cache_key": {
 
342
  },
343
  "cache_reused": true,
344
  "resume_probe_seconds": [
345
+ 3.301654667999969,
346
+ 0.11550889399995867
347
  ]
348
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:27deb3f82952932a7dca09467002556e1f8621d42020878ad6f833309e2c4d60
3
- size 721770687
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef5519c0d2203b921c794e8e2cec4418be2e30e6f0ee6e640ab5ca3fc6c954f5
3
+ size 721774271
training_config.json CHANGED
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 60,
50
- "prefetch_low_water": 30,
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": 62,
50
+ "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,