appvoid commited on
Commit
9b9306a
·
verified ·
1 Parent(s): 74ce2a1

Qwen90M step 845000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 842507,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "6f8382631e83ae42ed9759e16b99adad61f343fa6de9cb2d60b9dd6c9148bc70",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "a2b248de903e43cc04b64f31262641a2a596e6720f6e8b83622de3cb8a1ad33a",
12
- "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
- "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7",
14
- "shard_scheduler.json": "76298eb667981f18a15e74c27926fbcd7a9ae27df4f2ad6d655bc8499e520b76",
15
- "coverage.json": "b6f294dc008ba78304b1fa2017cef88912e1ceaaf314ebcebdb825482582a1b3",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
 
1
  {
2
+ "step": 845000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "28d525df26bddcdad1f7e0ee21d00c4e2f6f2750838939feccfbef122663248b",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "87fc13206a6115414dc5698ba185a42d91483c871e4ad1d988dd31693f1ea24e",
12
+ "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58",
13
+ "performance.json": "172134497605e50805257548c05018674e515323c63930c6237cd46d43786367",
14
+ "shard_scheduler.json": "57f3f5db4104672d1623ef748776a5e35f680600431f1b4c0d4f28fc3f966af9",
15
+ "coverage.json": "e2b9a86150908d61f021a350e36374564851cf14a57027b0b71fa60173ea1a90",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 842507,
3
  "stats": {
4
- "climbmix/blocks": 30330252,
5
- "climbmix/targets": 50701917572,
6
- "climbmix/input_tokens": 50705130186,
7
- "ultra_qa/blocks": 11373844,
8
- "ultra_qa/targets": 18555701427,
9
- "ultra_qa/input_tokens": 18555737118,
10
- "cortex/blocks": 11373845,
11
- "cortex/targets": 432503890,
12
- "cortex/input_tokens": 18079643538,
13
- "ultra_style/blocks": 11373844,
14
- "ultra_style/targets": 17989747078,
15
- "ultra_style/input_tokens": 17990690711,
16
- "openmath/blocks": 7582563,
17
- "openmath/targets": 11716447202,
18
- "openmath/input_tokens": 11720351418,
19
- "rewrite6/blocks": 3791282,
20
- "rewrite6/targets": 6215294727,
21
- "rewrite6/input_tokens": 6222108705
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 4153,
26
- "row": 3572,
27
  "start": {
28
  "file": 201,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 2,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 17014260
38
  },
39
  "rewrite6": {
40
- "page": 13,
41
- "row": 316,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 53248
47
  }
48
  },
49
  "epoch": 47,
50
  "offset": 2048,
51
- "documents": 9707552
52
  },
53
  "openmath": {
54
- "page": 2596,
55
- "row": 750,
56
  "start": {
57
  "file": 24,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 153,
61
- "row": 617
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 0,
66
- "documents": 10633966
67
  },
68
  "ultra_qa": {
69
- "page": 1122,
70
- "row": 1847,
71
  "start": {
72
  "file": 8,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
- "row": 51467
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 4597559
82
  },
83
  "ultra_style": {
84
- "page": 1713,
85
- "row": 1089,
86
  "start": {
87
  "file": 10,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 1,
91
- "row": 81704
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 7017537
97
  },
98
  "cortex": {
99
- "page": 220582,
100
- "row": 48,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 28234544
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 845000,
3
  "stats": {
4
+ "climbmix/blocks": 30420000,
5
+ "climbmix/targets": 50852257328,
6
+ "climbmix/input_tokens": 50855479577,
7
+ "ultra_qa/blocks": 11407500,
8
+ "ultra_qa/targets": 18610655674,
9
+ "ultra_qa/input_tokens": 18610691463,
10
+ "cortex/blocks": 11407500,
11
+ "cortex/targets": 433787145,
12
+ "cortex/input_tokens": 18133094420,
13
+ "ultra_style/blocks": 11407500,
14
+ "ultra_style/targets": 18043082009,
15
+ "ultra_style/input_tokens": 18044028468,
16
+ "openmath/blocks": 7605000,
17
+ "openmath/targets": 11751095221,
18
+ "openmath/input_tokens": 11755011037,
19
+ "rewrite6/blocks": 3802500,
20
+ "rewrite6/targets": 6233650055,
21
+ "rewrite6/input_tokens": 6240484559
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 4166,
26
+ "row": 476,
27
  "start": {
28
  "file": 201,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 54,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 0,
37
+ "documents": 17064412
38
  },
39
  "rewrite6": {
40
+ "page": 20,
41
+ "row": 812,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 81920
47
  }
48
  },
49
  "epoch": 47,
50
  "offset": 2048,
51
+ "documents": 9736720
52
  },
53
  "openmath": {
54
+ "page": 2603,
55
+ "row": 3612,
56
  "start": {
57
  "file": 24,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 182,
61
+ "row": 289
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 2048,
66
+ "documents": 10665500
67
  },
68
  "ultra_qa": {
69
+ "page": 1125,
70
+ "row": 3194,
71
  "start": {
72
  "file": 8,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
+ "row": 63755
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 4611194
82
  },
83
  "ultra_style": {
84
+ "page": 1718,
85
+ "row": 1338,
86
  "start": {
87
  "file": 10,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 2,
91
+ "row": 17547
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 4096,
96
+ "documents": 7038266
97
  },
98
  "cortex": {
99
+ "page": 221233,
100
+ "row": 96,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 28317920
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 842507, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "6f8382631e83ae42ed9759e16b99adad61f343fa6de9cb2d60b9dd6c9148bc70", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "a2b248de903e43cc04b64f31262641a2a596e6720f6e8b83622de3cb8a1ad33a", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7", "shard_scheduler.json": "76298eb667981f18a15e74c27926fbcd7a9ae27df4f2ad6d655bc8499e520b76", "coverage.json": "b6f294dc008ba78304b1fa2017cef88912e1ceaaf314ebcebdb825482582a1b3", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000842507"}
 
1
+ {"step": 845000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "28d525df26bddcdad1f7e0ee21d00c4e2f6f2750838939feccfbef122663248b", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "87fc13206a6115414dc5698ba185a42d91483c871e4ad1d988dd31693f1ea24e", "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58", "performance.json": "172134497605e50805257548c05018674e515323c63930c6237cd46d43786367", "shard_scheduler.json": "57f3f5db4104672d1623ef748776a5e35f680600431f1b4c0d4f28fc3f966af9", "coverage.json": "e2b9a86150908d61f021a350e36374564851cf14a57027b0b71fa60173ea1a90", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000845000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6f8382631e83ae42ed9759e16b99adad61f343fa6de9cb2d60b9dd6c9148bc70
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:28d525df26bddcdad1f7e0ee21d00c4e2f6f2750838939feccfbef122663248b
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9411944559999483,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9411944559999483,
56
- "input_tokens_per_second": 195836.25766704488
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9390849369999614,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9390849369999614,
56
+ "input_tokens_per_second": 196276.1756022156
57
  },
58
  "measurements": [
59
  {
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 76375.6706640414,
26
- "required_pages_per_second_upper_estimate": 0.005283846112509921
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
36
  "tokens_per_second": 8533.957213384612,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
- "required_token_positions_per_second": 9546.958833005176,
40
- "required_pages_per_second_upper_estimate": 0.003997112300208492
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
50
  "tokens_per_second": 380144.35761751566,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
- "required_token_positions_per_second": 19093.91766601035,
54
- "required_pages_per_second_upper_estimate": 0.003976759509359474
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
64
  "tokens_per_second": 43742.16556775435,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
- "required_token_positions_per_second": 28640.876499015525,
68
- "required_pages_per_second_upper_estimate": 0.0017169768140297367
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
78
  "tokens_per_second": 22018.86276523721,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
- "required_token_positions_per_second": 28640.876499015525,
82
- "required_pages_per_second_upper_estimate": 0.002226702169896493
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 23,
95
- "required_token_positions_per_second": 28640.876499015525,
96
- "required_pages_per_second_upper_estimate": 0.36737441155212897
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,412 +155,154 @@
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
- "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.15020428112444506,
162
- "producer_p95_seconds": 0.3366278989997227,
163
- "gpu_update_seconds": 0.9653335854072185,
164
- "producer_batches_per_second": 6.657599853438895,
165
- "gpu_batches_per_second": 1.0359113317062907,
166
- "producer_headroom": 6.426804736726741,
167
  "sustainable": true,
168
- "buffer_seconds": 60.81601588065477,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 79,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 273762.60107156524,
174
- "upload_bytes_per_second": 4419757.457076919,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 58.74308132276743,
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": 1.4560003849983332,
184
- "bytes": 6473170,
185
- "pages": 37,
186
- "shards": 2
187
- },
188
- {
189
- "seconds": 1.6064382549957372,
190
- "bytes": 6604109,
191
- "pages": 34,
192
- "shards": 3
193
- },
194
- {
195
- "seconds": 1.684055683996121,
196
- "bytes": 5176791,
197
- "pages": 11,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.8713450549985282,
202
- "bytes": 942860,
203
- "pages": 27,
204
  "shards": 1
205
  },
206
  {
207
- "seconds": 1.5147400999994716,
208
- "bytes": 6664610,
209
- "pages": 37,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.4935083919990575,
214
- "bytes": 2459570,
215
- "pages": 28,
216
- "shards": 2
217
- },
218
- {
219
- "seconds": 2.1903198990039527,
220
- "bytes": 16888519,
221
- "pages": 41,
222
- "shards": 4
223
- },
224
- {
225
- "seconds": 2.158359011998982,
226
- "bytes": 7618380,
227
- "pages": 40,
228
- "shards": 3
229
- },
230
- {
231
- "seconds": 1.546953492994362,
232
- "bytes": 982772,
233
- "pages": 27,
234
  "shards": 1
235
  },
236
  {
237
- "seconds": 1.5179005070021958,
238
- "bytes": 6221682,
239
- "pages": 38,
240
- "shards": 2
241
- },
242
- {
243
- "seconds": 1.4849864080024417,
244
- "bytes": 6331011,
245
- "pages": 36,
246
- "shards": 3
247
- },
248
- {
249
- "seconds": 1.7072867230017437,
250
- "bytes": 11821281,
251
- "pages": 36,
252
- "shards": 3
253
- },
254
- {
255
- "seconds": 1.688430515001528,
256
- "bytes": 5778338,
257
- "pages": 29,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.6583789540018188,
262
- "bytes": 2527741,
263
- "pages": 35,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 1.606587074995332,
268
- "bytes": 10744249,
269
- "pages": 39,
270
- "shards": 4
271
- },
272
- {
273
- "seconds": 1.0940176939984667,
274
- "bytes": 7184499,
275
- "pages": 30,
276
- "shards": 3
277
- },
278
- {
279
- "seconds": 1.0425019949980197,
280
- "bytes": 1083508,
281
- "pages": 36,
282
  "shards": 1
283
  },
284
  {
285
- "seconds": 1.5655531079974025,
286
- "bytes": 12067871,
287
- "pages": 35,
288
- "shards": 3
289
- },
290
- {
291
- "seconds": 1.9511204809969058,
292
- "bytes": 6917854,
293
- "pages": 33,
294
- "shards": 3
295
- },
296
- {
297
- "seconds": 1.1665768739985651,
298
- "bytes": 5996687,
299
- "pages": 35,
300
- "shards": 2
301
- },
302
- {
303
- "seconds": 1.6431592789958813,
304
- "bytes": 7717315,
305
- "pages": 37,
306
- "shards": 3
307
- },
308
- {
309
- "seconds": 1.3486441710047075,
310
- "bytes": 976930,
311
- "pages": 30,
312
  "shards": 1
313
  },
314
  {
315
- "seconds": 1.5532393190005678,
316
- "bytes": 10410458,
317
- "pages": 39,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 1.5896227589983027,
322
- "bytes": 8024141,
323
- "pages": 26,
324
  "shards": 3
325
  },
326
  {
327
- "seconds": 1.570418814997538,
328
- "bytes": 6148151,
329
- "pages": 41,
330
- "shards": 2
331
- },
332
- {
333
- "seconds": 1.876006619000691,
334
- "bytes": 11510011,
335
- "pages": 30,
336
  "shards": 4
337
  },
338
  {
339
- "seconds": 1.1747664850045112,
340
- "bytes": 1216277,
341
- "pages": 36,
342
- "shards": 1
343
- },
344
- {
345
- "seconds": 1.0712760539972805,
346
- "bytes": 6301436,
347
  "pages": 37,
348
  "shards": 2
349
  },
350
  {
351
- "seconds": 1.2888189970035455,
352
- "bytes": 7385771,
353
- "pages": 32,
354
- "shards": 3
355
- },
356
- {
357
- "seconds": 1.5881777600006899,
358
- "bytes": 5494719,
359
- "pages": 25,
360
- "shards": 2
361
- },
362
- {
363
- "seconds": 1.394116926996503,
364
- "bytes": 11997464,
365
- "pages": 39,
366
- "shards": 3
367
- },
368
- {
369
- "seconds": 1.7342578780007898,
370
- "bytes": 2696654,
371
- "pages": 39,
372
- "shards": 2
373
- },
374
- {
375
- "seconds": 1.611772108997684,
376
- "bytes": 5903925,
377
- "pages": 29,
378
- "shards": 2
379
- },
380
- {
381
- "seconds": 1.512940188003995,
382
- "bytes": 7372289,
383
  "pages": 35,
384
- "shards": 3
385
- },
386
- {
387
- "seconds": 0.9355933799961349,
388
- "bytes": 5063118,
389
- "pages": 34,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 1.774848651002685,
394
- "bytes": 5953016,
395
- "pages": 29,
396
- "shards": 2
397
  },
398
  {
399
- "seconds": 1.877187440994021,
400
- "bytes": 13513556,
401
- "pages": 41,
402
  "shards": 4
403
  },
404
  {
405
- "seconds": 1.8802343610004755,
406
- "bytes": 5656226,
407
- "pages": 32,
408
  "shards": 2
409
  },
410
  {
411
- "seconds": 1.6476056760002393,
412
- "bytes": 7839870,
413
- "pages": 38,
414
- "shards": 3
415
- },
416
- {
417
- "seconds": 1.3661578879982699,
418
- "bytes": 1247022,
419
- "pages": 35,
420
- "shards": 1
421
- },
422
- {
423
- "seconds": 1.4103239969990682,
424
- "bytes": 5948663,
425
  "pages": 28,
426
  "shards": 2
427
  },
428
  {
429
- "seconds": 1.2626624109980185,
430
- "bytes": 7374458,
431
- "pages": 32,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 1.3394468149999739,
436
- "bytes": 10758114,
437
- "pages": 29,
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 1.5901181910012383,
442
- "bytes": 260770,
443
- "pages": 9,
444
- "shards": 1
445
- },
446
- {
447
- "seconds": 1.5258419460005825,
448
- "bytes": 5807963,
449
- "pages": 29,
450
- "shards": 2
451
- },
452
- {
453
- "seconds": 1.2447194750056951,
454
- "bytes": 7832230,
455
- "pages": 41,
456
- "shards": 4
457
- },
458
- {
459
- "seconds": 0.9895193189950078,
460
- "bytes": 741196,
461
- "pages": 23,
462
- "shards": 1
463
- },
464
- {
465
- "seconds": 1.489227063000726,
466
- "bytes": 11724055,
467
- "pages": 40,
468
- "shards": 4
469
- },
470
- {
471
- "seconds": 0.9323788800029433,
472
- "bytes": 1111114,
473
- "pages": 34,
474
- "shards": 1
475
- },
476
- {
477
- "seconds": 1.5997961780012702,
478
- "bytes": 11835967,
479
- "pages": 35,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 1.3631727270039846,
484
- "bytes": 2146631,
485
- "pages": 21,
486
  "shards": 2
487
  },
488
  {
489
- "seconds": 1.2327626959959161,
490
- "bytes": 11052547,
491
- "pages": 48,
492
- "shards": 3
493
- },
494
- {
495
- "seconds": 1.4408944379974855,
496
- "bytes": 7298137,
497
- "pages": 31,
498
- "shards": 3
499
- },
500
- {
501
- "seconds": 1.4694450449969736,
502
- "bytes": 1110429,
503
- "pages": 36,
504
  "shards": 1
505
  },
506
  {
507
- "seconds": 1.0526011850015493,
508
- "bytes": 6120517,
509
- "pages": 31,
510
- "shards": 2
511
- },
512
- {
513
- "seconds": 1.7904460849968018,
514
- "bytes": 12133874,
515
- "pages": 26,
516
- "shards": 4
517
- },
518
- {
519
- "seconds": 1.269238109998696,
520
- "bytes": 6277214,
521
- "pages": 37,
522
- "shards": 2
523
- },
524
- {
525
- "seconds": 1.02439347699692,
526
- "bytes": 5979878,
527
- "pages": 29,
528
- "shards": 2
529
  },
530
  {
531
- "seconds": 1.1111059189934167,
532
- "bytes": 2867400,
533
- "pages": 39,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 1.2084345140028745,
538
- "bytes": 10491799,
539
- "pages": 30,
540
- "shards": 3
541
- },
542
- {
543
- "seconds": 0.8811795219953638,
544
- "bytes": 776994,
545
- "pages": 26,
546
- "shards": 1
547
- },
548
- {
549
- "seconds": 1.837297693004075,
550
- "bytes": 13040260,
551
- "pages": 38,
552
- "shards": 4
553
- },
554
- {
555
- "seconds": 1.5299789529963164,
556
- "bytes": 6406874,
557
- "pages": 38,
558
  "shards": 2
559
  },
560
  {
561
- "seconds": 1.7032136210036697,
562
- "bytes": 6726184,
563
- "pages": 31,
564
  "shards": 3
565
  }
566
  ],
@@ -606,7 +348,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 3.328129581999974,
610
- 0.10924991900014902
611
  ]
612
  }
 
22
  "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 76039.94462746628,
26
+ "required_pages_per_second_upper_estimate": 0.005260619806307925
27
  },
28
  {
29
  "source": "rewrite6",
 
36
  "tokens_per_second": 8533.957213384612,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
+ "required_token_positions_per_second": 9504.993078433285,
40
+ "required_pages_per_second_upper_estimate": 0.003979542115113849
41
  },
42
  {
43
  "source": "openmath",
 
50
  "tokens_per_second": 380144.35761751566,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
+ "required_token_positions_per_second": 19009.98615686657,
54
+ "required_pages_per_second_upper_estimate": 0.003959278789427566
55
  },
56
  {
57
  "source": "ultra_qa",
 
64
  "tokens_per_second": 43742.16556775435,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
+ "required_token_positions_per_second": 28514.979235299852,
68
+ "required_pages_per_second_upper_estimate": 0.001709429465303973
69
  },
70
  {
71
  "source": "ultra_style",
 
78
  "tokens_per_second": 22018.86276523721,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
+ "required_token_positions_per_second": 28514.979235299852,
82
+ "required_pages_per_second_upper_estimate": 0.002216914211406139
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 22,
95
+ "required_token_positions_per_second": 28514.979235299852,
96
+ "required_pages_per_second_upper_estimate": 0.3657595366311342
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
+ "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.05139413005469606,
162
+ "producer_p95_seconds": 0.15724994399988645,
163
+ "gpu_update_seconds": 0.969595656088482,
164
+ "producer_batches_per_second": 19.457474986652226,
165
+ "gpu_batches_per_second": 1.0313577559063893,
166
+ "producer_headroom": 18.86588322550829,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.11493067748588,
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": 403027.47604566225,
174
+ "upload_bytes_per_second": 3951706.6780117266,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 86.48031435992623,
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": 1.4923919180000098,
184
+ "bytes": 40304,
185
+ "pages": 1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 1.3177459699999758,
190
+ "bytes": 94124,
191
+ "pages": 4,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192
  "shards": 1
193
  },
194
  {
195
+ "seconds": 1.576328019000016,
196
+ "bytes": 582303,
197
+ "pages": 17,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
198
  "shards": 1
199
  },
200
  {
201
+ "seconds": 1.4797563120000632,
202
+ "bytes": 958250,
203
+ "pages": 27,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
204
  "shards": 1
205
  },
206
  {
207
+ "seconds": 1.1572930810000344,
208
+ "bytes": 7429293,
209
+ "pages": 34,
 
 
 
 
 
 
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.7505123790000425,
214
+ "bytes": 15799691,
215
+ "pages": 34,
 
 
 
 
 
 
216
  "shards": 4
217
  },
218
  {
219
+ "seconds": 1.5947456709999415,
220
+ "bytes": 2710887,
 
 
 
 
 
 
221
  "pages": 37,
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 2.055465744000003,
226
+ "bytes": 6120157,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
227
  "pages": 35,
 
 
 
 
 
 
228
  "shards": 2
229
  },
230
  {
231
+ "seconds": 1.5483520990001125,
232
+ "bytes": 929612,
233
+ "pages": 27,
234
+ "shards": 1
235
  },
236
  {
237
+ "seconds": 1.549194875000012,
238
+ "bytes": 12173751,
239
+ "pages": 40,
240
  "shards": 4
241
  },
242
  {
243
+ "seconds": 1.4142575109999598,
244
+ "bytes": 6051595,
245
+ "pages": 36,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.509009006000042,
250
+ "bytes": 6599366,
 
 
 
 
 
 
 
 
 
 
 
 
251
  "pages": 28,
252
  "shards": 2
253
  },
254
  {
255
+ "seconds": 1.1468410420000055,
256
+ "bytes": 7711330,
257
+ "pages": 36,
258
  "shards": 3
259
  },
260
  {
261
+ "seconds": 1.7056902470001205,
262
+ "bytes": 10320081,
263
+ "pages": 32,
264
  "shards": 3
265
  },
266
  {
267
+ "seconds": 1.3121002539999154,
268
+ "bytes": 2619780,
269
+ "pages": 36,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 1.2438676930000838,
274
+ "bytes": 6205549,
275
+ "pages": 34,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 1.3983366309998928,
280
+ "bytes": 1085324,
281
+ "pages": 32,
 
 
 
 
 
 
 
 
 
 
 
 
282
  "shards": 1
283
  },
284
  {
285
+ "seconds": 1.6373932590004188,
286
+ "bytes": 17208473,
287
+ "pages": 36,
288
+ "shards": 5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
289
  },
290
  {
291
+ "seconds": 1.523682868999913,
292
+ "bytes": 6058267,
293
+ "pages": 36,
294
  "shards": 2
295
  },
296
  {
297
+ "seconds": 1.493248520999714,
298
+ "bytes": 2669063,
299
+ "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
300
  "shards": 2
301
  },
302
  {
303
+ "seconds": 1.4799904070000593,
304
+ "bytes": 10661870,
305
+ "pages": 35,
306
  "shards": 3
307
  }
308
  ],
 
348
  },
349
  "cache_reused": true,
350
  "resume_probe_seconds": [
351
+ 2.576500048000014,
352
+ 0.012398547999964649
353
  ]
354
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a2b248de903e43cc04b64f31262641a2a596e6720f6e8b83622de3cb8a1ad33a
3
- size 721782015
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87fc13206a6115414dc5698ba185a42d91483c871e4ad1d988dd31693f1ea24e
3
+ size 721767423
training_config.json CHANGED
@@ -53,7 +53,7 @@
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
- "prefetch_batches": 63,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 62,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,