appvoid commited on
Commit
1dd00e8
·
verified ·
1 Parent(s): 850836d

Qwen90M step 170000; 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 +260 -218
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 165000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "2055e13f7b5b304d0c2206d7fb9fd41c1dbf127325878ef730f5b741d6555462",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "4aadd2411e27d45ea0852b13570a9cc811de9a400c6b05705bce64c357639250",
12
- "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08",
14
- "shard_scheduler.json": "37c2d3b415e95706fcd09442dbfed95fbe0847366375512b565a88834bde0eeb",
15
- "coverage.json": "6b16c159bcce254a2303d601ede22d8e32b19a35679351555cb18cba36d7f12d",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 170000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "eae93b5a987696907c76cdf5c39f1ece9ef2dc6c0dacd54567e9b61a70909330",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "68b2bd34b599c12b63f5eb6b5156f7905dc9454684b6f321753aaa307c958f77",
12
+ "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08",
14
+ "shard_scheduler.json": "6b80239d2fad6471bd75c8996f07573047714dceff819ccf239934bc4693936c",
15
+ "coverage.json": "33a1ecc073a8506314c0c7e8b1455eaacfff9bce622ac5436c9d5ec2fcbfbf5d",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 165000,
3
  "stats": {
4
- "climbmix/blocks": 5940000,
5
- "climbmix/targets": 9927628187,
6
- "climbmix/input_tokens": 9928257678,
7
- "ultra_qa/blocks": 2227500,
8
- "ultra_qa/targets": 3633811973,
9
- "ultra_qa/input_tokens": 3633818963,
10
- "cortex/blocks": 2227500,
11
- "cortex/targets": 84725586,
12
- "cortex/input_tokens": 3540651120,
13
- "ultra_style/blocks": 2227500,
14
- "ultra_style/targets": 3523134670,
15
- "ultra_style/input_tokens": 3523319504,
16
- "openmath/blocks": 1485000,
17
- "openmath/targets": 2294270266,
18
- "openmath/input_tokens": 2295034673,
19
- "rewrite6/blocks": 742500,
20
- "rewrite6/targets": 1217215870,
21
- "rewrite6/input_tokens": 1218550219
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 814,
26
- "row": 242,
27
  "start": {
28
- "file": 39,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 27,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 3334386
38
  },
39
  "rewrite6": {
40
- "page": 12,
41
- "row": 3197,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 49152
47
  }
48
  },
49
  "epoch": 9,
50
- "offset": 0,
51
- "documents": 1900985
52
  },
53
  "openmath": {
54
- "page": 508,
55
- "row": 1788,
56
  "start": {
57
  "file": 4,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 334,
61
- "row": 168
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 0,
66
- "documents": 2082556
67
  },
68
  "ultra_qa": {
69
- "page": 219,
70
- "row": 3510,
71
  "start": {
72
  "file": 1,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 5,
76
- "row": 54490
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 900534
82
  },
83
  "ultra_style": {
84
- "page": 335,
85
- "row": 2292,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 0,
91
- "row": 2278
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 1374452
97
  },
98
  "cortex": {
99
- "page": 43197,
100
- "row": 83,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 5529299
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 170000,
3
  "stats": {
4
+ "climbmix/blocks": 6120000,
5
+ "climbmix/targets": 10228213717,
6
+ "climbmix/input_tokens": 10228862440,
7
+ "ultra_qa/blocks": 2295000,
8
+ "ultra_qa/targets": 3743992865,
9
+ "ultra_qa/input_tokens": 3744000053,
10
+ "cortex/blocks": 2295000,
11
+ "cortex/targets": 87303716,
12
+ "cortex/input_tokens": 3647892000,
13
+ "ultra_style/blocks": 2295000,
14
+ "ultra_style/targets": 3630136904,
15
+ "ultra_style/input_tokens": 3630327119,
16
+ "openmath/blocks": 1530000,
17
+ "openmath/targets": 2363841526,
18
+ "openmath/input_tokens": 2364629111,
19
+ "rewrite6/blocks": 765000,
20
+ "rewrite6/targets": 1254067034,
21
+ "rewrite6/input_tokens": 1255442274
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 838,
26
+ "row": 3535,
27
  "start": {
28
+ "file": 40,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 40,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 3435983
38
  },
39
  "rewrite6": {
40
+ "page": 26,
41
+ "row": 4050,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 106496
47
  }
48
  },
49
  "epoch": 9,
50
+ "offset": 2048,
51
+ "documents": 1959182
52
  },
53
  "openmath": {
54
+ "page": 523,
55
+ "row": 3399,
56
  "start": {
57
  "file": 4,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 395,
61
+ "row": 608
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 2048,
66
+ "documents": 2145607
67
  },
68
  "ultra_qa": {
69
+ "page": 226,
70
+ "row": 2070,
71
  "start": {
72
  "file": 1,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 6,
76
+ "row": 18609
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 927766
82
  },
83
  "ultra_style": {
84
+ "page": 345,
85
+ "row": 2329,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 0,
91
+ "row": 43238
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 1415449
97
  },
98
  "cortex": {
99
+ "page": 44505,
100
+ "row": 35,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 5696675
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 165000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "2055e13f7b5b304d0c2206d7fb9fd41c1dbf127325878ef730f5b741d6555462", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "4aadd2411e27d45ea0852b13570a9cc811de9a400c6b05705bce64c357639250", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08", "shard_scheduler.json": "37c2d3b415e95706fcd09442dbfed95fbe0847366375512b565a88834bde0eeb", "coverage.json": "6b16c159bcce254a2303d601ede22d8e32b19a35679351555cb18cba36d7f12d", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000165000"}
 
1
+ {"step": 170000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "eae93b5a987696907c76cdf5c39f1ece9ef2dc6c0dacd54567e9b61a70909330", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "68b2bd34b599c12b63f5eb6b5156f7905dc9454684b6f321753aaa307c958f77", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08", "shard_scheduler.json": "6b80239d2fad6471bd75c8996f07573047714dceff819ccf239934bc4693936c", "coverage.json": "33a1ecc073a8506314c0c7e8b1455eaacfff9bce622ac5436c9d5ec2fcbfbf5d", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000170000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2055e13f7b5b304d0c2206d7fb9fd41c1dbf127325878ef730f5b741d6555462
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eae93b5a987696907c76cdf5c39f1ece9ef2dc6c0dacd54567e9b61a70909330
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": 76024.17600051314,
26
- "required_pages_per_second_upper_estimate": 0.006589582894211986
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": 9503.022000064142,
40
- "required_pages_per_second_upper_estimate": 0.0037908975586660853
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": 19006.044000128284,
54
- "required_pages_per_second_upper_estimate": 0.004396713784479697
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": 28509.06600019243,
68
- "required_pages_per_second_upper_estimate": 0.0017360525439022204
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": 28509.06600019243,
82
- "required_pages_per_second_upper_estimate": 0.002194706818840488
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": 28509.06600019243,
96
- "required_pages_per_second_upper_estimate": 0.35398718601627116
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,370 +155,412 @@
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
157
  "selected": {
158
- "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
- "producer_batch_seconds": 0.09048123760162241,
162
- "producer_p95_seconds": 0.1997182420000172,
163
- "gpu_update_seconds": 0.9697967656959854,
164
- "producer_batches_per_second": 11.05201505314146,
165
- "gpu_batches_per_second": 1.0311438802152932,
166
- "producer_headroom": 10.718208452959932,
167
  "sustainable": true,
168
- "buffer_seconds": 60.127399473151094,
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": 314248.84728296526,
174
- "upload_bytes_per_second": 5212065.682698314,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 67.43048728815764,
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": 1.0642098359999181,
184
- "bytes": 45649,
185
- "pages": 1,
186
  "shards": 1
187
  },
188
  {
189
- "seconds": 1.2323853809998582,
190
- "bytes": 7278265,
 
 
 
 
 
 
 
 
 
 
 
 
191
  "pages": 34,
 
 
 
 
 
 
192
  "shards": 3
193
  },
194
  {
195
- "seconds": 0.9598245239999414,
196
- "bytes": 1062466,
197
  "pages": 33,
198
  "shards": 1
199
  },
200
  {
201
- "seconds": 1.1868087180000657,
202
- "bytes": 11845509,
203
- "pages": 35,
204
- "shards": 4
205
  },
206
  {
207
- "seconds": 0.9347497089997887,
208
- "bytes": 6852819,
209
- "pages": 34,
210
- "shards": 2
211
  },
212
  {
213
- "seconds": 1.0938459449998845,
214
- "bytes": 6142145,
215
- "pages": 35,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 1.2391148739998243,
220
- "bytes": 7141661,
221
- "pages": 29,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 3.5350073540002995,
226
- "bytes": 5274454,
227
- "pages": 34,
228
- "shards": 2
229
  },
230
  {
231
- "seconds": 1.0228857370002515,
232
- "bytes": 6079721,
233
  "pages": 34,
234
- "shards": 2
235
  },
236
  {
237
- "seconds": 1.587895904999641,
238
- "bytes": 7946846,
239
- "pages": 38,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 1.0177612440002122,
244
- "bytes": 6784132,
245
- "pages": 31,
246
  "shards": 2
247
  },
248
  {
249
- "seconds": 1.3712029609996534,
250
- "bytes": 12398330,
251
- "pages": 38,
252
- "shards": 4
253
  },
254
  {
255
- "seconds": 0.7755766290001702,
256
- "bytes": 1144894,
257
- "pages": 33,
258
- "shards": 1
259
  },
260
  {
261
- "seconds": 1.4222684020005545,
262
- "bytes": 6265294,
263
- "pages": 27,
264
  "shards": 2
265
  },
266
  {
267
- "seconds": 0.9120372190000126,
268
- "bytes": 2656908,
269
- "pages": 35,
270
- "shards": 2
271
  },
272
  {
273
- "seconds": 1.6420339440001044,
274
- "bytes": 5755818,
275
- "pages": 29,
276
- "shards": 2
277
  },
278
  {
279
- "seconds": 1.4588776000000507,
280
- "bytes": 16176314,
281
- "pages": 44,
282
- "shards": 4
283
  },
284
  {
285
- "seconds": 0.8323371429996769,
286
- "bytes": 2088001,
287
- "pages": 18,
288
  "shards": 2
289
  },
290
  {
291
- "seconds": 1.2336509919996388,
292
- "bytes": 6127983,
293
- "pages": 36,
294
  "shards": 2
295
  },
296
  {
297
- "seconds": 1.0918112409999594,
298
- "bytes": 968274,
299
- "pages": 28,
300
- "shards": 1
301
  },
302
  {
303
- "seconds": 1.2902597590000369,
304
- "bytes": 12650540,
305
- "pages": 36,
306
- "shards": 4
307
  },
308
  {
309
- "seconds": 1.0202606470002138,
310
- "bytes": 1124000,
311
- "pages": 34,
312
  "shards": 2
313
  },
314
  {
315
- "seconds": 1.3055697850004435,
316
- "bytes": 13351938,
317
- "pages": 38,
318
- "shards": 4
319
  },
320
  {
321
- "seconds": 1.0015427250000357,
322
- "bytes": 6023702,
323
- "pages": 35,
324
- "shards": 2
325
  },
326
  {
327
- "seconds": 1.052357557000505,
328
- "bytes": 4323709,
329
- "pages": 27,
330
- "shards": 2
331
  },
332
  {
333
- "seconds": 1.0870202730002347,
334
- "bytes": 7523830,
335
- "pages": 37,
336
  "shards": 3
337
  },
338
  {
339
- "seconds": 0.8949287559998993,
340
- "bytes": 6264749,
341
  "pages": 35,
342
  "shards": 2
343
  },
344
  {
345
- "seconds": 1.0034355950001554,
346
- "bytes": 860910,
347
- "pages": 25,
348
  "shards": 1
349
  },
350
  {
351
- "seconds": 1.6520775649996722,
352
- "bytes": 17392082,
 
 
 
 
 
 
353
  "pages": 43,
354
- "shards": 5
355
  },
356
  {
357
- "seconds": 1.2575243370001772,
358
- "bytes": 1055853,
359
- "pages": 33,
360
  "shards": 1
361
  },
362
  {
363
- "seconds": 2.737182930999552,
364
- "bytes": 7527517,
365
- "pages": 37,
366
  "shards": 3
367
  },
368
  {
369
- "seconds": 0.9472420219999549,
370
- "bytes": 626425,
371
- "pages": 19,
372
- "shards": 1
373
- },
374
- {
375
- "seconds": 1.3963090409997676,
376
- "bytes": 10793699,
377
- "pages": 45,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 1.3661847930006843,
382
- "bytes": 7242166,
383
  "pages": 29,
 
 
 
 
 
 
384
  "shards": 3
385
  },
386
  {
387
- "seconds": 1.0256425439993109,
388
- "bytes": 848800,
389
- "pages": 26,
390
  "shards": 1
391
  },
392
  {
393
- "seconds": 1.479707024000163,
394
- "bytes": 13318632,
395
- "pages": 43,
396
  "shards": 4
397
  },
398
  {
399
- "seconds": 1.157918292999966,
400
- "bytes": 10292312,
401
- "pages": 36,
402
- "shards": 3
403
  },
404
  {
405
- "seconds": 2.16527700000006,
406
- "bytes": 1181305,
407
- "pages": 34,
408
- "shards": 1
409
  },
410
  {
411
- "seconds": 1.226145381000606,
412
- "bytes": 7407097,
413
- "pages": 30,
414
- "shards": 3
415
  },
416
  {
417
- "seconds": 0.7928119780008274,
418
- "bytes": 1141631,
419
- "pages": 33,
420
- "shards": 1
421
  },
422
  {
423
- "seconds": 1.3008312719994137,
424
- "bytes": 10073702,
425
- "pages": 31,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 1.3391879739992874,
430
- "bytes": 13717263,
431
- "pages": 44,
432
- "shards": 4
433
  },
434
  {
435
- "seconds": 0.882563871999082,
436
- "bytes": 1071171,
437
  "pages": 33,
438
  "shards": 1
439
  },
440
  {
441
- "seconds": 1.5854626939999434,
442
- "bytes": 7158059,
443
- "pages": 27,
444
- "shards": 3
445
  },
446
  {
447
- "seconds": 1.1073119590000715,
448
- "bytes": 10577372,
449
- "pages": 37,
450
- "shards": 3
451
  },
452
  {
453
- "seconds": 0.7750392130001273,
454
- "bytes": 1199165,
455
- "pages": 34,
456
- "shards": 1
457
  },
458
  {
459
- "seconds": 1.0709359819993551,
460
- "bytes": 7553738,
461
- "pages": 32,
462
  "shards": 3
463
  },
464
  {
465
- "seconds": 0.9643511739996029,
466
- "bytes": 6918272,
467
- "pages": 35,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.0286668139997346,
472
- "bytes": 9530498,
473
- "pages": 29,
474
  "shards": 3
475
  },
476
  {
477
- "seconds": 0.9772154700003739,
478
- "bytes": 2553165,
479
- "pages": 32,
480
  "shards": 2
481
  },
482
  {
483
- "seconds": 1.3266675979994034,
484
- "bytes": 5992849,
485
- "pages": 30,
 
 
 
 
 
 
 
 
 
 
 
 
486
  "shards": 2
487
  },
488
  {
489
- "seconds": 1.2236139670003467,
490
- "bytes": 7766768,
491
- "pages": 37,
492
  "shards": 3
493
  },
494
  {
495
- "seconds": 1.4731977419996838,
496
- "bytes": 6241189,
497
- "pages": 37,
498
- "shards": 3
499
  },
500
  {
501
- "seconds": 1.5922990050003136,
502
- "bytes": 11693360,
503
- "pages": 32,
 
 
 
 
 
 
504
  "shards": 3
505
  },
506
  {
507
- "seconds": 0.8287107050000486,
508
- "bytes": 2337614,
509
- "pages": 28,
510
- "shards": 2
511
  },
512
  {
513
- "seconds": 0.9556833940005163,
514
- "bytes": 6215473,
515
  "pages": 35,
516
- "shards": 2
517
  },
518
  {
519
- "seconds": 1.3753932430008717,
520
- "bytes": 9934254,
521
- "pages": 39,
522
  "shards": 3
523
  }
524
  ],
 
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": 74872.19943677516,
26
+ "required_pages_per_second_upper_estimate": 0.0064897324853250855
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": 9359.024929596895,
40
+ "required_pages_per_second_upper_estimate": 0.0037334549743086387
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": 18718.04985919379,
54
+ "required_pages_per_second_upper_estimate": 0.004330091408498271
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": 28077.074788790687,
68
+ "required_pages_per_second_upper_estimate": 0.0017097465456105776
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": 28077.074788790687,
82
+ "required_pages_per_second_upper_estimate": 0.002161450939558572
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": 28077.074788790687,
96
+ "required_pages_per_second_upper_estimate": 0.34862330095224164
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": 688946,
161
+ "producer_batch_seconds": 0.06762013767189501,
162
+ "producer_p95_seconds": 0.20516962799956673,
163
+ "gpu_update_seconds": 0.9847179668103464,
164
+ "producer_batches_per_second": 14.78849399645086,
165
+ "gpu_batches_per_second": 1.015519198089941,
166
+ "producer_headroom": 14.562495740372105,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.06779597543113,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 80,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 428682.5490002683,
174
+ "upload_bytes_per_second": 5225926.143782697,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 91.98529579645191,
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": 1.0256425439993109,
184
+ "bytes": 848800,
185
+ "pages": 26,
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 1.479707024000163,
190
+ "bytes": 13318632,
191
+ "pages": 43,
192
+ "shards": 4
193
+ },
194
+ {
195
+ "seconds": 1.157918292999966,
196
+ "bytes": 10292312,
197
+ "pages": 36,
198
+ "shards": 3
199
+ },
200
+ {
201
+ "seconds": 2.16527700000006,
202
+ "bytes": 1181305,
203
  "pages": 34,
204
+ "shards": 1
205
+ },
206
+ {
207
+ "seconds": 1.226145381000606,
208
+ "bytes": 7407097,
209
+ "pages": 30,
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 0.7928119780008274,
214
+ "bytes": 1141631,
215
  "pages": 33,
216
  "shards": 1
217
  },
218
  {
219
+ "seconds": 1.3008312719994137,
220
+ "bytes": 10073702,
221
+ "pages": 31,
222
+ "shards": 3
223
  },
224
  {
225
+ "seconds": 1.3391879739992874,
226
+ "bytes": 13717263,
227
+ "pages": 44,
228
+ "shards": 4
229
  },
230
  {
231
+ "seconds": 0.882563871999082,
232
+ "bytes": 1071171,
233
+ "pages": 33,
234
+ "shards": 1
235
  },
236
  {
237
+ "seconds": 1.5854626939999434,
238
+ "bytes": 7158059,
239
+ "pages": 27,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 1.1073119590000715,
244
+ "bytes": 10577372,
245
+ "pages": 37,
246
+ "shards": 3
247
  },
248
  {
249
+ "seconds": 0.7750392130001273,
250
+ "bytes": 1199165,
251
  "pages": 34,
252
+ "shards": 1
253
  },
254
  {
255
+ "seconds": 1.0709359819993551,
256
+ "bytes": 7553738,
257
+ "pages": 32,
258
  "shards": 3
259
  },
260
  {
261
+ "seconds": 0.9643511739996029,
262
+ "bytes": 6918272,
263
+ "pages": 35,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.0286668139997346,
268
+ "bytes": 9530498,
269
+ "pages": 29,
270
+ "shards": 3
271
  },
272
  {
273
+ "seconds": 0.9772154700003739,
274
+ "bytes": 2553165,
275
+ "pages": 32,
276
+ "shards": 2
277
  },
278
  {
279
+ "seconds": 1.3266675979994034,
280
+ "bytes": 5992849,
281
+ "pages": 30,
282
  "shards": 2
283
  },
284
  {
285
+ "seconds": 1.2236139670003467,
286
+ "bytes": 7766768,
287
+ "pages": 37,
288
+ "shards": 3
289
  },
290
  {
291
+ "seconds": 1.4731977419996838,
292
+ "bytes": 6241189,
293
+ "pages": 37,
294
+ "shards": 3
295
  },
296
  {
297
+ "seconds": 1.5922990050003136,
298
+ "bytes": 11693360,
299
+ "pages": 32,
300
+ "shards": 3
301
  },
302
  {
303
+ "seconds": 0.8287107050000486,
304
+ "bytes": 2337614,
305
+ "pages": 28,
306
  "shards": 2
307
  },
308
  {
309
+ "seconds": 0.9556833940005163,
310
+ "bytes": 6215473,
311
+ "pages": 35,
312
  "shards": 2
313
  },
314
  {
315
+ "seconds": 1.3753932430008717,
316
+ "bytes": 9934254,
317
+ "pages": 39,
318
+ "shards": 3
319
  },
320
  {
321
+ "seconds": 0.9314350770000601,
322
+ "bytes": 2320145,
323
+ "pages": 27,
324
+ "shards": 2
325
  },
326
  {
327
+ "seconds": 1.1146210560000327,
328
+ "bytes": 5994268,
329
+ "pages": 27,
330
  "shards": 2
331
  },
332
  {
333
+ "seconds": 0.7972013050002715,
334
+ "bytes": 688709,
335
+ "pages": 19,
336
+ "shards": 1
337
  },
338
  {
339
+ "seconds": 1.845211140000174,
340
+ "bytes": 18359200,
341
+ "pages": 44,
342
+ "shards": 5
343
  },
344
  {
345
+ "seconds": 0.8465310680003313,
346
+ "bytes": 1116330,
347
+ "pages": 33,
348
+ "shards": 1
349
  },
350
  {
351
+ "seconds": 1.1592603440003586,
352
+ "bytes": 7563694,
353
+ "pages": 38,
354
  "shards": 3
355
  },
356
  {
357
+ "seconds": 1.1315413200009061,
358
+ "bytes": 5996697,
359
  "pages": 35,
360
  "shards": 2
361
  },
362
  {
363
+ "seconds": 2.1342484679989866,
364
+ "bytes": 1010482,
365
+ "pages": 32,
366
  "shards": 1
367
  },
368
  {
369
+ "seconds": 1.0687636729999213,
370
+ "bytes": 11353142,
371
+ "pages": 26,
372
+ "shards": 4
373
+ },
374
+ {
375
+ "seconds": 1.1774445949995425,
376
+ "bytes": 12065836,
377
  "pages": 43,
378
+ "shards": 3
379
  },
380
  {
381
+ "seconds": 1.8629912129999866,
382
+ "bytes": 1107242,
383
+ "pages": 34,
384
  "shards": 1
385
  },
386
  {
387
+ "seconds": 1.2984945649986912,
388
+ "bytes": 7220956,
389
+ "pages": 28,
390
  "shards": 3
391
  },
392
  {
393
+ "seconds": 1.4053506099990045,
394
+ "bytes": 11055172,
395
+ "pages": 41,
 
 
 
 
 
 
396
  "shards": 3
397
  },
398
  {
399
+ "seconds": 0.8915826160009601,
400
+ "bytes": 2309383,
401
  "pages": 29,
402
+ "shards": 2
403
+ },
404
+ {
405
+ "seconds": 1.056229023000924,
406
+ "bytes": 11737562,
407
+ "pages": 38,
408
  "shards": 3
409
  },
410
  {
411
+ "seconds": 0.9255671579994669,
412
+ "bytes": 1181038,
413
+ "pages": 34,
414
  "shards": 1
415
  },
416
  {
417
+ "seconds": 3.174646825998934,
418
+ "bytes": 11736199,
419
+ "pages": 33,
420
  "shards": 4
421
  },
422
  {
423
+ "seconds": 1.0644383409999136,
424
+ "bytes": 5968582,
425
+ "pages": 35,
426
+ "shards": 2
427
  },
428
  {
429
+ "seconds": 0.9677518210010021,
430
+ "bytes": 2711632,
431
+ "pages": 36,
432
+ "shards": 2
433
  },
434
  {
435
+ "seconds": 1.13261040100042,
436
+ "bytes": 5584296,
437
+ "pages": 27,
438
+ "shards": 2
439
  },
440
  {
441
+ "seconds": 1.2312986870001623,
442
+ "bytes": 11373221,
443
+ "pages": 36,
444
+ "shards": 3
445
  },
446
  {
447
+ "seconds": 1.0926308529997186,
448
+ "bytes": 7644433,
449
+ "pages": 38,
450
  "shards": 3
451
  },
452
  {
453
+ "seconds": 1.0316966930004128,
454
+ "bytes": 5875792,
455
+ "pages": 27,
456
+ "shards": 2
457
  },
458
  {
459
+ "seconds": 0.9227722110008472,
460
+ "bytes": 1091891,
461
  "pages": 33,
462
  "shards": 1
463
  },
464
  {
465
+ "seconds": 1.4008473989997583,
466
+ "bytes": 11311428,
467
+ "pages": 36,
468
+ "shards": 4
469
  },
470
  {
471
+ "seconds": 1.1716579149997415,
472
+ "bytes": 6474393,
473
+ "pages": 35,
474
+ "shards": 2
475
  },
476
  {
477
+ "seconds": 0.9960888349996821,
478
+ "bytes": 2528980,
479
+ "pages": 33,
480
+ "shards": 3
481
  },
482
  {
483
+ "seconds": 1.481450629999017,
484
+ "bytes": 11596666,
485
+ "pages": 35,
486
  "shards": 3
487
  },
488
  {
489
+ "seconds": 1.1495631180005148,
490
+ "bytes": 5725671,
491
+ "pages": 33,
492
  "shards": 2
493
  },
494
  {
495
+ "seconds": 1.1277136550015712,
496
+ "bytes": 7537147,
497
+ "pages": 37,
498
  "shards": 3
499
  },
500
  {
501
+ "seconds": 1.212965857999734,
502
+ "bytes": 6163951,
503
+ "pages": 36,
504
  "shards": 2
505
  },
506
  {
507
+ "seconds": 1.403984237998884,
508
+ "bytes": 849560,
509
+ "pages": 25,
510
+ "shards": 1
511
+ },
512
+ {
513
+ "seconds": 2.95492070399996,
514
+ "bytes": 11249015,
515
+ "pages": 38,
516
+ "shards": 4
517
+ },
518
+ {
519
+ "seconds": 1.3934125020005013,
520
+ "bytes": 6804482,
521
+ "pages": 34,
522
  "shards": 2
523
  },
524
  {
525
+ "seconds": 1.1164342709998891,
526
+ "bytes": 7715646,
527
+ "pages": 35,
528
  "shards": 3
529
  },
530
  {
531
+ "seconds": 1.6021141640012502,
532
+ "bytes": 6045584,
533
+ "pages": 30,
534
+ "shards": 2
535
  },
536
  {
537
+ "seconds": 1.4088935970012244,
538
+ "bytes": 4960887,
539
+ "pages": 38,
540
+ "shards": 2
541
+ },
542
+ {
543
+ "seconds": 2.058075558999917,
544
+ "bytes": 7841052,
545
+ "pages": 38,
546
  "shards": 3
547
  },
548
  {
549
+ "seconds": 1.1063804570003413,
550
+ "bytes": 836703,
551
+ "pages": 25,
552
+ "shards": 1
553
  },
554
  {
555
+ "seconds": 1.6271591859986074,
556
+ "bytes": 16803522,
557
  "pages": 35,
558
+ "shards": 4
559
  },
560
  {
561
+ "seconds": 1.121872658999564,
562
+ "bytes": 7611150,
563
+ "pages": 37,
564
  "shards": 3
565
  }
566
  ],
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4aadd2411e27d45ea0852b13570a9cc811de9a400c6b05705bce64c357639250
3
- size 721766463
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68b2bd34b599c12b63f5eb6b5156f7905dc9454684b6f321753aaa307c958f77
3
+ size 721776127
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,