appvoid commited on
Commit
414bd2b
·
verified ·
1 Parent(s): e293a40

Qwen90M step 825000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +50 -50
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +243 -243
  6. trainer.pt +2 -2
  7. training_config.json +1 -1
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 820000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "be65b8dd85d32e3c046a0a29d18dd9c9636bd3518c6180704165ccc3b345f97a",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "72a6fa391959ac41be57caac41e5d9e33fa242f668e7e959e3b49726861e8403",
12
- "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
  "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7",
14
- "shard_scheduler.json": "cca80237086820dcf5c494e00db56e92fb63f5bff088142a981f0d3db418bb87",
15
- "coverage.json": "c13407e1cced45349ea4cc9f52ed38e019bd1fa9d3e900afc3652ad06d436a2c",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
 
1
  {
2
+ "step": 825000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "c94a248685b839a2edfdd0022e869649e1d13c7266eb40bee2cd5b65ffe131e7",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "de221e9ed21e83c30b24525b864378be67505342cacb8dfb75a9ebfcd0672c2b",
12
+ "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58",
13
  "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7",
14
+ "shard_scheduler.json": "3a5912414de65c80e1ba8c4b974f6204935ab0245d673f1014baff16d50c52ea",
15
+ "coverage.json": "384700a0a13bd6ae2e77220c1d1c48f9e386f807b0b80377087bdf716560b06c",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 820000,
3
  "stats": {
4
- "climbmix/blocks": 29520000,
5
- "climbmix/targets": 49348064056,
6
- "climbmix/input_tokens": 49351191260,
7
- "ultra_qa/blocks": 11070000,
8
- "ultra_qa/targets": 18059605638,
9
- "ultra_qa/input_tokens": 18059640442,
10
- "cortex/blocks": 11070000,
11
- "cortex/targets": 420935370,
12
- "cortex/input_tokens": 17596608233,
13
- "ultra_style/blocks": 11070000,
14
- "ultra_style/targets": 17509519663,
15
- "ultra_style/input_tokens": 17510438016,
16
- "openmath/blocks": 7380000,
17
- "openmath/targets": 11403359242,
18
- "openmath/input_tokens": 11407158801,
19
- "rewrite6/blocks": 3690000,
20
- "rewrite6/targets": 6049287294,
21
- "rewrite6/input_tokens": 6055919347
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 4042,
26
- "row": 3493,
27
  "start": {
28
- "file": 195,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 54,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 16559525
38
  },
39
  "rewrite6": {
40
- "page": 50,
41
- "row": 410,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 204800
47
  }
48
  },
49
- "epoch": 45,
50
  "offset": 0,
51
- "documents": 9448390
52
  },
53
  "openmath": {
54
- "page": 2526,
55
- "row": 3013,
56
  "start": {
57
  "file": 23,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 303,
61
- "row": 546
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 10349509
67
  },
68
  "ultra_qa": {
69
- "page": 1092,
70
- "row": 2051,
71
  "start": {
72
  "file": 8,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
- "row_group": 4,
76
- "row": 57600
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 4474883
82
  },
83
  "ultra_style": {
84
- "page": 1667,
85
- "row": 1796,
86
  "start": {
87
- "file": 9,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 7,
91
- "row": 68335
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 2048,
96
- "documents": 6829828
97
  },
98
  "cortex": {
99
- "page": 214689,
100
- "row": 86,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 27480278
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 825000,
3
  "stats": {
4
+ "climbmix/blocks": 29700000,
5
+ "climbmix/targets": 49648833395,
6
+ "climbmix/input_tokens": 49651979405,
7
+ "ultra_qa/blocks": 11137500,
8
+ "ultra_qa/targets": 18169783675,
9
+ "ultra_qa/input_tokens": 18169818678,
10
+ "cortex/blocks": 11137500,
11
+ "cortex/targets": 423506385,
12
+ "cortex/input_tokens": 17703865080,
13
+ "ultra_style/blocks": 11137500,
14
+ "ultra_style/targets": 17616349514,
15
+ "ultra_style/input_tokens": 17617273526,
16
+ "openmath/blocks": 7425000,
17
+ "openmath/targets": 11472953325,
18
+ "openmath/input_tokens": 11476776268,
19
+ "rewrite6/blocks": 3712500,
20
+ "rewrite6/targets": 6086134136,
21
+ "rewrite6/input_tokens": 6092806682
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 4067,
26
+ "row": 1808,
27
  "start": {
28
+ "file": 196,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 71,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 4096,
37
+ "documents": 16660240
38
  },
39
  "rewrite6": {
40
+ "page": 14,
41
+ "row": 93,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 57344
47
  }
48
  },
49
+ "epoch": 46,
50
  "offset": 0,
51
+ "documents": 9506021
52
  },
53
  "openmath": {
54
+ "page": 2542,
55
+ "row": 913,
56
  "start": {
57
  "file": 23,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 369,
61
+ "row": 82
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 10412945
67
  },
68
  "ultra_qa": {
69
+ "page": 1099,
70
+ "row": 700,
71
  "start": {
72
  "file": 8,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
+ "row_group": 5,
76
+ "row": 21781
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 4502204
82
  },
83
  "ultra_style": {
84
+ "page": 1677,
85
+ "row": 2616,
86
  "start": {
87
+ "file": 10,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 0,
91
+ "row": 19577
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 0,
96
+ "documents": 6871608
97
  },
98
  "cortex": {
99
+ "page": 215998,
100
+ "row": 28,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 27647772
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 820000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "be65b8dd85d32e3c046a0a29d18dd9c9636bd3518c6180704165ccc3b345f97a", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "72a6fa391959ac41be57caac41e5d9e33fa242f668e7e959e3b49726861e8403", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7", "shard_scheduler.json": "cca80237086820dcf5c494e00db56e92fb63f5bff088142a981f0d3db418bb87", "coverage.json": "c13407e1cced45349ea4cc9f52ed38e019bd1fa9d3e900afc3652ad06d436a2c", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000820000"}
 
1
+ {"step": 825000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "c94a248685b839a2edfdd0022e869649e1d13c7266eb40bee2cd5b65ffe131e7", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "de221e9ed21e83c30b24525b864378be67505342cacb8dfb75a9ebfcd0672c2b", "training_config.json": "0ab64376a29947d9a9f670c0ece06c1900ab6f9ec82482957af5fc2c70beec58", "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7", "shard_scheduler.json": "3a5912414de65c80e1ba8c4b974f6204935ab0245d673f1014baff16d50c52ea", "coverage.json": "384700a0a13bd6ae2e77220c1d1c48f9e386f807b0b80377087bdf716560b06c", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000825000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:be65b8dd85d32e3c046a0a29d18dd9c9636bd3518c6180704165ccc3b345f97a
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c94a248685b839a2edfdd0022e869649e1d13c7266eb40bee2cd5b65ffe131e7
3
  size 360622744
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 76942.25922787549,
26
- "required_pages_per_second_upper_estimate": 0.005323044024022557
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": 9617.782403484436,
40
- "required_pages_per_second_upper_estimate": 0.004026764650203828
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": 19235.564806968872,
54
- "required_pages_per_second_upper_estimate": 0.004006260873334826
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": 28853.347210453307,
68
- "required_pages_per_second_upper_estimate": 0.00172971410875644
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": 28853.347210453307,
82
- "required_pages_per_second_upper_estimate": 0.0022432208331508883
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": 28853.347210453307,
96
- "required_pages_per_second_upper_estimate": 0.37009975770517706
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,413 +155,413 @@
155
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
156
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
157
  "selected": {
158
- "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.08518336581272479,
162
- "producer_p95_seconds": 0.2713440530023945,
163
- "gpu_update_seconds": 0.958225047455963,
164
- "producer_batches_per_second": 11.739381162731878,
165
- "gpu_batches_per_second": 1.0435961809336411,
166
- "producer_headroom": 11.248969071762392,
167
  "sustainable": true,
168
- "buffer_seconds": 60.36817798972567,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 78,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 237498.7448770881,
174
- "upload_bytes_per_second": 3973102.9977444257,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 50.96170196279986,
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": 1.6904037299991614,
184
- "bytes": 12097177,
185
  "pages": 34,
 
 
 
 
 
 
186
  "shards": 3
187
  },
188
  {
189
- "seconds": 1.2079874939990987,
190
- "bytes": 1225243,
191
- "pages": 35,
192
- "shards": 1
193
  },
194
  {
195
- "seconds": 1.7503577689985832,
196
- "bytes": 11722749,
197
- "pages": 33,
198
- "shards": 4
199
  },
200
  {
201
- "seconds": 1.717468271999678,
202
- "bytes": 6201366,
203
  "pages": 37,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 1.4180366079999658,
208
- "bytes": 818969,
209
- "pages": 25,
210
- "shards": 1
211
  },
212
  {
213
- "seconds": 1.935665021999739,
214
- "bytes": 7829235,
215
- "pages": 39,
216
  "shards": 3
217
  },
218
  {
219
- "seconds": 1.7852341269990575,
220
- "bytes": 11660999,
221
- "pages": 37,
222
- "shards": 3
223
  },
224
  {
225
- "seconds": 1.8856627909990493,
226
- "bytes": 7118698,
227
- "pages": 30,
228
- "shards": 3
 
 
 
 
 
 
 
 
 
 
 
 
229
  },
230
  {
231
- "seconds": 1.6447518009990745,
232
- "bytes": 6129393,
233
  "pages": 37,
 
 
 
 
 
 
234
  "shards": 2
235
  },
236
  {
237
- "seconds": 1.743864690999544,
238
- "bytes": 1054003,
239
- "pages": 35,
240
  "shards": 1
241
  },
242
  {
243
- "seconds": 1.5393589430004795,
244
- "bytes": 11424070,
245
- "pages": 35,
246
  "shards": 4
247
  },
248
  {
249
- "seconds": 1.9194490070003667,
250
- "bytes": 6288888,
251
- "pages": 35,
252
  "shards": 2
253
  },
254
  {
255
- "seconds": 1.4686280049991183,
256
- "bytes": 8470485,
257
  "pages": 37,
258
- "shards": 4
259
  },
260
  {
261
- "seconds": 1.9930366190001223,
262
- "bytes": 5866239,
263
- "pages": 26,
264
  "shards": 2
265
  },
266
  {
267
- "seconds": 2.011558195001271,
268
- "bytes": 5842602,
269
- "pages": 41,
270
- "shards": 2
271
  },
272
  {
273
- "seconds": 1.6693870859999151,
274
- "bytes": 7612861,
275
  "pages": 29,
276
  "shards": 3
277
  },
278
  {
279
- "seconds": 1.3870315900003334,
280
- "bytes": 1261477,
281
  "pages": 35,
282
  "shards": 1
283
  },
284
  {
285
- "seconds": 1.6420892379992438,
286
- "bytes": 7607765,
287
- "pages": 39,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 1.5253431410001212,
292
- "bytes": 5688723,
293
- "pages": 19,
294
  "shards": 2
295
  },
296
  {
297
- "seconds": 1.593651556999248,
298
- "bytes": 11614889,
299
- "pages": 46,
300
  "shards": 3
301
  },
302
  {
303
- "seconds": 1.818591990000641,
304
- "bytes": 7461958,
305
- "pages": 30,
306
- "shards": 3
307
  },
308
  {
309
- "seconds": 1.4790617950002343,
310
- "bytes": 1048131,
311
- "pages": 34,
312
- "shards": 1
313
  },
314
  {
315
- "seconds": 1.3689376410002296,
316
- "bytes": 5834070,
317
- "pages": 28,
 
 
 
 
 
 
318
  "shards": 2
319
  },
320
  {
321
- "seconds": 1.2757619459989655,
322
- "bytes": 12484592,
323
- "pages": 45,
324
  "shards": 4
325
  },
326
  {
327
- "seconds": 1.4620103640008892,
328
- "bytes": 316633,
329
- "pages": 8,
330
- "shards": 1
331
  },
332
  {
333
- "seconds": 1.8397933680007554,
334
- "bytes": 6758434,
335
  "pages": 29,
336
  "shards": 2
337
  },
338
  {
339
- "seconds": 1.9822834630012949,
340
- "bytes": 7671850,
341
- "pages": 38,
342
  "shards": 3
343
  },
344
  {
345
- "seconds": 1.43599132899908,
346
- "bytes": 5917528,
347
- "pages": 29,
348
- "shards": 2
349
  },
350
  {
351
- "seconds": 1.7868269310019969,
352
- "bytes": 5421842,
353
- "pages": 40,
354
- "shards": 2
355
  },
356
  {
357
- "seconds": 1.8639257160029956,
358
- "bytes": 7235620,
359
- "pages": 29,
 
 
 
 
 
 
360
  "shards": 3
361
  },
362
  {
363
- "seconds": 1.8414086459997634,
364
- "bytes": 5860537,
365
  "pages": 36,
366
  "shards": 2
367
  },
368
  {
369
- "seconds": 1.8819817990006413,
370
- "bytes": 6541422,
371
- "pages": 23,
372
  "shards": 2
373
  },
374
  {
375
- "seconds": 1.735375592001219,
376
- "bytes": 12429197,
377
- "pages": 48,
378
- "shards": 4
379
  },
380
  {
381
- "seconds": 1.311772810000548,
382
- "bytes": 931562,
383
- "pages": 26,
384
- "shards": 1
385
  },
386
  {
387
- "seconds": 1.873525888997392,
388
- "bytes": 7538765,
389
- "pages": 38,
390
- "shards": 3
391
  },
392
  {
393
- "seconds": 1.4924008759990102,
394
- "bytes": 6022523,
395
- "pages": 32,
396
  "shards": 2
397
  },
398
  {
399
- "seconds": 1.568882242001564,
400
- "bytes": 7013425,
401
  "pages": 40,
402
  "shards": 3
403
  },
404
  {
405
- "seconds": 1.719156404997193,
406
- "bytes": 11941861,
407
- "pages": 32,
408
- "shards": 3
409
  },
410
  {
411
- "seconds": 1.5756286089999776,
412
- "bytes": 6163874,
413
  "pages": 36,
414
  "shards": 2
415
  },
416
  {
417
- "seconds": 1.85335265699905,
418
- "bytes": 5996093,
419
  "pages": 28,
420
- "shards": 3
421
- },
422
- {
423
- "seconds": 1.6912986729985278,
424
- "bytes": 6223259,
425
- "pages": 38,
426
- "shards": 2
427
- },
428
- {
429
- "seconds": 1.3231591360017774,
430
- "bytes": 1106930,
431
- "pages": 34,
432
  "shards": 1
433
  },
434
  {
435
- "seconds": 2.400186033999489,
436
- "bytes": 7382284,
437
- "pages": 30,
438
- "shards": 3
439
- },
440
- {
441
- "seconds": 2.1097241989991744,
442
- "bytes": 16842562,
443
  "pages": 44,
444
  "shards": 5
445
  },
446
  {
447
- "seconds": 1.6551350050031033,
448
- "bytes": 2311821,
449
  "pages": 28,
450
  "shards": 2
451
  },
452
  {
453
- "seconds": 1.769397926000238,
454
- "bytes": 6044362,
455
  "pages": 37,
456
  "shards": 2
457
  },
458
  {
459
- "seconds": 1.4953640580024512,
460
- "bytes": 6379384,
461
- "pages": 37,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.7474238050017448,
466
- "bytes": 6290249,
467
- "pages": 31,
468
- "shards": 3
469
- },
470
- {
471
- "seconds": 1.5282406320002337,
472
- "bytes": 5939964,
473
- "pages": 36,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.2798854280008527,
478
- "bytes": 545338,
479
- "pages": 18,
480
- "shards": 1
481
  },
482
  {
483
- "seconds": 1.8976086379989283,
484
- "bytes": 13629725,
485
- "pages": 49,
486
- "shards": 4
487
  },
488
  {
489
- "seconds": 1.6741066590002447,
490
- "bytes": 5692006,
491
- "pages": 27,
492
  "shards": 2
493
  },
494
  {
495
- "seconds": 1.5403353149995382,
496
- "bytes": 7551970,
497
- "pages": 37,
498
  "shards": 3
499
  },
500
  {
501
- "seconds": 1.5439765690025524,
502
- "bytes": 6266973,
503
  "pages": 36,
504
- "shards": 2
505
- },
506
- {
507
- "seconds": 1.459664652000356,
508
- "bytes": 1024497,
509
- "pages": 33,
510
  "shards": 1
511
  },
512
  {
513
- "seconds": 1.8518995710001036,
514
- "bytes": 10973445,
515
- "pages": 22,
516
- "shards": 4
517
- },
518
- {
519
- "seconds": 1.523874344999058,
520
- "bytes": 6793396,
521
- "pages": 30,
522
- "shards": 2
523
  },
524
  {
525
- "seconds": 1.551560215997597,
526
- "bytes": 6347247,
527
- "pages": 37,
528
  "shards": 2
529
  },
530
  {
531
- "seconds": 1.3656726940025692,
532
- "bytes": 2327821,
533
- "pages": 28,
534
- "shards": 2
535
  },
536
  {
537
- "seconds": 1.6191135740009486,
538
- "bytes": 10591273,
539
- "pages": 42,
540
  "shards": 3
541
  },
542
  {
543
- "seconds": 1.5766091279983812,
544
- "bytes": 7345474,
545
- "pages": 29,
546
- "shards": 3
547
  },
548
  {
549
- "seconds": 1.195227366002655,
550
- "bytes": 1189304,
551
- "pages": 35,
552
- "shards": 1
553
  },
554
  {
555
- "seconds": 2.000728038001398,
556
- "bytes": 11892571,
557
- "pages": 34,
558
  "shards": 3
559
  },
560
  {
561
- "seconds": 1.4507886779974797,
562
- "bytes": 2119215,
563
- "pages": 21,
564
- "shards": 2
565
  }
566
  ],
567
  "cache_key": {
 
22
  "tokens_per_second": 598315.063954628,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 75686.52414391923,
26
+ "required_pages_per_second_upper_estimate": 0.005236169357207637
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": 9460.815517989904,
40
+ "required_pages_per_second_upper_estimate": 0.0039610458930885726
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": 18921.631035979808,
54
+ "required_pages_per_second_upper_estimate": 0.003940876747828083
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": 28382.44655396971,
68
+ "required_pages_per_second_upper_estimate": 0.0017014843334239179
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": 28382.44655396971,
82
+ "required_pages_per_second_upper_estimate": 0.0022066103783823774
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": 28382.44655396971,
96
+ "required_pages_per_second_upper_estimate": 0.36405954969753734
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": 0.3011315108518602,
162
+ "producer_p95_seconds": 0.2949342350002553,
163
+ "gpu_update_seconds": 0.9741232119446381,
164
+ "producer_batches_per_second": 3.320808231497048,
165
+ "gpu_batches_per_second": 1.0265641838096684,
166
+ "producer_headroom": 3.2348763807180982,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.39563914056757,
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": 533735.996489511,
174
+ "upload_bytes_per_second": 3295611.2481271825,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 114.52732010854719,
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": 1.3231591360017774,
184
+ "bytes": 1106930,
185
  "pages": 34,
186
+ "shards": 1
187
+ },
188
+ {
189
+ "seconds": 2.400186033999489,
190
+ "bytes": 7382284,
191
+ "pages": 30,
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 2.1097241989991744,
196
+ "bytes": 16842562,
197
+ "pages": 44,
198
+ "shards": 5
199
  },
200
  {
201
+ "seconds": 1.6551350050031033,
202
+ "bytes": 2311821,
203
+ "pages": 28,
204
+ "shards": 2
205
  },
206
  {
207
+ "seconds": 1.769397926000238,
208
+ "bytes": 6044362,
209
  "pages": 37,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.4953640580024512,
214
+ "bytes": 6379384,
215
+ "pages": 37,
216
+ "shards": 2
217
  },
218
  {
219
+ "seconds": 1.7474238050017448,
220
+ "bytes": 6290249,
221
+ "pages": 31,
222
  "shards": 3
223
  },
224
  {
225
+ "seconds": 1.5282406320002337,
226
+ "bytes": 5939964,
227
+ "pages": 36,
228
+ "shards": 2
229
  },
230
  {
231
+ "seconds": 1.2798854280008527,
232
+ "bytes": 545338,
233
+ "pages": 18,
234
+ "shards": 1
235
+ },
236
+ {
237
+ "seconds": 1.8976086379989283,
238
+ "bytes": 13629725,
239
+ "pages": 49,
240
+ "shards": 4
241
+ },
242
+ {
243
+ "seconds": 1.6741066590002447,
244
+ "bytes": 5692006,
245
+ "pages": 27,
246
+ "shards": 2
247
  },
248
  {
249
+ "seconds": 1.5403353149995382,
250
+ "bytes": 7551970,
251
  "pages": 37,
252
+ "shards": 3
253
+ },
254
+ {
255
+ "seconds": 1.5439765690025524,
256
+ "bytes": 6266973,
257
+ "pages": 36,
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 1.459664652000356,
262
+ "bytes": 1024497,
263
+ "pages": 33,
264
  "shards": 1
265
  },
266
  {
267
+ "seconds": 1.8518995710001036,
268
+ "bytes": 10973445,
269
+ "pages": 22,
270
  "shards": 4
271
  },
272
  {
273
+ "seconds": 1.523874344999058,
274
+ "bytes": 6793396,
275
+ "pages": 30,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 1.551560215997597,
280
+ "bytes": 6347247,
281
  "pages": 37,
282
+ "shards": 2
283
  },
284
  {
285
+ "seconds": 1.3656726940025692,
286
+ "bytes": 2327821,
287
+ "pages": 28,
288
  "shards": 2
289
  },
290
  {
291
+ "seconds": 1.6191135740009486,
292
+ "bytes": 10591273,
293
+ "pages": 42,
294
+ "shards": 3
295
  },
296
  {
297
+ "seconds": 1.5766091279983812,
298
+ "bytes": 7345474,
299
  "pages": 29,
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 1.195227366002655,
304
+ "bytes": 1189304,
305
  "pages": 35,
306
  "shards": 1
307
  },
308
  {
309
+ "seconds": 2.000728038001398,
310
+ "bytes": 11892571,
311
+ "pages": 34,
312
  "shards": 3
313
  },
314
  {
315
+ "seconds": 1.4507886779974797,
316
+ "bytes": 2119215,
317
+ "pages": 21,
318
  "shards": 2
319
  },
320
  {
321
+ "seconds": 1.5652879460030817,
322
+ "bytes": 11222900,
323
+ "pages": 47,
324
  "shards": 3
325
  },
326
  {
327
+ "seconds": 1.6341935630007356,
328
+ "bytes": 5648901,
329
+ "pages": 19,
330
+ "shards": 2
331
  },
332
  {
333
+ "seconds": 1.5995988550021139,
334
+ "bytes": 2515627,
335
+ "pages": 36,
336
+ "shards": 2
337
  },
338
  {
339
+ "seconds": 1.3335332679998828,
340
+ "bytes": 5896707,
341
+ "pages": 29,
342
+ "shards": 2
343
+ },
344
+ {
345
+ "seconds": 1.9101363129993842,
346
+ "bytes": 4340262,
347
+ "pages": 24,
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 2.043565675998252,
352
+ "bytes": 13514454,
353
+ "pages": 41,
354
  "shards": 4
355
  },
356
  {
357
+ "seconds": 1.75038528100049,
358
+ "bytes": 5975710,
359
+ "pages": 36,
360
+ "shards": 2
361
  },
362
  {
363
+ "seconds": 2.091957274999004,
364
+ "bytes": 2323345,
365
  "pages": 29,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.9117556889978005,
370
+ "bytes": 10784078,
371
+ "pages": 43,
372
  "shards": 3
373
  },
374
  {
375
+ "seconds": 1.6485553290003736,
376
+ "bytes": 917525,
377
+ "pages": 26,
378
+ "shards": 1
379
  },
380
  {
381
+ "seconds": 2.0703852710030333,
382
+ "bytes": 13336105,
383
+ "pages": 42,
384
+ "shards": 4
385
  },
386
  {
387
+ "seconds": 1.4275249089987483,
388
+ "bytes": 5696136,
389
+ "pages": 26,
390
+ "shards": 3
391
+ },
392
+ {
393
+ "seconds": 1.6552812369991443,
394
+ "bytes": 6589515,
395
+ "pages": 36,
396
  "shards": 3
397
  },
398
  {
399
+ "seconds": 1.7584444239982986,
400
+ "bytes": 6116166,
401
  "pages": 36,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 1.7439238019978802,
406
+ "bytes": 6171251,
407
+ "pages": 36,
408
  "shards": 2
409
  },
410
  {
411
+ "seconds": 1.5497890789993107,
412
+ "bytes": 2026869,
413
+ "pages": 23,
414
+ "shards": 2
415
  },
416
  {
417
+ "seconds": 2.0294503110017104,
418
+ "bytes": 16328104,
419
+ "pages": 41,
420
+ "shards": 4
421
  },
422
  {
423
+ "seconds": 1.913415155999246,
424
+ "bytes": 2704493,
425
+ "pages": 37,
426
+ "shards": 2
427
  },
428
  {
429
+ "seconds": 1.6119605509993562,
430
+ "bytes": 6041654,
431
+ "pages": 28,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.6828175019982154,
436
+ "bytes": 10664698,
437
  "pages": 40,
438
  "shards": 3
439
  },
440
  {
441
+ "seconds": 1.565426788998593,
442
+ "bytes": 2415576,
443
+ "pages": 30,
444
+ "shards": 2
445
  },
446
  {
447
+ "seconds": 1.5388405060002697,
448
+ "bytes": 6065532,
449
  "pages": 36,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.3961634170009347,
454
+ "bytes": 973137,
455
  "pages": 28,
 
 
 
 
 
 
 
 
 
 
 
 
456
  "shards": 1
457
  },
458
  {
459
+ "seconds": 2.0744263130000036,
460
+ "bytes": 17948904,
 
 
 
 
 
 
461
  "pages": 44,
462
  "shards": 5
463
  },
464
  {
465
+ "seconds": 1.631489851999504,
466
+ "bytes": 5794503,
467
  "pages": 28,
468
  "shards": 2
469
  },
470
  {
471
+ "seconds": 1.5992371809988981,
472
+ "bytes": 2589704,
473
  "pages": 37,
474
  "shards": 2
475
  },
476
  {
477
+ "seconds": 1.4273169570005848,
478
+ "bytes": 6151606,
479
+ "pages": 36,
480
  "shards": 2
481
  },
482
  {
483
+ "seconds": 1.4220550049976737,
484
+ "bytes": 4624242,
485
+ "pages": 30,
 
 
 
 
 
 
486
  "shards": 2
487
  },
488
  {
489
+ "seconds": 2.047509969001112,
490
+ "bytes": 7597965,
491
+ "pages": 31,
492
+ "shards": 3
493
  },
494
  {
495
+ "seconds": 1.836660891000065,
496
+ "bytes": 12233872,
497
+ "pages": 41,
498
+ "shards": 3
499
  },
500
  {
501
+ "seconds": 1.7772512390001793,
502
+ "bytes": 2617112,
503
+ "pages": 37,
504
  "shards": 2
505
  },
506
  {
507
+ "seconds": 1.5108418660020106,
508
+ "bytes": 11046966,
509
+ "pages": 31,
510
  "shards": 3
511
  },
512
  {
513
+ "seconds": 1.4716807509976206,
514
+ "bytes": 1094934,
515
  "pages": 36,
 
 
 
 
 
 
516
  "shards": 1
517
  },
518
  {
519
+ "seconds": 1.608142429999134,
520
+ "bytes": 7266564,
521
+ "pages": 31,
522
+ "shards": 3
 
 
 
 
 
 
523
  },
524
  {
525
+ "seconds": 1.734035388999473,
526
+ "bytes": 6027691,
527
+ "pages": 29,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.5976459840021562,
532
+ "bytes": 11304870,
533
+ "pages": 44,
534
+ "shards": 3
535
  },
536
  {
537
+ "seconds": 1.606354307001311,
538
+ "bytes": 7330991,
539
+ "pages": 32,
540
  "shards": 3
541
  },
542
  {
543
+ "seconds": 1.5559647560003214,
544
+ "bytes": 6263907,
545
+ "pages": 37,
546
+ "shards": 2
547
  },
548
  {
549
+ "seconds": 1.8874185050008236,
550
+ "bytes": 2469051,
551
+ "pages": 30,
552
+ "shards": 2
553
  },
554
  {
555
+ "seconds": 24.49332847400001,
556
+ "bytes": 9957694,
557
+ "pages": 37,
558
  "shards": 3
559
  },
560
  {
561
+ "seconds": 1.9510848309982975,
562
+ "bytes": 11992256,
563
+ "pages": 40,
564
+ "shards": 3
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:72a6fa391959ac41be57caac41e5d9e33fa242f668e7e959e3b49726861e8403
3
- size 721780159
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de221e9ed21e83c30b24525b864378be67505342cacb8dfb75a9ebfcd0672c2b
3
+ size 721770943
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,