appvoid commited on
Commit
66ec0ab
·
verified ·
1 Parent(s): 26e6662

Qwen90M step 595000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +46 -46
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +249 -249
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 590000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "2d5cd0fbe3e3ae4f47501307fc03d8c9a67b97fce7c8cd1d26c5286378e25c0e",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "e6a73d2db07c024ebb237c7d0b6fcb773796b3b6ab1195a52d429d55bc86e73a",
12
- "training_config.json": "eb51b7316ffa2242848fb50069dd8c6e0ecf28534fa6202f12a9fcfdcf6b16eb",
13
  "performance.json": "b7cb16c0913422fe54a766da97356133a598f4da7e691939f51028a8bce75667",
14
- "shard_scheduler.json": "44d308ae26c6e914df1a735812c742d18304f11ef2e74faf2202b5f592eda180",
15
- "coverage.json": "33583797aa946780892ff16dcb005f14f82af20df711e2f0964d6f600512e644",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 595000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "947d55251820f157a3efb0410a2db9cfd0393475c09cc94095d69f9cbc248585",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "04c2d546da542c12c58a0435f1ba889816d6f1028daf377c51a35cccae028102",
12
+ "training_config.json": "42b690efbefcf434fbcb690b27d17a2af5e3b65c3c6aab71efa1d83d1c6a6905",
13
  "performance.json": "b7cb16c0913422fe54a766da97356133a598f4da7e691939f51028a8bce75667",
14
+ "shard_scheduler.json": "a238ae925bbc8d31568673428b064582cd86c23da8f0df6bda1ed49585c86033",
15
+ "coverage.json": "80387086e945d13c57f5522f96608f203c1992dacef7f48f7cc07e389650ef8a",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 590000,
3
  "stats": {
4
- "climbmix/blocks": 21240000,
5
- "climbmix/targets": 35505139268,
6
- "climbmix/input_tokens": 35507389756,
7
- "ultra_qa/blocks": 7965000,
8
- "ultra_qa/targets": 12994238467,
9
- "ultra_qa/input_tokens": 12994263542,
10
- "cortex/blocks": 7965000,
11
- "cortex/targets": 302869277,
12
- "cortex/input_tokens": 12660969641,
13
- "ultra_style/blocks": 7965000,
14
- "ultra_style/targets": 12598185008,
15
- "ultra_style/input_tokens": 12598846165,
16
- "openmath/blocks": 5310000,
17
- "openmath/targets": 8204944979,
18
- "openmath/input_tokens": 8207677843,
19
- "rewrite6/blocks": 2655000,
20
- "rewrite6/targets": 4352536815,
21
- "rewrite6/input_tokens": 4357308639
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 2909,
26
- "row": 1352,
27
  "start": {
28
- "file": 140,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 64,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 0,
37
- "documents": 11916616
38
  },
39
  "rewrite6": {
40
- "page": 4,
41
- "row": 3444,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 16384
47
  }
48
  },
49
  "epoch": 33,
50
- "offset": 2048,
51
- "documents": 6798160
52
  },
53
  "openmath": {
54
- "page": 1817,
55
- "row": 2899,
56
  "start": {
57
  "file": 17,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 19,
61
- "row": 382
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 0,
66
- "documents": 7445331
67
  },
68
  "ultra_qa": {
69
- "page": 786,
70
- "row": 115,
71
  "start": {
72
  "file": 6,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
- "row": 37192
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 4096,
81
- "documents": 3219571
82
  },
83
  "ultra_style": {
84
- "page": 1199,
85
- "row": 3499,
86
  "start": {
87
  "file": 7,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 1,
91
- "row": 31050
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 4914603
97
  },
98
  "cortex": {
99
- "page": 154475,
100
- "row": 120,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 19772920
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 595000,
3
  "stats": {
4
+ "climbmix/blocks": 21420000,
5
+ "climbmix/targets": 35806090870,
6
+ "climbmix/input_tokens": 35808360523,
7
+ "ultra_qa/blocks": 8032500,
8
+ "ultra_qa/targets": 13104381210,
9
+ "ultra_qa/input_tokens": 13104406506,
10
+ "cortex/blocks": 8032500,
11
+ "cortex/targets": 305435983,
12
+ "cortex/input_tokens": 12768242199,
13
+ "ultra_style/blocks": 8032500,
14
+ "ultra_style/targets": 12704953076,
15
+ "ultra_style/input_tokens": 12705619593,
16
+ "openmath/blocks": 5355000,
17
+ "openmath/targets": 8274625787,
18
+ "openmath/input_tokens": 8277382013,
19
+ "rewrite6/blocks": 2677500,
20
+ "rewrite6/targets": 4389363744,
21
+ "rewrite6/input_tokens": 4394176236
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 2934,
26
+ "row": 289,
27
  "start": {
28
+ "file": 141,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 81,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 0,
37
+ "documents": 12017953
38
  },
39
  "rewrite6": {
40
+ "page": 18,
41
+ "row": 3998,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 73728
47
  }
48
  },
49
  "epoch": 33,
50
+ "offset": 0,
51
+ "documents": 6856058
52
  },
53
  "openmath": {
54
+ "page": 1833,
55
+ "row": 642,
56
  "start": {
57
  "file": 17,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 84,
61
+ "row": 918
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 2048,
66
+ "documents": 7508610
67
  },
68
  "ultra_qa": {
69
+ "page": 792,
70
+ "row": 2870,
71
  "start": {
72
  "file": 6,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 1,
76
+ "row": 61768
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 4096,
81
+ "documents": 3246902
82
  },
83
  "ultra_style": {
84
+ "page": 1209,
85
+ "row": 3875,
86
  "start": {
87
  "file": 7,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 1,
91
+ "row": 72010
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 4955939
97
  },
98
  "cortex": {
99
+ "page": 155783,
100
+ "row": 98,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 19940322
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 590000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "2d5cd0fbe3e3ae4f47501307fc03d8c9a67b97fce7c8cd1d26c5286378e25c0e", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "e6a73d2db07c024ebb237c7d0b6fcb773796b3b6ab1195a52d429d55bc86e73a", "training_config.json": "eb51b7316ffa2242848fb50069dd8c6e0ecf28534fa6202f12a9fcfdcf6b16eb", "performance.json": "b7cb16c0913422fe54a766da97356133a598f4da7e691939f51028a8bce75667", "shard_scheduler.json": "44d308ae26c6e914df1a735812c742d18304f11ef2e74faf2202b5f592eda180", "coverage.json": "33583797aa946780892ff16dcb005f14f82af20df711e2f0964d6f600512e644", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000590000"}
 
1
+ {"step": 595000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "947d55251820f157a3efb0410a2db9cfd0393475c09cc94095d69f9cbc248585", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "04c2d546da542c12c58a0435f1ba889816d6f1028daf377c51a35cccae028102", "training_config.json": "42b690efbefcf434fbcb690b27d17a2af5e3b65c3c6aab71efa1d83d1c6a6905", "performance.json": "b7cb16c0913422fe54a766da97356133a598f4da7e691939f51028a8bce75667", "shard_scheduler.json": "a238ae925bbc8d31568673428b064582cd86c23da8f0df6bda1ed49585c86033", "coverage.json": "80387086e945d13c57f5522f96608f203c1992dacef7f48f7cc07e389650ef8a", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000595000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2d5cd0fbe3e3ae4f47501307fc03d8c9a67b97fce7c8cd1d26c5286378e25c0e
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:947d55251820f157a3efb0410a2db9cfd0393475c09cc94095d69f9cbc248585
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": 77450.23149600206,
26
- "required_pages_per_second_upper_estimate": 0.005358186724189602
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": 9681.278937000257,
40
- "required_pages_per_second_upper_estimate": 0.004053349322828503
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": 19362.557874000515,
54
- "required_pages_per_second_upper_estimate": 0.004032710180165127
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": 29043.83681100077,
68
- "required_pages_per_second_upper_estimate": 0.0017411336694484819
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": 29043.83681100077,
82
- "required_pages_per_second_upper_estimate": 0.00225803056171825
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 23,
95
- "required_token_positions_per_second": 29043.83681100077,
96
- "required_pages_per_second_upper_estimate": 0.37254315376920216
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": 704566,
161
- "producer_batch_seconds": 0.09924472569522891,
162
- "producer_p95_seconds": 0.2953243410011055,
163
- "gpu_update_seconds": 0.9519403438297768,
164
- "producer_batches_per_second": 10.076102210921562,
165
- "gpu_batches_per_second": 1.050485995768257,
166
- "producer_headroom": 9.591848203128645,
167
  "sustainable": true,
168
- "buffer_seconds": 60.92418200510571,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
  "upload_pages_per_commit": 87,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 464883.5979835728,
174
- "upload_bytes_per_second": 4976341.388513764,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 99.75319819097889,
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": 1.124440429997776,
184
- "bytes": 5889635,
185
- "pages": 28,
186
  "shards": 2
187
  },
188
  {
189
- "seconds": 1.0922325869978522,
190
- "bytes": 6027814,
191
- "pages": 36,
192
- "shards": 2
193
  },
194
  {
195
- "seconds": 1.0737022979992616,
196
- "bytes": 6591396,
197
- "pages": 31,
198
  "shards": 3
199
  },
200
  {
201
- "seconds": 1.1665754960013146,
202
- "bytes": 11803743,
203
- "pages": 35,
 
 
 
 
 
 
204
  "shards": 3
205
  },
206
  {
207
- "seconds": 0.962029407000955,
208
- "bytes": 2665696,
209
- "pages": 38,
210
  "shards": 2
211
  },
212
  {
213
- "seconds": 1.1741747699998086,
214
- "bytes": 5919198,
215
- "pages": 26,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 1.178343194998888,
220
- "bytes": 10619543,
221
- "pages": 42,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 1.036487166002189,
226
- "bytes": 2690894,
 
 
 
 
 
 
 
 
 
 
 
 
227
  "pages": 37,
228
  "shards": 2
229
  },
230
  {
231
- "seconds": 1.1432905639994715,
232
- "bytes": 5745742,
233
- "pages": 28,
234
- "shards": 2
235
  },
236
  {
237
- "seconds": 1.232796011998289,
238
- "bytes": 13189628,
239
- "pages": 37,
240
  "shards": 4
241
  },
242
  {
243
- "seconds": 0.943955818998802,
244
- "bytes": 5281939,
245
- "pages": 35,
246
  "shards": 2
247
  },
248
  {
249
- "seconds": 1.0029552209998656,
250
- "bytes": 6176365,
251
- "pages": 37,
252
- "shards": 2
253
  },
254
  {
255
- "seconds": 1.1109868420026032,
256
- "bytes": 2592499,
257
- "pages": 35,
258
  "shards": 2
259
  },
260
  {
261
- "seconds": 1.24596934100191,
262
- "bytes": 5407802,
263
- "pages": 21,
264
- "shards": 2
265
  },
266
  {
267
- "seconds": 1.361496919002093,
268
- "bytes": 10822221,
269
  "pages": 38,
270
- "shards": 3
271
  },
272
  {
273
- "seconds": 1.3625388020009268,
274
- "bytes": 8506674,
275
- "pages": 41,
276
  "shards": 3
277
  },
278
  {
279
- "seconds": 1.141177229997993,
280
- "bytes": 6001543,
281
- "pages": 29,
282
  "shards": 2
283
  },
284
  {
285
- "seconds": 0.9294679570011795,
286
- "bytes": 2588442,
 
 
 
 
 
 
287
  "pages": 35,
288
- "shards": 2
289
  },
290
  {
291
- "seconds": 1.1605264330028149,
292
- "bytes": 9613122,
293
- "pages": 38,
294
- "shards": 4
295
  },
296
  {
297
- "seconds": 1.0401736989988422,
298
- "bytes": 6081990,
299
  "pages": 37,
300
  "shards": 2
301
  },
302
  {
303
- "seconds": 0.9755379390007874,
304
- "bytes": 2340646,
305
- "pages": 28,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 1.4557110550013022,
310
- "bytes": 16684165,
311
- "pages": 36,
312
- "shards": 4
313
- },
314
- {
315
- "seconds": 1.2760817240014148,
316
- "bytes": 7386323,
317
  "pages": 39,
318
- "shards": 3
319
  },
320
  {
321
- "seconds": 1.0108026320012868,
322
- "bytes": 815856,
323
- "pages": 26,
324
  "shards": 1
325
  },
326
  {
327
- "seconds": 0.99785261299985,
328
- "bytes": 326991,
329
- "pages": 9,
330
- "shards": 1
331
  },
332
  {
333
- "seconds": 1.214567277002061,
334
- "bytes": 5649266,
335
- "pages": 28,
336
  "shards": 2
337
  },
338
  {
339
- "seconds": 1.2461040010020952,
340
- "bytes": 11349081,
341
- "pages": 37,
342
- "shards": 4
343
- },
344
- {
345
- "seconds": 1.0920391100007691,
346
- "bytes": 1249324,
347
- "pages": 36,
348
- "shards": 1
349
  },
350
  {
351
- "seconds": 1.1804772159994172,
352
- "bytes": 13462589,
353
- "pages": 36,
354
  "shards": 4
355
  },
356
  {
357
- "seconds": 0.8390263229994162,
358
- "bytes": 877453,
359
- "pages": 26,
360
  "shards": 1
361
  },
362
  {
363
- "seconds": 1.1113367050020315,
364
- "bytes": 10487773,
365
  "pages": 35,
366
- "shards": 3
367
  },
368
  {
369
- "seconds": 1.1902512640008354,
370
- "bytes": 7677939,
371
- "pages": 39,
372
- "shards": 3
373
  },
374
  {
375
- "seconds": 1.0722998880009982,
376
- "bytes": 987240,
377
- "pages": 32,
378
  "shards": 1
379
  },
380
  {
381
- "seconds": 1.0394909079986974,
382
- "bytes": 6071704,
383
  "pages": 32,
384
- "shards": 2
385
  },
386
  {
387
- "seconds": 1.4701623389992164,
388
- "bytes": 17195357,
389
- "pages": 39,
390
- "shards": 5
391
  },
392
  {
393
- "seconds": 0.8250811970028735,
394
- "bytes": 1111187,
395
- "pages": 36,
396
- "shards": 1
397
  },
398
  {
399
- "seconds": 1.2127139630028978,
400
- "bytes": 7187251,
401
- "pages": 30,
402
  "shards": 3
403
  },
404
  {
405
- "seconds": 1.2223667069993098,
406
- "bytes": 5868379,
407
- "pages": 38,
408
- "shards": 2
409
  },
410
  {
411
- "seconds": 0.9594618119990628,
412
- "bytes": 5817766,
413
- "pages": 28,
414
- "shards": 2
415
  },
416
  {
417
- "seconds": 1.0792979020006896,
418
- "bytes": 7541421,
419
- "pages": 39,
420
  "shards": 3
421
  },
422
  {
423
- "seconds": 0.9343277210027736,
424
- "bytes": 6872118,
425
- "pages": 34,
426
  "shards": 2
427
  },
428
  {
429
- "seconds": 1.3191045509993273,
430
- "bytes": 10179142,
431
- "pages": 32,
432
- "shards": 3
433
- },
434
- {
435
- "seconds": 1.5255058740003733,
436
- "bytes": 7690292,
437
  "pages": 40,
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 1.0536525450006593,
442
- "bytes": 865655,
443
  "pages": 27,
444
  "shards": 1
445
  },
446
  {
447
- "seconds": 1.1389114759986114,
448
- "bytes": 7332921,
449
- "pages": 39,
450
- "shards": 3
451
  },
452
  {
453
- "seconds": 0.8940556989982724,
454
- "bytes": 4992530,
455
- "pages": 32,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 1.0987542440016114,
460
- "bytes": 12106318,
461
- "pages": 36,
462
- "shards": 3
463
  },
464
  {
465
- "seconds": 1.2125764899974456,
466
- "bytes": 7340281,
467
- "pages": 31,
468
- "shards": 3
469
  },
470
  {
471
- "seconds": 0.8264875540007779,
472
- "bytes": 1223987,
473
- "pages": 36,
474
- "shards": 1
475
  },
476
  {
477
- "seconds": 1.2477019600009953,
478
- "bytes": 12351474,
479
  "pages": 36,
480
- "shards": 5
481
  },
482
  {
483
- "seconds": 1.4946348649973515,
484
- "bytes": 6250315,
485
- "pages": 37,
486
- "shards": 2
487
  },
488
  {
489
- "seconds": 1.1129696229982073,
490
- "bytes": 1097015,
491
- "pages": 34,
492
  "shards": 1
493
  },
494
  {
495
- "seconds": 1.8527486320017488,
496
- "bytes": 12857725,
497
- "pages": 32,
498
  "shards": 4
499
  },
500
  {
501
- "seconds": 0.9267861950029328,
502
- "bytes": 5234150,
503
- "pages": 32,
504
- "shards": 2
505
- },
506
- {
507
- "seconds": 1.1693340630008606,
508
- "bytes": 7789873,
509
- "pages": 40,
510
- "shards": 3
511
- },
512
- {
513
- "seconds": 1.150511206000374,
514
- "bytes": 5980828,
515
  "pages": 34,
516
- "shards": 2
517
  },
518
  {
519
- "seconds": 0.8373393530018802,
520
- "bytes": 1013057,
521
- "pages": 28,
522
  "shards": 1
523
  },
524
  {
525
- "seconds": 1.2470868550008163,
526
- "bytes": 11490913,
527
- "pages": 38,
528
- "shards": 4
529
- },
530
- {
531
- "seconds": 13.443922794998798,
532
- "bytes": 11916771,
533
- "pages": 38,
534
  "shards": 3
535
  },
536
  {
537
- "seconds": 0.9173718489983003,
538
- "bytes": 2617778,
539
  "pages": 36,
540
  "shards": 2
541
  },
542
  {
543
- "seconds": 1.2762991070012504,
544
- "bytes": 10864789,
545
- "pages": 36,
 
 
 
 
 
 
546
  "shards": 3
547
  },
548
  {
549
- "seconds": 1.24567458599995,
550
- "bytes": 1188999,
551
- "pages": 35,
552
- "shards": 1
553
  },
554
  {
555
- "seconds": 1.4343414930008294,
556
- "bytes": 7437168,
557
- "pages": 31,
558
- "shards": 3
559
  },
560
  {
561
- "seconds": 1.4832610999983444,
562
- "bytes": 5959171,
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": 75999.11957643424,
26
+ "required_pages_per_second_upper_estimate": 0.005257795434550359
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": 9499.88994705428,
40
+ "required_pages_per_second_upper_estimate": 0.003977405540570961
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": 18999.77989410856,
54
+ "required_pages_per_second_upper_estimate": 0.003957153094052322
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": 28499.66984116284,
68
+ "required_pages_per_second_upper_estimate": 0.0017085116905015505
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": 28499.66984116284,
82
+ "required_pages_per_second_upper_estimate": 0.0022157239733508944
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 22,
95
+ "required_token_positions_per_second": 28499.66984116284,
96
+ "required_pages_per_second_upper_estimate": 0.3655631641610913
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
+ "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 704566,
161
+ "producer_batch_seconds": 0.15374018804675416,
162
+ "producer_p95_seconds": 0.36335070499626454,
163
+ "gpu_update_seconds": 0.9701165014924927,
164
+ "producer_batches_per_second": 6.504480140845727,
165
+ "gpu_batches_per_second": 1.030804030713355,
166
+ "producer_headroom": 6.310103518264653,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.14722309253455,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
  "upload_pages_per_commit": 87,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 292008.66332315706,
174
+ "upload_bytes_per_second": 4983567.38343802,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 62.65826153536829,
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": 0.9343277210027736,
184
+ "bytes": 6872118,
185
+ "pages": 34,
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 1.3191045509993273,
190
+ "bytes": 10179142,
191
+ "pages": 32,
192
+ "shards": 3
193
  },
194
  {
195
+ "seconds": 1.5255058740003733,
196
+ "bytes": 7690292,
197
+ "pages": 40,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 1.0536525450006593,
202
+ "bytes": 865655,
203
+ "pages": 27,
204
+ "shards": 1
205
+ },
206
+ {
207
+ "seconds": 1.1389114759986114,
208
+ "bytes": 7332921,
209
+ "pages": 39,
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 0.8940556989982724,
214
+ "bytes": 4992530,
215
+ "pages": 32,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.0987542440016114,
220
+ "bytes": 12106318,
221
+ "pages": 36,
222
+ "shards": 3
223
  },
224
  {
225
+ "seconds": 1.2125764899974456,
226
+ "bytes": 7340281,
227
+ "pages": 31,
228
  "shards": 3
229
  },
230
  {
231
+ "seconds": 0.8264875540007779,
232
+ "bytes": 1223987,
233
+ "pages": 36,
234
+ "shards": 1
235
+ },
236
+ {
237
+ "seconds": 1.2477019600009953,
238
+ "bytes": 12351474,
239
+ "pages": 36,
240
+ "shards": 5
241
+ },
242
+ {
243
+ "seconds": 1.4946348649973515,
244
+ "bytes": 6250315,
245
  "pages": 37,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.1129696229982073,
250
+ "bytes": 1097015,
251
+ "pages": 34,
252
+ "shards": 1
253
  },
254
  {
255
+ "seconds": 1.8527486320017488,
256
+ "bytes": 12857725,
257
+ "pages": 32,
258
  "shards": 4
259
  },
260
  {
261
+ "seconds": 0.9267861950029328,
262
+ "bytes": 5234150,
263
+ "pages": 32,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.1693340630008606,
268
+ "bytes": 7789873,
269
+ "pages": 40,
270
+ "shards": 3
271
  },
272
  {
273
+ "seconds": 1.150511206000374,
274
+ "bytes": 5980828,
275
+ "pages": 34,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 0.8373393530018802,
280
+ "bytes": 1013057,
281
+ "pages": 28,
282
+ "shards": 1
283
  },
284
  {
285
+ "seconds": 1.2470868550008163,
286
+ "bytes": 11490913,
287
  "pages": 38,
288
+ "shards": 4
289
  },
290
  {
291
+ "seconds": 13.443922794998798,
292
+ "bytes": 11916771,
293
+ "pages": 38,
294
  "shards": 3
295
  },
296
  {
297
+ "seconds": 0.9173718489983003,
298
+ "bytes": 2617778,
299
+ "pages": 36,
300
  "shards": 2
301
  },
302
  {
303
+ "seconds": 1.2762991070012504,
304
+ "bytes": 10864789,
305
+ "pages": 36,
306
+ "shards": 3
307
+ },
308
+ {
309
+ "seconds": 1.24567458599995,
310
+ "bytes": 1188999,
311
  "pages": 35,
312
+ "shards": 1
313
  },
314
  {
315
+ "seconds": 1.4343414930008294,
316
+ "bytes": 7437168,
317
+ "pages": 31,
318
+ "shards": 3
319
  },
320
  {
321
+ "seconds": 1.4832610999983444,
322
+ "bytes": 5959171,
323
  "pages": 37,
324
  "shards": 2
325
  },
326
  {
327
+ "seconds": 0.9623825809976552,
328
+ "bytes": 6329110,
329
+ "pages": 20,
330
  "shards": 2
331
  },
332
  {
333
+ "seconds": 1.5357303359996877,
334
+ "bytes": 11699991,
 
 
 
 
 
 
335
  "pages": 39,
336
+ "shards": 4
337
  },
338
  {
339
+ "seconds": 1.0364311270022881,
340
+ "bytes": 982135,
341
+ "pages": 27,
342
  "shards": 1
343
  },
344
  {
345
+ "seconds": 1.1774842960003298,
346
+ "bytes": 7504480,
347
+ "pages": 39,
348
+ "shards": 3
349
  },
350
  {
351
+ "seconds": 1.4250826019997476,
352
+ "bytes": 6016198,
353
+ "pages": 29,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 1.177016456000274,
358
+ "bytes": 5382045,
359
+ "pages": 34,
360
+ "shards": 2
 
 
 
 
 
 
361
  },
362
  {
363
+ "seconds": 1.5666041279982892,
364
+ "bytes": 13422269,
365
+ "pages": 37,
366
  "shards": 4
367
  },
368
  {
369
+ "seconds": 1.324093607996474,
370
+ "bytes": 1086878,
371
+ "pages": 34,
372
  "shards": 1
373
  },
374
  {
375
+ "seconds": 1.0723825640016003,
376
+ "bytes": 6205162,
377
  "pages": 35,
378
+ "shards": 2
379
  },
380
  {
381
+ "seconds": 1.3036935650015948,
382
+ "bytes": 11727766,
383
+ "pages": 34,
384
+ "shards": 4
385
  },
386
  {
387
+ "seconds": 0.7984904570039362,
388
+ "bytes": 1227271,
389
+ "pages": 35,
390
  "shards": 1
391
  },
392
  {
393
+ "seconds": 1.0684821660033776,
394
+ "bytes": 7548903,
395
  "pages": 32,
396
+ "shards": 3
397
  },
398
  {
399
+ "seconds": 1.2989959650003584,
400
+ "bytes": 11854087,
401
+ "pages": 38,
402
+ "shards": 3
403
  },
404
  {
405
+ "seconds": 1.0731057049997617,
406
+ "bytes": 5340693,
407
+ "pages": 33,
408
+ "shards": 2
409
  },
410
  {
411
+ "seconds": 1.125931346003199,
412
+ "bytes": 7735546,
413
+ "pages": 41,
414
  "shards": 3
415
  },
416
  {
417
+ "seconds": 0.949357431003591,
418
+ "bytes": 856694,
419
+ "pages": 26,
420
+ "shards": 1
421
  },
422
  {
423
+ "seconds": 1.058450859003642,
424
+ "bytes": 7478371,
425
+ "pages": 37,
426
+ "shards": 4
427
  },
428
  {
429
+ "seconds": 1.0343096720025642,
430
+ "bytes": 11122347,
431
+ "pages": 34,
432
  "shards": 3
433
  },
434
  {
435
+ "seconds": 1.0296640300002764,
436
+ "bytes": 6697132,
437
+ "pages": 29,
438
  "shards": 2
439
  },
440
  {
441
+ "seconds": 1.1917317950064898,
442
+ "bytes": 7979357,
 
 
 
 
 
 
443
  "pages": 40,
444
  "shards": 3
445
  },
446
  {
447
+ "seconds": 0.9077869020038634,
448
+ "bytes": 931830,
449
  "pages": 27,
450
  "shards": 1
451
  },
452
  {
453
+ "seconds": 1.1493673139993916,
454
+ "bytes": 11660613,
455
+ "pages": 43,
456
+ "shards": 4
457
  },
458
  {
459
+ "seconds": 1.1604952449997654,
460
+ "bytes": 5793744,
461
+ "pages": 28,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 0.8524275599993416,
466
+ "bytes": 1090009,
467
+ "pages": 35,
468
+ "shards": 1
469
  },
470
  {
471
+ "seconds": 1.149581274999946,
472
+ "bytes": 13144315,
473
+ "pages": 34,
474
+ "shards": 4
475
  },
476
  {
477
+ "seconds": 1.5471423709968803,
478
+ "bytes": 10740667,
479
+ "pages": 33,
480
+ "shards": 3
481
  },
482
  {
483
+ "seconds": 0.8488108419987839,
484
+ "bytes": 1104603,
485
  "pages": 36,
486
+ "shards": 1
487
  },
488
  {
489
+ "seconds": 0.9554288359940983,
490
+ "bytes": 7500029,
491
+ "pages": 31,
492
+ "shards": 3
493
  },
494
  {
495
+ "seconds": 0.8447882219988969,
496
+ "bytes": 1059224,
497
+ "pages": 35,
498
  "shards": 1
499
  },
500
  {
501
+ "seconds": 1.164817237004172,
502
+ "bytes": 10956336,
503
+ "pages": 35,
504
  "shards": 4
505
  },
506
  {
507
+ "seconds": 1.2077032870001858,
508
+ "bytes": 11770097,
 
 
 
 
 
 
 
 
 
 
 
 
509
  "pages": 34,
510
+ "shards": 3
511
  },
512
  {
513
+ "seconds": 0.8165655029952177,
514
+ "bytes": 1121083,
515
+ "pages": 36,
516
  "shards": 1
517
  },
518
  {
519
+ "seconds": 0.944058788001712,
520
+ "bytes": 7283383,
521
+ "pages": 27,
 
 
 
 
 
 
522
  "shards": 3
523
  },
524
  {
525
+ "seconds": 1.194725619003293,
526
+ "bytes": 5214434,
527
  "pages": 36,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.2110685700026806,
532
+ "bytes": 6209123,
533
+ "pages": 38,
534
+ "shards": 2
535
+ },
536
+ {
537
+ "seconds": 1.3596407239965629,
538
+ "bytes": 7496236,
539
+ "pages": 30,
540
  "shards": 3
541
  },
542
  {
543
+ "seconds": 1.0467049810031313,
544
+ "bytes": 6666834,
545
+ "pages": 30,
546
+ "shards": 2
547
  },
548
  {
549
+ "seconds": 1.164912510997965,
550
+ "bytes": 12226977,
551
+ "pages": 41,
552
+ "shards": 4
553
  },
554
  {
555
+ "seconds": 1.2401886210063822,
556
+ "bytes": 6134452,
557
+ "pages": 38,
558
  "shards": 2
559
+ },
560
+ {
561
+ "seconds": 1.449164785000903,
562
+ "bytes": 969493,
563
+ "pages": 29,
564
+ "shards": 1
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e6a73d2db07c024ebb237c7d0b6fcb773796b3b6ab1195a52d429d55bc86e73a
3
- size 721766911
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04c2d546da542c12c58a0435f1ba889816d6f1028daf377c51a35cccae028102
3
+ size 721764095
training_config.json CHANGED
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 64,
50
- "prefetch_low_water": 32,
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,