appvoid commited on
Commit
cc0e517
·
verified ·
1 Parent(s): 21c28c9

Qwen90M step 270000; 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 +237 -237
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 265000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "ee0020789a0e214a0997c223fbd46efde8cdecccc05fb984287e0e416e538818",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "351fececeaaf7d61cfa47cb55a4c1fc10dd091aef0f1465531852527c3727ddd",
12
- "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "ac15a69eb79c2ac9ef8406edc12d9ae8abafb561b0f74b54b23d5920ee4fe3f7",
14
- "shard_scheduler.json": "2f2de5af2999d7b5fe09209f6bbf509804ef8b94f6516c38e9e688d203d65511",
15
- "coverage.json": "c14d992a21f757f9e0584bc908a2da7fa8b6a7ebe5a0d5d6d02f098aff871438",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 270000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "ff80e0adb68dde4f1a64a84ef34645bbab4563c2ba6a6e22e8ecc9e7f0d64151",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "daa4557c6d9877c2d582a143485f465f77ac3654b5943b52633de0782589f2a9",
12
+ "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "ac15a69eb79c2ac9ef8406edc12d9ae8abafb561b0f74b54b23d5920ee4fe3f7",
14
+ "shard_scheduler.json": "af7ddb35dfaf68caf6ea156cec645638aef2aea57e28dc84fbde94349cb2bc7f",
15
+ "coverage.json": "62a499fb026b3c3a39bc48e383984377f6842cc6452ea627930a87e4ce594788",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 265000,
3
  "stats": {
4
- "climbmix/blocks": 9540000,
5
- "climbmix/targets": 15947557328,
6
- "climbmix/input_tokens": 15948568353,
7
- "ultra_qa/blocks": 3577500,
8
- "ultra_qa/targets": 5836258589,
9
- "ultra_qa/input_tokens": 5836269832,
10
- "cortex/blocks": 3577500,
11
- "cortex/targets": 136096913,
12
- "cortex/input_tokens": 5686718997,
13
- "ultra_style/blocks": 3577500,
14
- "ultra_style/targets": 5659331784,
15
- "ultra_style/input_tokens": 5659628550,
16
- "openmath/blocks": 2385000,
17
- "openmath/targets": 3685319164,
18
- "openmath/input_tokens": 3686547926,
19
- "rewrite6/blocks": 1192500,
20
- "rewrite6/targets": 1954941231,
21
- "rewrite6/input_tokens": 1957084611
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 1306,
26
- "row": 2467,
27
  "start": {
28
- "file": 63,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 18,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 0,
37
- "documents": 5351843
38
  },
39
  "rewrite6": {
40
- "page": 43,
41
- "row": 1780,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 176128
47
  }
48
  },
49
- "epoch": 14,
50
- "offset": 2048,
51
- "documents": 3053564
52
  },
53
  "openmath": {
54
- "page": 816,
55
- "row": 3499,
56
  "start": {
57
  "file": 7,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 285,
61
- "row": 786
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 3345835
67
  },
68
  "ultra_qa": {
69
- "page": 353,
70
- "row": 292,
71
  "start": {
72
  "file": 2,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
- "row": 19137
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 1446180
82
  },
83
  "ultra_style": {
84
- "page": 538,
85
- "row": 3285,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 1,
91
- "row": 63801
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 2206933
97
  },
98
  "cortex": {
99
- "page": 69382,
100
- "row": 24,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 8880920
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 270000,
3
  "stats": {
4
+ "climbmix/blocks": 9720000,
5
+ "climbmix/targets": 16248306295,
6
+ "climbmix/input_tokens": 16249336141,
7
+ "ultra_qa/blocks": 3645000,
8
+ "ultra_qa/targets": 5946191498,
9
+ "ultra_qa/input_tokens": 5946202929,
10
+ "cortex/blocks": 3645000,
11
+ "cortex/targets": 138670584,
12
+ "cortex/input_tokens": 5793992704,
13
+ "ultra_style/blocks": 3645000,
14
+ "ultra_style/targets": 5765190778,
15
+ "ultra_style/input_tokens": 5765493512,
16
+ "openmath/blocks": 2430000,
17
+ "openmath/targets": 3754982954,
18
+ "openmath/input_tokens": 3756234898,
19
+ "rewrite6/blocks": 1215000,
20
+ "rewrite6/targets": 1991834205,
21
+ "rewrite6/input_tokens": 1994017932
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 1331,
26
+ "row": 746,
27
  "start": {
28
+ "file": 64,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 35,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 12288,
37
+ "documents": 5452522
38
  },
39
  "rewrite6": {
40
+ "page": 7,
41
+ "row": 1260,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 28672
47
  }
48
  },
49
+ "epoch": 15,
50
+ "offset": 0,
51
+ "documents": 3110992
52
  },
53
  "openmath": {
54
+ "page": 832,
55
+ "row": 915,
56
  "start": {
57
  "file": 7,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 351,
61
+ "row": 322
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 3408787
67
  },
68
  "ultra_qa": {
69
+ "page": 359,
70
+ "row": 2968,
71
  "start": {
72
  "file": 2,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
+ "row": 43713
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 1473432
82
  },
83
  "ultra_style": {
84
+ "page": 549,
85
+ "row": 1375,
86
  "start": {
87
  "file": 3,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 2,
91
+ "row": 23029
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 2250079
97
  },
98
  "cortex": {
99
+ "page": 70690,
100
+ "row": 82,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 9048402
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 265000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "ee0020789a0e214a0997c223fbd46efde8cdecccc05fb984287e0e416e538818", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "351fececeaaf7d61cfa47cb55a4c1fc10dd091aef0f1465531852527c3727ddd", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "ac15a69eb79c2ac9ef8406edc12d9ae8abafb561b0f74b54b23d5920ee4fe3f7", "shard_scheduler.json": "2f2de5af2999d7b5fe09209f6bbf509804ef8b94f6516c38e9e688d203d65511", "coverage.json": "c14d992a21f757f9e0584bc908a2da7fa8b6a7ebe5a0d5d6d02f098aff871438", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000265000"}
 
1
+ {"step": 270000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "ff80e0adb68dde4f1a64a84ef34645bbab4563c2ba6a6e22e8ecc9e7f0d64151", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "daa4557c6d9877c2d582a143485f465f77ac3654b5943b52633de0782589f2a9", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "ac15a69eb79c2ac9ef8406edc12d9ae8abafb561b0f74b54b23d5920ee4fe3f7", "shard_scheduler.json": "af7ddb35dfaf68caf6ea156cec645638aef2aea57e28dc84fbde94349cb2bc7f", "coverage.json": "62a499fb026b3c3a39bc48e383984377f6842cc6452ea627930a87e4ce594788", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000270000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ee0020789a0e214a0997c223fbd46efde8cdecccc05fb984287e0e416e538818
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ff80e0adb68dde4f1a64a84ef34645bbab4563c2ba6a6e22e8ecc9e7f0d64151
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": 75598.5316342505,
26
- "required_pages_per_second_upper_estimate": 0.0065526891193301235
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": 9449.816454281312,
40
- "required_pages_per_second_upper_estimate": 0.003769673070959515
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": 18899.632908562624,
54
- "required_pages_per_second_upper_estimate": 0.00437209745121723
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": 28349.449362843934,
68
- "required_pages_per_second_upper_estimate": 0.0017263327281314721
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": 28349.449362843934,
82
- "required_pages_per_second_upper_estimate": 0.002182419088250269
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": 28349.449362843934,
96
- "required_pages_per_second_upper_estimate": 0.3520052815829238
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": 0.1434574083060457,
157
  "selected": {
158
- "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 690670,
161
- "producer_batch_seconds": 0.263509006851649,
162
- "producer_p95_seconds": 0.32680167599755805,
163
- "gpu_update_seconds": 0.9752570374871801,
164
- "producer_batches_per_second": 3.794936696653343,
165
- "gpu_batches_per_second": 1.0253707090148994,
166
- "producer_headroom": 3.7010387202295245,
167
  "sustainable": true,
168
- "buffer_seconds": 60.46593632420517,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 82,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 372935.845468488,
174
- "upload_bytes_per_second": 4393328.842840834,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 80.02335093537312,
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": 1.0260517099995923,
184
- "bytes": 2406820,
185
- "pages": 27,
186
- "shards": 2
187
  },
188
  {
189
- "seconds": 1.0375558390005608,
190
- "bytes": 11763695,
191
- "pages": 40,
192
  "shards": 3
193
  },
194
  {
195
- "seconds": 1.2267565980000654,
196
- "bytes": 917900,
197
- "pages": 27,
198
  "shards": 1
199
  },
200
  {
201
- "seconds": 1.2721305819977715,
202
- "bytes": 12594000,
203
- "pages": 39,
204
  "shards": 4
205
  },
206
  {
207
- "seconds": 1.3089486960016075,
208
- "bytes": 6052165,
209
- "pages": 33,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.166924328001187,
214
- "bytes": 2378485,
215
- "pages": 30,
216
  "shards": 2
217
  },
218
  {
219
- "seconds": 1.0251769999995304,
220
- "bytes": 6275188,
221
  "pages": 36,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 0.9176649480032211,
226
- "bytes": 4600807,
227
- "pages": 27,
228
- "shards": 2
229
- },
230
- {
231
- "seconds": 1.5337045840024075,
232
- "bytes": 13355543,
233
- "pages": 42,
234
- "shards": 4
235
- },
236
- {
237
- "seconds": 1.0210915280003974,
238
- "bytes": 6074322,
239
  "pages": 30,
240
  "shards": 2
241
  },
242
  {
243
- "seconds": 0.9218759530012903,
244
- "bytes": 1088353,
245
- "pages": 34,
246
- "shards": 1
247
- },
248
- {
249
- "seconds": 1.2536445320001803,
250
- "bytes": 11359516,
251
- "pages": 35,
252
- "shards": 4
253
  },
254
  {
255
- "seconds": 1.0509260079998057,
256
- "bytes": 6234779,
257
  "pages": 36,
258
  "shards": 2
259
  },
260
  {
261
- "seconds": 1.154916157000116,
262
- "bytes": 2367312,
263
- "pages": 28,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 1.8538454429981357,
268
- "bytes": 11918988,
269
- "pages": 37,
270
- "shards": 3
271
  },
272
  {
273
- "seconds": 1.6541002780031704,
274
- "bytes": 6139021,
275
- "pages": 33,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.9353761299971666,
280
- "bytes": 7270095,
281
- "pages": 29,
282
  "shards": 3
283
  },
284
  {
285
- "seconds": 2.698638427998958,
286
- "bytes": 6167283,
287
  "pages": 36,
288
  "shards": 2
289
  },
290
  {
291
- "seconds": 2.6425060760011547,
292
- "bytes": 2623268,
293
- "pages": 37,
294
  "shards": 2
295
  },
296
  {
297
- "seconds": 2.5355882220028434,
298
- "bytes": 9765836,
299
- "pages": 33,
300
  "shards": 3
301
  },
302
  {
303
- "seconds": 1.361022408997087,
304
- "bytes": 6242966,
305
- "pages": 18,
306
  "shards": 3
307
  },
308
  {
309
- "seconds": 1.2147977610002272,
310
- "bytes": 6193785,
311
- "pages": 35,
312
  "shards": 2
313
  },
314
  {
315
- "seconds": 0.9691767799995432,
316
- "bytes": 2430413,
317
- "pages": 30,
318
  "shards": 2
319
  },
320
  {
321
- "seconds": 1.0292037019971758,
322
- "bytes": 5575102,
323
- "pages": 18,
324
  "shards": 2
325
  },
326
  {
327
- "seconds": 1.288112906000606,
328
- "bytes": 5599143,
329
- "pages": 33,
 
 
 
 
 
 
330
  "shards": 2
331
  },
332
  {
333
- "seconds": 1.0466477340014535,
334
- "bytes": 7416772,
335
- "pages": 30,
336
- "shards": 3
337
  },
338
  {
339
- "seconds": 1.263268196998979,
340
- "bytes": 1135440,
341
- "pages": 34,
342
- "shards": 1
343
  },
344
  {
345
- "seconds": 1.37565966299735,
346
- "bytes": 13270818,
347
  "pages": 36,
348
  "shards": 4
349
  },
350
  {
351
- "seconds": 0.7815917669977352,
352
- "bytes": 913950,
353
- "pages": 28,
354
  "shards": 1
355
  },
356
  {
357
- "seconds": 1.2434416440009954,
358
- "bytes": 10283141,
359
- "pages": 41,
360
  "shards": 3
361
  },
362
  {
363
- "seconds": 1.034428533999744,
364
- "bytes": 7372684,
365
- "pages": 29,
366
- "shards": 3
367
  },
368
  {
369
- "seconds": 1.070391503002611,
370
- "bytes": 1123754,
371
  "pages": 35,
372
- "shards": 1
373
- },
374
- {
375
- "seconds": 1.18286236700078,
376
- "bytes": 5852988,
377
- "pages": 29,
378
  "shards": 2
379
  },
380
  {
381
- "seconds": 1.7732943640003214,
382
- "bytes": 17450620,
383
- "pages": 33,
384
- "shards": 5
385
  },
386
  {
387
- "seconds": 0.8563921040004061,
388
- "bytes": 1024816,
389
  "pages": 34,
390
  "shards": 1
391
  },
392
  {
393
- "seconds": 1.4308035579997522,
394
- "bytes": 7307496,
395
- "pages": 39,
396
- "shards": 3
397
  },
398
  {
399
- "seconds": 1.6164238569981535,
400
- "bytes": 892499,
401
- "pages": 26,
402
- "shards": 1
403
  },
404
  {
405
- "seconds": 1.2368712060015241,
406
- "bytes": 10018327,
407
- "pages": 37,
408
- "shards": 3
409
  },
410
  {
411
- "seconds": 1.856085086998064,
412
- "bytes": 7579977,
413
  "pages": 38,
414
  "shards": 3
415
  },
416
  {
417
- "seconds": 1.2339968710002722,
418
- "bytes": 6738522,
419
- "pages": 29,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 2.4403519309998956,
424
- "bytes": 5986840,
425
- "pages": 30,
426
- "shards": 2
427
  },
428
  {
429
- "seconds": 2.588106572999095,
430
- "bytes": 12282184,
431
  "pages": 39,
432
- "shards": 4
433
  },
434
  {
435
- "seconds": 2.1326607650007645,
436
- "bytes": 1150711,
437
- "pages": 35,
438
  "shards": 1
439
  },
440
  {
441
- "seconds": 2.922968579998269,
442
- "bytes": 7646925,
443
- "pages": 36,
 
 
 
 
 
 
444
  "shards": 3
445
  },
446
  {
447
- "seconds": 2.558325848000095,
448
- "bytes": 890978,
449
- "pages": 28,
450
- "shards": 1
451
  },
452
  {
453
- "seconds": 2.274885417002224,
454
- "bytes": 15797221,
455
- "pages": 34,
456
  "shards": 4
457
  },
458
  {
459
- "seconds": 3.0481821479988866,
460
- "bytes": 2407311,
461
- "pages": 29,
 
 
 
 
 
 
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.2172690299994429,
466
- "bytes": 6177200,
467
  "pages": 36,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.8479619009995076,
472
- "bytes": 5907251,
473
- "pages": 30,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.345177949002391,
478
- "bytes": 6941210,
479
  "pages": 37,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 1.2569710710013169,
484
- "bytes": 6280484,
485
- "pages": 36,
486
- "shards": 2
487
  },
488
  {
489
- "seconds": 1.521943290001218,
490
- "bytes": 13325077,
491
  "pages": 36,
492
- "shards": 4
493
  },
494
  {
495
- "seconds": 1.0711962380009936,
496
- "bytes": 889212,
497
- "pages": 26,
498
  "shards": 2
499
  },
500
  {
501
- "seconds": 1.3221969680016628,
502
- "bytes": 10781457,
503
- "pages": 38,
504
  "shards": 3
505
  },
506
  {
507
- "seconds": 1.231661190999148,
508
- "bytes": 2563678,
509
  "pages": 36,
510
  "shards": 2
511
  },
512
  {
513
- "seconds": 1.1248420010015252,
514
- "bytes": 5951433,
515
- "pages": 29,
516
  "shards": 2
517
  },
518
  {
519
- "seconds": 1.6543720410008973,
520
- "bytes": 7466727,
521
- "pages": 38,
522
  "shards": 3
523
  },
524
  {
525
- "seconds": 1.3103912280021177,
526
- "bytes": 10646431,
527
- "pages": 28,
 
 
 
 
 
 
528
  "shards": 3
529
  },
530
  {
531
- "seconds": 1.3748210780031513,
532
- "bytes": 6494530,
533
- "pages": 37,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 0.9434360529994592,
538
- "bytes": 2587569,
539
- "pages": 36,
540
  "shards": 2
541
  },
542
  {
543
- "seconds": 1.083952794000652,
544
- "bytes": 5627431,
545
  "pages": 28,
 
 
 
 
 
 
546
  "shards": 2
547
  },
548
  {
549
- "seconds": 1.2045630370012077,
550
- "bytes": 11050050,
551
- "pages": 39,
552
  "shards": 3
553
  },
554
  {
555
- "seconds": 1.3764796949981246,
556
- "bytes": 2478260,
557
- "pages": 28,
558
  "shards": 2
559
  },
560
  {
561
- "seconds": 1.1128817630014964,
562
- "bytes": 6162674,
563
- "pages": 35,
564
- "shards": 2
 
 
 
 
 
 
565
  }
566
  ],
567
  "cache_key": {
 
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": 74602.20012608299,
26
+ "required_pages_per_second_upper_estimate": 0.006466329629381285
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": 9325.275015760373,
40
+ "required_pages_per_second_upper_estimate": 0.0037199916290730707
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": 18650.550031520746,
54
+ "required_pages_per_second_upper_estimate": 0.004314476511322505
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": 27975.825047281116,
68
+ "required_pages_per_second_upper_estimate": 0.0017035809675689838
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": 27975.825047281116,
82
+ "required_pages_per_second_upper_estimate": 0.0021536564541799422
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": 27975.825047281116,
96
+ "required_pages_per_second_upper_estimate": 0.3473661180237793
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": 0.2140537983123636,
162
+ "producer_p95_seconds": 0.3658770080000977,
163
+ "gpu_update_seconds": 0.9882818452457767,
164
+ "producer_batches_per_second": 4.671722753271231,
165
+ "gpu_batches_per_second": 1.0118570980642765,
166
+ "producer_headroom": 4.616978783079572,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.285192559992375,
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": 268198.67309396865,
174
+ "upload_bytes_per_second": 4767939.973542438,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 57.54919190038962,
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": 2.1326607650007645,
184
+ "bytes": 1150711,
185
+ "pages": 35,
186
+ "shards": 1
187
  },
188
  {
189
+ "seconds": 2.922968579998269,
190
+ "bytes": 7646925,
191
+ "pages": 36,
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 2.558325848000095,
196
+ "bytes": 890978,
197
+ "pages": 28,
198
  "shards": 1
199
  },
200
  {
201
+ "seconds": 2.274885417002224,
202
+ "bytes": 15797221,
203
+ "pages": 34,
204
  "shards": 4
205
  },
206
  {
207
+ "seconds": 3.0481821479988866,
208
+ "bytes": 2407311,
209
+ "pages": 29,
 
 
 
 
 
 
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.2172690299994429,
214
+ "bytes": 6177200,
215
  "pages": 36,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.8479619009995076,
220
+ "bytes": 5907251,
 
 
 
 
 
 
 
 
 
 
 
 
221
  "pages": 30,
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 1.345177949002391,
226
+ "bytes": 6941210,
227
+ "pages": 37,
228
+ "shards": 3
 
 
 
 
 
 
229
  },
230
  {
231
+ "seconds": 1.2569710710013169,
232
+ "bytes": 6280484,
233
  "pages": 36,
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.521943290001218,
238
+ "bytes": 13325077,
239
+ "pages": 36,
240
+ "shards": 4
 
 
 
 
 
 
241
  },
242
  {
243
+ "seconds": 1.0711962380009936,
244
+ "bytes": 889212,
245
+ "pages": 26,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.3221969680016628,
250
+ "bytes": 10781457,
251
+ "pages": 38,
252
  "shards": 3
253
  },
254
  {
255
+ "seconds": 1.231661190999148,
256
+ "bytes": 2563678,
257
  "pages": 36,
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 1.1248420010015252,
262
+ "bytes": 5951433,
263
+ "pages": 29,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.6543720410008973,
268
+ "bytes": 7466727,
269
+ "pages": 38,
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 1.3103912280021177,
274
+ "bytes": 10646431,
275
+ "pages": 28,
276
  "shards": 3
277
  },
278
  {
279
+ "seconds": 1.3748210780031513,
280
+ "bytes": 6494530,
281
+ "pages": 37,
282
  "shards": 2
283
  },
284
  {
285
+ "seconds": 0.9434360529994592,
286
+ "bytes": 2587569,
287
+ "pages": 36,
288
  "shards": 2
289
  },
290
  {
291
+ "seconds": 1.083952794000652,
292
+ "bytes": 5627431,
293
+ "pages": 28,
294
  "shards": 2
295
  },
296
  {
297
+ "seconds": 1.2045630370012077,
298
+ "bytes": 11050050,
299
+ "pages": 39,
300
+ "shards": 3
301
+ },
302
+ {
303
+ "seconds": 1.3764796949981246,
304
+ "bytes": 2478260,
305
+ "pages": 28,
306
  "shards": 2
307
  },
308
  {
309
+ "seconds": 1.1128817630014964,
310
+ "bytes": 6162674,
311
+ "pages": 35,
312
+ "shards": 2
313
  },
314
  {
315
+ "seconds": 1.2386194030004845,
316
+ "bytes": 6597417,
317
+ "pages": 26,
318
+ "shards": 2
319
  },
320
  {
321
+ "seconds": 1.1337901490005606,
322
+ "bytes": 11234450,
323
  "pages": 36,
324
  "shards": 4
325
  },
326
  {
327
+ "seconds": 0.9795328820000577,
328
+ "bytes": 930296,
329
+ "pages": 29,
330
  "shards": 1
331
  },
332
  {
333
+ "seconds": 1.2613436630017532,
334
+ "bytes": 7588531,
335
+ "pages": 35,
336
  "shards": 3
337
  },
338
  {
339
+ "seconds": 0.8843855299965071,
340
+ "bytes": 5949979,
341
+ "pages": 34,
342
+ "shards": 2
343
  },
344
  {
345
+ "seconds": 1.1865619210002478,
346
+ "bytes": 5319050,
347
  "pages": 35,
 
 
 
 
 
 
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 1.3752340050014027,
352
+ "bytes": 13115816,
353
+ "pages": 30,
354
+ "shards": 4
355
  },
356
  {
357
+ "seconds": 1.269822567999654,
358
+ "bytes": 1251620,
359
  "pages": 34,
360
  "shards": 1
361
  },
362
  {
363
+ "seconds": 0.9414359830007015,
364
+ "bytes": 5894779,
365
+ "pages": 30,
366
+ "shards": 2
367
  },
368
  {
369
+ "seconds": 1.1529198649986938,
370
+ "bytes": 11606253,
371
+ "pages": 40,
372
+ "shards": 4
373
  },
374
  {
375
+ "seconds": 0.9286228500022844,
376
+ "bytes": 849903,
377
+ "pages": 26,
378
+ "shards": 1
379
  },
380
  {
381
+ "seconds": 1.0139027430013812,
382
+ "bytes": 7512493,
383
  "pages": 38,
384
  "shards": 3
385
  },
386
  {
387
+ "seconds": 0.9002696629977436,
388
+ "bytes": 6383764,
389
+ "pages": 19,
390
  "shards": 2
391
  },
392
  {
393
+ "seconds": 1.1763414359993476,
394
+ "bytes": 9929012,
395
+ "pages": 38,
396
+ "shards": 3
397
  },
398
  {
399
+ "seconds": 1.3798298289984814,
400
+ "bytes": 7738436,
401
  "pages": 39,
402
+ "shards": 3
403
  },
404
  {
405
+ "seconds": 0.956485711998539,
406
+ "bytes": 1216085,
407
+ "pages": 34,
408
  "shards": 1
409
  },
410
  {
411
+ "seconds": 0.9746390660002362,
412
+ "bytes": 5704060,
413
+ "pages": 20,
414
+ "shards": 2
415
+ },
416
+ {
417
+ "seconds": 1.1505642880001687,
418
+ "bytes": 6924827,
419
+ "pages": 43,
420
  "shards": 3
421
  },
422
  {
423
+ "seconds": 0.9783257499984757,
424
+ "bytes": 5971368,
425
+ "pages": 30,
426
+ "shards": 2
427
  },
428
  {
429
+ "seconds": 1.5463005599995086,
430
+ "bytes": 13315083,
431
+ "pages": 40,
432
  "shards": 4
433
  },
434
  {
435
+ "seconds": 1.3599293949991988,
436
+ "bytes": 4987866,
437
+ "pages": 33,
438
+ "shards": 3
439
+ },
440
+ {
441
+ "seconds": 1.145182983000268,
442
+ "bytes": 6203180,
443
+ "pages": 32,
444
  "shards": 2
445
  },
446
  {
447
+ "seconds": 0.9317472719994839,
448
+ "bytes": 2642711,
449
  "pages": 36,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 0.9964947309999843,
454
+ "bytes": 5697842,
455
+ "pages": 28,
456
  "shards": 2
457
  },
458
  {
459
+ "seconds": 1.1368190850007522,
460
+ "bytes": 10318857,
461
  "pages": 37,
462
  "shards": 3
463
  },
464
  {
465
+ "seconds": 1.0222451929985255,
466
+ "bytes": 8375215,
467
+ "pages": 35,
468
+ "shards": 3
469
  },
470
  {
471
+ "seconds": 0.9998172479972709,
472
+ "bytes": 6024653,
473
  "pages": 36,
474
+ "shards": 2
475
  },
476
  {
477
+ "seconds": 0.9568293169977551,
478
+ "bytes": 2417450,
479
+ "pages": 29,
480
  "shards": 2
481
  },
482
  {
483
+ "seconds": 1.2115873000002466,
484
+ "bytes": 10372111,
485
+ "pages": 33,
486
  "shards": 3
487
  },
488
  {
489
+ "seconds": 1.1448558080010116,
490
+ "bytes": 6260573,
491
  "pages": 36,
492
  "shards": 2
493
  },
494
  {
495
+ "seconds": 1.1358172180007386,
496
+ "bytes": 2620953,
497
+ "pages": 36,
498
  "shards": 2
499
  },
500
  {
501
+ "seconds": 1.4711359560023993,
502
+ "bytes": 11859463,
503
+ "pages": 29,
504
  "shards": 3
505
  },
506
  {
507
+ "seconds": 1.2465269989988883,
508
+ "bytes": 5535585,
509
+ "pages": 33,
510
+ "shards": 2
511
+ },
512
+ {
513
+ "seconds": 1.2969893279987446,
514
+ "bytes": 7578437,
515
+ "pages": 39,
516
  "shards": 3
517
  },
518
  {
519
+ "seconds": 1.5002685959989321,
520
+ "bytes": 6193769,
521
+ "pages": 27,
522
  "shards": 2
523
  },
524
  {
525
+ "seconds": 1.1753300150012365,
526
+ "bytes": 2646901,
527
+ "pages": 37,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.4061265680029464,
532
+ "bytes": 9691299,
533
  "pages": 28,
534
+ "shards": 3
535
+ },
536
+ {
537
+ "seconds": 1.0368577580011333,
538
+ "bytes": 6834427,
539
+ "pages": 31,
540
  "shards": 2
541
  },
542
  {
543
+ "seconds": 1.4560243180021644,
544
+ "bytes": 7277270,
545
+ "pages": 30,
546
  "shards": 3
547
  },
548
  {
549
+ "seconds": 1.4158115079990239,
550
+ "bytes": 6180346,
551
+ "pages": 38,
552
  "shards": 2
553
  },
554
  {
555
+ "seconds": 1.5717087800003355,
556
+ "bytes": 560796,
557
+ "pages": 18,
558
+ "shards": 1
559
+ },
560
+ {
561
+ "seconds": 4.375384323997423,
562
+ "bytes": 11988381,
563
+ "pages": 43,
564
+ "shards": 4
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:351fececeaaf7d61cfa47cb55a4c1fc10dd091aef0f1465531852527c3727ddd
3
- size 721773055
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daa4557c6d9877c2d582a143485f465f77ac3654b5943b52633de0782589f2a9
3
+ size 721768959
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,