appvoid commited on
Commit
dc55ede
·
verified ·
1 Parent(s): 1aee838

Qwen90M step 200000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 195000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "e7124a2db2f213964f3cd249326d01e111d4002da26c586796a207bae4728dda",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "ba31b41d133fff78caf1b22d96d15106f826e2e03648c9dbff5ca60049e9ff2e",
12
- "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
- "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08",
14
- "shard_scheduler.json": "9382ee0711b8a32284204139fa207dcc6bdbed89a4d1e96d772bcf51dbded624",
15
- "coverage.json": "a52c19a6ab0dd4e5feeaf4d905f52d95523ae03ff4b83abba914a7c129d58cfb",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 200000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "9e1098c2f4131285db26132369fd315a230c2a1ed6433490bf04e1d55302bd5c",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "121ebbd424d06e2eca68e9473ccb7613c34366b9b444edaa47833ef5a5344919",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
+ "performance.json": "7f72f0a0e868a4baec06f238f922d0c5d7e397a51386c11ac32acc776d4d3522",
14
+ "shard_scheduler.json": "d840cc690a9276bb980e6c1c2e254aea3ef8a4071771396fd2e073516734a949",
15
+ "coverage.json": "33255e32c3d706b5219a7812b83fcf6ed8998528c0f3efa9d1b474d0202a80c0",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 195000,
3
  "stats": {
4
- "climbmix/blocks": 7020000,
5
- "climbmix/targets": 11732742810,
6
- "climbmix/input_tokens": 11733486473,
7
- "ultra_qa/blocks": 2632500,
8
- "ultra_qa/targets": 4295100300,
9
- "ultra_qa/input_tokens": 4295108575,
10
- "cortex/blocks": 2632500,
11
- "cortex/targets": 100135715,
12
- "cortex/input_tokens": 4184383493,
13
- "ultra_style/blocks": 2632500,
14
- "ultra_style/targets": 4163945713,
15
- "ultra_style/input_tokens": 4164164467,
16
- "openmath/blocks": 1755000,
17
- "openmath/targets": 2711634953,
18
- "openmath/input_tokens": 2712538826,
19
- "rewrite6/blocks": 877500,
20
- "rewrite6/targets": 1438575118,
21
- "rewrite6/input_tokens": 1440152299
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 961,
26
- "row": 3746,
27
  "start": {
28
- "file": 46,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 39,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 3940002
38
  },
39
  "rewrite6": {
40
- "page": 47,
41
- "row": 360,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 192512
47
  }
48
  },
49
- "epoch": 10,
50
  "offset": 0,
51
- "documents": 2246912
52
  },
53
  "openmath": {
54
- "page": 601,
55
- "row": 98,
56
  "start": {
57
  "file": 5,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 278,
61
- "row": 446
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 2048,
66
- "documents": 2461794
67
  },
68
  "ultra_qa": {
69
- "page": 259,
70
- "row": 3323,
71
  "start": {
72
  "file": 2,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
- "row": 21613
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 1064187
82
  },
83
  "ultra_style": {
84
- "page": 396,
85
- "row": 2960,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 2,
91
- "row": 81748
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 1624976
97
  },
98
  "cortex": {
99
- "page": 51053,
100
- "row": 14,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 6534798
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 200000,
3
  "stats": {
4
+ "climbmix/blocks": 7200000,
5
+ "climbmix/targets": 12033775305,
6
+ "climbmix/input_tokens": 12034537799,
7
+ "ultra_qa/blocks": 2700000,
8
+ "ultra_qa/targets": 4405129990,
9
+ "ultra_qa/input_tokens": 4405138485,
10
+ "cortex/blocks": 2700000,
11
+ "cortex/targets": 102703673,
12
+ "cortex/input_tokens": 4291696006,
13
+ "ultra_style/blocks": 2700000,
14
+ "ultra_style/targets": 4270419549,
15
+ "ultra_style/input_tokens": 4270644269,
16
+ "openmath/blocks": 1800000,
17
+ "openmath/targets": 2781111203,
18
+ "openmath/input_tokens": 2782038425,
19
+ "rewrite6/blocks": 900000,
20
+ "rewrite6/targets": 1475418140,
21
+ "rewrite6/input_tokens": 1477035705
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 986,
26
+ "row": 1904,
27
  "start": {
28
+ "file": 47,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 56,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 4096,
37
+ "documents": 4040560
38
  },
39
  "rewrite6": {
40
+ "page": 10,
41
+ "row": 4037,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 40960
47
  }
48
  },
49
+ "epoch": 11,
50
  "offset": 0,
51
+ "documents": 2304441
52
  },
53
  "openmath": {
54
+ "page": 616,
55
+ "row": 2034,
56
  "start": {
57
  "file": 5,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 339,
61
+ "row": 886
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 0,
66
+ "documents": 2525170
67
  },
68
  "ultra_qa": {
69
+ "page": 266,
70
+ "row": 1939,
71
  "start": {
72
  "file": 2,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
+ "row": 50285
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 1091475
82
  },
83
  "ultra_style": {
84
+ "page": 407,
85
+ "row": 311,
86
  "start": {
87
  "file": 2,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 3,
91
+ "row": 41668
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 1667383
97
  },
98
  "cortex": {
99
+ "page": 52363,
100
+ "row": 94,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 6702558
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 195000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "e7124a2db2f213964f3cd249326d01e111d4002da26c586796a207bae4728dda", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "ba31b41d133fff78caf1b22d96d15106f826e2e03648c9dbff5ca60049e9ff2e", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "d8fc0ab4ba9b75a14bc27c39b9cdaf8c3a3b38cccaa4c038b8370fdf2cd49c08", "shard_scheduler.json": "9382ee0711b8a32284204139fa207dcc6bdbed89a4d1e96d772bcf51dbded624", "coverage.json": "a52c19a6ab0dd4e5feeaf4d905f52d95523ae03ff4b83abba914a7c129d58cfb", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000195000"}
 
1
+ {"step": 200000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "9e1098c2f4131285db26132369fd315a230c2a1ed6433490bf04e1d55302bd5c", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "121ebbd424d06e2eca68e9473ccb7613c34366b9b444edaa47833ef5a5344919", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "7f72f0a0e868a4baec06f238f922d0c5d7e397a51386c11ac32acc776d4d3522", "shard_scheduler.json": "d840cc690a9276bb980e6c1c2e254aea3ef8a4071771396fd2e073516734a949", "coverage.json": "33255e32c3d706b5219a7812b83fcf6ed8998528c0f3efa9d1b474d0202a80c0", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000200000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e7124a2db2f213964f3cd249326d01e111d4002da26c586796a207bae4728dda
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e1098c2f4131285db26132369fd315a230c2a1ed6433490bf04e1d55302bd5c
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.8980338759999995,
53
  "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9548798630000306,
56
- "input_tokens_per_second": 196126.2316567667
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9016621779999809,
53
  "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9578150019999612,
56
+ "input_tokens_per_second": 195337.01678679453
57
  },
58
  "measurements": [
59
  {
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 74454.29039832999,
26
- "required_pages_per_second_upper_estimate": 0.006453509189053433
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": 9306.786299791249,
40
- "required_pages_per_second_upper_estimate": 0.0037126162038420493
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": 18613.572599582498,
54
- "required_pages_per_second_upper_estimate": 0.004305922433224167
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": 27920.35889937375,
68
- "required_pages_per_second_upper_estimate": 0.001700203370169813
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": 27920.35889937375,
82
- "required_pages_per_second_upper_estimate": 0.0021493865165739087
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.042780299000014566,
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 21,
95
- "required_token_positions_per_second": 27920.35889937375,
96
- "required_pages_per_second_upper_estimate": 0.34667741410002545
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,413 +155,125 @@
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": 0.10249432328106423,
162
- "producer_p95_seconds": 0.39124788300250657,
163
- "gpu_update_seconds": 0.9902451504883822,
164
- "producer_batches_per_second": 9.756637909182132,
165
- "gpu_batches_per_second": 1.0098509439877659,
166
- "producer_headroom": 9.661463374638716,
167
  "sustainable": true,
168
- "buffer_seconds": 60.40495417979132,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 84,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 328429.3339946892,
174
- "upload_bytes_per_second": 4846781.735776416,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 70.47328963165768,
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": 1.207307631000731,
184
- "bytes": 6036390,
185
- "pages": 36,
186
- "shards": 2
187
  },
188
  {
189
- "seconds": 1.257034783000563,
190
- "bytes": 2415905,
191
  "pages": 31,
192
  "shards": 3
193
  },
194
  {
195
- "seconds": 1.2379260390007403,
196
- "bytes": 10087351,
197
- "pages": 39,
198
- "shards": 3
199
- },
200
- {
201
- "seconds": 1.0402059740008553,
202
- "bytes": 6293747,
203
- "pages": 28,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 1.1847501439988264,
208
- "bytes": 2572744,
209
- "pages": 33,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.268939194000268,
214
- "bytes": 16767391,
215
- "pages": 42,
216
  "shards": 4
217
  },
218
  {
219
- "seconds": 1.2854500899993582,
220
- "bytes": 7334524,
221
- "pages": 31,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 1.0401804770008312,
226
- "bytes": 1254870,
227
  "pages": 34,
228
  "shards": 1
229
  },
230
  {
231
- "seconds": 1.428139841002121,
232
- "bytes": 6304931,
233
- "pages": 37,
234
- "shards": 2
235
- },
236
- {
237
- "seconds": 0.987766936996195,
238
- "bytes": 2378061,
239
- "pages": 29,
240
  "shards": 2
241
  },
242
  {
243
- "seconds": 1.0273188179999124,
244
- "bytes": 9379394,
245
- "pages": 33,
246
- "shards": 3
247
- },
248
- {
249
- "seconds": 1.4798621720037772,
250
- "bytes": 13649971,
251
- "pages": 37,
252
  "shards": 4
253
  },
254
  {
255
- "seconds": 0.8202600639997399,
256
- "bytes": 1256174,
257
  "pages": 35,
258
- "shards": 1
259
- },
260
- {
261
- "seconds": 1.0791088629994192,
262
- "bytes": 10296860,
263
- "pages": 35,
264
- "shards": 3
265
- },
266
- {
267
- "seconds": 1.2926067090011202,
268
- "bytes": 2391669,
269
- "pages": 32,
270
- "shards": 2
271
- },
272
- {
273
- "seconds": 1.033323986994219,
274
- "bytes": 6001182,
275
- "pages": 33,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 3.9172589509980753,
280
- "bytes": 5890320,
281
- "pages": 32,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 1.3426188599987654,
286
- "bytes": 12956908,
287
- "pages": 37,
288
- "shards": 4
289
- },
290
- {
291
- "seconds": 1.2346246180022717,
292
- "bytes": 6315095,
293
- "pages": 36,
294
- "shards": 2
295
- },
296
- {
297
- "seconds": 3.684220068003924,
298
- "bytes": 2396656,
299
- "pages": 29,
300
- "shards": 2
301
- },
302
- {
303
- "seconds": 1.1850955770059954,
304
- "bytes": 6553717,
305
- "pages": 37,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 1.714512132006348,
310
- "bytes": 6167274,
311
- "pages": 33,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.2202410780009814,
316
- "bytes": 6552255,
317
- "pages": 33,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 1.1842839309974806,
322
- "bytes": 12073153,
323
- "pages": 39,
324
- "shards": 3
325
- },
326
- {
327
- "seconds": 0.9938246039964724,
328
- "bytes": 235889,
329
- "pages": 8,
330
- "shards": 1
331
- },
332
- {
333
- "seconds": 1.3448229619971244,
334
- "bytes": 7303500,
335
- "pages": 31,
336
  "shards": 3
337
  },
338
  {
339
- "seconds": 1.1307224229967687,
340
- "bytes": 5856127,
341
- "pages": 37,
342
- "shards": 2
343
- },
344
- {
345
- "seconds": 1.2856477119930787,
346
- "bytes": 6096134,
347
- "pages": 29,
348
- "shards": 2
349
- },
350
- {
351
- "seconds": 1.163583995999943,
352
- "bytes": 2620539,
353
- "pages": 37,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.2240531869974802,
358
- "bytes": 11650354,
359
- "pages": 31,
360
  "shards": 3
361
  },
362
  {
363
- "seconds": 1.2925304770033108,
364
- "bytes": 12240192,
365
- "pages": 41,
366
- "shards": 4
367
- },
368
- {
369
- "seconds": 0.9206129219965078,
370
- "bytes": 852268,
371
- "pages": 27,
372
- "shards": 1
373
- },
374
- {
375
- "seconds": 1.2395337650013971,
376
- "bytes": 6262507,
377
  "pages": 38,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 1.0192943180009024,
382
- "bytes": 2301578,
383
  "pages": 28,
384
- "shards": 2
385
- },
386
- {
387
- "seconds": 2.074865057998977,
388
- "bytes": 9907182,
389
- "pages": 39,
390
- "shards": 3
391
- },
392
- {
393
- "seconds": 1.498775358995772,
394
- "bytes": 5765034,
395
- "pages": 28,
396
- "shards": 2
397
- },
398
- {
399
- "seconds": 1.3661099450037,
400
- "bytes": 8604864,
401
- "pages": 41,
402
- "shards": 3
403
  },
404
  {
405
- "seconds": 1.0984054020009353,
406
- "bytes": 6085035,
407
  "pages": 29,
408
  "shards": 2
409
  },
410
  {
411
- "seconds": 1.4394773240055656,
412
- "bytes": 12207803,
413
- "pages": 42,
414
- "shards": 4
415
- },
416
- {
417
- "seconds": 0.8468366760062054,
418
- "bytes": 871307,
419
- "pages": 27,
420
- "shards": 1
421
- },
422
- {
423
- "seconds": 1.2028667299964582,
424
- "bytes": 6129254,
425
  "pages": 38,
426
- "shards": 2
427
- },
428
- {
429
- "seconds": 1.6012624290015083,
430
- "bytes": 8291413,
431
- "pages": 33,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 1.3447834969992982,
436
- "bytes": 10108218,
437
  "pages": 32,
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 1.1404380909953034,
442
- "bytes": 7821336,
443
- "pages": 39,
444
- "shards": 3
445
- },
446
- {
447
- "seconds": 1.2159917039971333,
448
- "bytes": 870322,
449
- "pages": 29,
450
- "shards": 1
451
- },
452
- {
453
- "seconds": 1.047054365000804,
454
- "bytes": 6014250,
455
- "pages": 35,
456
- "shards": 2
457
- },
458
- {
459
- "seconds": 1.1608161260010093,
460
- "bytes": 6519078,
461
- "pages": 32,
462
- "shards": 3
463
- },
464
- {
465
- "seconds": 1.5200698819971876,
466
- "bytes": 6082599,
467
- "pages": 34,
468
- "shards": 2
469
- },
470
- {
471
- "seconds": 1.4744428199992399,
472
- "bytes": 11695751,
473
- "pages": 38,
474
- "shards": 3
475
- },
476
- {
477
- "seconds": 1.3584106329944916,
478
- "bytes": 2483287,
479
  "pages": 36,
480
  "shards": 2
481
- },
482
- {
483
- "seconds": 1.1772192700009327,
484
- "bytes": 10757429,
485
- "pages": 34,
486
- "shards": 3
487
- },
488
- {
489
- "seconds": 2.1233109069944476,
490
- "bytes": 2398655,
491
- "pages": 29,
492
- "shards": 2
493
- },
494
- {
495
- "seconds": 1.0890797859974555,
496
- "bytes": 6225655,
497
- "pages": 38,
498
- "shards": 2
499
- },
500
- {
501
- "seconds": 3.324964125000406,
502
- "bytes": 5816918,
503
- "pages": 28,
504
- "shards": 2
505
- },
506
- {
507
- "seconds": 1.304179352002393,
508
- "bytes": 11867177,
509
- "pages": 37,
510
- "shards": 4
511
- },
512
- {
513
- "seconds": 1.3306422289970214,
514
- "bytes": 6277468,
515
- "pages": 38,
516
- "shards": 2
517
- },
518
- {
519
- "seconds": 0.9982111080025788,
520
- "bytes": 877422,
521
- "pages": 26,
522
- "shards": 1
523
- },
524
- {
525
- "seconds": 1.6349642249988392,
526
- "bytes": 7479267,
527
- "pages": 33,
528
- "shards": 3
529
- },
530
- {
531
- "seconds": 1.1783448380010668,
532
- "bytes": 10185058,
533
- "pages": 38,
534
- "shards": 3
535
- },
536
- {
537
- "seconds": 0.9609743260007235,
538
- "bytes": 2529461,
539
- "pages": 33,
540
- "shards": 2
541
- },
542
- {
543
- "seconds": 1.5562968439990073,
544
- "bytes": 11835464,
545
- "pages": 39,
546
- "shards": 3
547
- },
548
- {
549
- "seconds": 0.8645425069989869,
550
- "bytes": 693988,
551
- "pages": 20,
552
- "shards": 1
553
- },
554
- {
555
- "seconds": 1.3442884890027926,
556
- "bytes": 12581350,
557
- "pages": 41,
558
- "shards": 4
559
- },
560
- {
561
- "seconds": 1.4273276979947696,
562
- "bytes": 6232998,
563
- "pages": 37,
564
- "shards": 3
565
  }
566
  ],
567
  "cache_key": {
@@ -606,7 +318,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 2.7487001069999906,
610
- 0.015045025999995687
611
  ]
612
  }
 
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 75216.18075821365,
26
+ "required_pages_per_second_upper_estimate": 0.006519547914454685
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": 9402.022594776707,
40
+ "required_pages_per_second_upper_estimate": 0.003750607385821249
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": 18804.045189553413,
54
+ "required_pages_per_second_upper_estimate": 0.0043499849147108464
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": 28206.06778433012,
68
+ "required_pages_per_second_upper_estimate": 0.0017176015422649853
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": 28206.06778433012,
82
+ "required_pages_per_second_upper_estimate": 0.002171381177430663
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.042780299000014566,
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 22,
95
+ "required_token_positions_per_second": 28206.06778433012,
96
+ "required_pages_per_second_upper_estimate": 0.35022496224505656
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
157
  "selected": {
158
+ "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
+ "producer_batch_seconds": 0.10476856211718655,
162
+ "producer_p95_seconds": 0.18885634199978085,
163
+ "gpu_update_seconds": 0.980214619471341,
164
+ "producer_batches_per_second": 9.544847994396184,
165
+ "gpu_batches_per_second": 1.0201847433568476,
166
+ "producer_headroom": 9.355999544738848,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.773306407223146,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 81,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 364056.88282280555,
174
+ "upload_bytes_per_second": 5423901.113376007,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 78.11813224328161,
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": 1.3317575169999145,
184
+ "bytes": 24816,
185
+ "pages": 1,
186
+ "shards": 1
187
  },
188
  {
189
+ "seconds": 1.540608198999962,
190
+ "bytes": 9815303,
191
  "pages": 31,
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 1.441364390000217,
196
+ "bytes": 13613764,
197
+ "pages": 38,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
198
  "shards": 4
199
  },
200
  {
201
+ "seconds": 0.8236200929995903,
202
+ "bytes": 1218836,
 
 
 
 
 
 
203
  "pages": 34,
204
  "shards": 1
205
  },
206
  {
207
+ "seconds": 1.1421018979999644,
208
+ "bytes": 6132811,
209
+ "pages": 34,
 
 
 
 
 
 
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.2374535429999014,
214
+ "bytes": 7166128,
215
+ "pages": 34,
 
 
 
 
 
 
216
  "shards": 4
217
  },
218
  {
219
+ "seconds": 1.0863462409997737,
220
+ "bytes": 6279325,
221
  "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 0.9579987879997134,
226
+ "bytes": 7302822,
227
+ "pages": 28,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
228
  "shards": 3
229
  },
230
  {
231
+ "seconds": 0.9735736169996017,
232
+ "bytes": 6894198,
233
+ "pages": 36,
 
 
 
 
 
 
 
 
 
 
 
 
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.282220471999608,
238
+ "bytes": 9425496,
239
+ "pages": 34,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 1.236358336000194,
244
+ "bytes": 7782594,
 
 
 
 
 
 
 
 
 
 
 
 
245
  "pages": 38,
246
  "shards": 3
247
  },
248
  {
249
+ "seconds": 1.0319873409998763,
250
+ "bytes": 964782,
251
  "pages": 28,
252
+ "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
253
  },
254
  {
255
+ "seconds": 1.3289059700000507,
256
+ "bytes": 6144064,
257
  "pages": 29,
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 1.1848053769999751,
262
+ "bytes": 6754320,
 
 
 
 
 
 
 
 
 
 
 
 
263
  "pages": 38,
 
 
 
 
 
 
264
  "shards": 3
265
  },
266
  {
267
+ "seconds": 1.6214569329995356,
268
+ "bytes": 11854578,
269
  "pages": 32,
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 0.9589299640001627,
274
+ "bytes": 2653813,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
275
  "pages": 36,
276
  "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
277
  }
278
  ],
279
  "cache_key": {
 
318
  },
319
  "cache_reused": true,
320
  "resume_probe_seconds": [
321
+ 3.044028898000022,
322
+ 0.01595121099995822
323
  ]
324
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ba31b41d133fff78caf1b22d96d15106f826e2e03648c9dbff5ca60049e9ff2e
3
- size 721763199
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:121ebbd424d06e2eca68e9473ccb7613c34366b9b444edaa47833ef5a5344919
3
+ size 721772159
training_config.json CHANGED
@@ -46,8 +46,8 @@
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,
 
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,