appvoid commited on
Commit
722b382
·
verified ·
1 Parent(s): 96d732d

Qwen90M step 940000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +46 -46
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +264 -162
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 935000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "32c5ec46c4d26de24f43e5e5ac475183bd0d5a9168b597e1a4995f06cda64d9b",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "c55cec8e778846c453320337a4022807364361a6bb535fd24e51ca5d5f695ab5",
12
- "training_config.json": "aafacc05aaf9ea874dcfa3d02457799efaf5525cf7b211a4208541f6e989efee",
13
  "performance.json": "a7562f2130d78c752d5494512bcbf6d4c62627eee4ff6943c08f7615044b2de4",
14
- "shard_scheduler.json": "9525e5f4fb5139b10175901a17eb2c1ac2df509c25c3edcfbc3108aaabd5b46e",
15
- "coverage.json": "ea1e9229096e365e0bf25587fc7c00524d8f83b2754ab83b1c551cc02c49d653",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
 
1
  {
2
+ "step": 940000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "d09a6f6d15b58b397b5cb80bf0442935d3486b863565dc4a3b414d4d29102f73",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "379eeb0d627e11d6aeb818be90d2beb7ad27d2bcf192919683768be57619bc1b",
12
+ "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
  "performance.json": "a7562f2130d78c752d5494512bcbf6d4c62627eee4ff6943c08f7615044b2de4",
14
+ "shard_scheduler.json": "e7f0f1e8787fc3b9f72fcf8f200650769ecf8ab43afd791855595c601ac93a27",
15
+ "coverage.json": "3d6b4fdad9262d21f1491946476e23ba4ec13ac73d7d5a5a0af7a9f9856d8a06",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 935000,
3
  "stats": {
4
- "climbmix/blocks": 33660000,
5
- "climbmix/targets": 56269025726,
6
- "climbmix/input_tokens": 56272591695,
7
- "ultra_qa/blocks": 12622500,
8
- "ultra_qa/targets": 20592546150,
9
- "ultra_qa/input_tokens": 20592585687,
10
- "cortex/blocks": 12622500,
11
- "cortex/targets": 479968870,
12
- "cortex/input_tokens": 20064566894,
13
- "ultra_style/blocks": 12622500,
14
- "ultra_style/targets": 19964286118,
15
- "ultra_style/input_tokens": 19965332428,
16
- "openmath/blocks": 8415000,
17
- "openmath/targets": 13002775878,
18
- "openmath/input_tokens": 13007108390,
19
- "rewrite6/blocks": 4207500,
20
- "rewrite6/targets": 6897626953,
21
- "rewrite6/input_tokens": 6905189624
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 4609,
26
- "row": 3963,
27
  "start": {
28
- "file": 223,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 9,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 20480,
37
- "documents": 18882427
38
  },
39
  "rewrite6": {
40
- "page": 22,
41
- "row": 2895,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 90112
47
  }
48
  },
49
  "epoch": 52,
50
  "offset": 0,
51
- "documents": 10774015
52
  },
53
  "openmath": {
54
- "page": 2881,
55
- "row": 724,
56
  "start": {
57
  "file": 27,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 11,
61
- "row": 30
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 11801300
67
  },
68
  "ultra_qa": {
69
- "page": 1245,
70
- "row": 2808,
71
  "start": {
72
  "file": 9,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
- "row": 35423
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 5102328
82
  },
83
  "ultra_style": {
84
- "page": 1901,
85
- "row": 453,
86
  "start": {
87
  "file": 11,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 2,
91
- "row": 81501
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 7786949
97
  },
98
  "cortex": {
99
- "page": 244798,
100
- "row": 127,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 31334271
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 940000,
3
  "stats": {
4
+ "climbmix/blocks": 33840000,
5
+ "climbmix/targets": 56570146487,
6
+ "climbmix/input_tokens": 56573731401,
7
+ "ultra_qa/blocks": 12690000,
8
+ "ultra_qa/targets": 20702763319,
9
+ "ultra_qa/input_tokens": 20702803093,
10
+ "cortex/blocks": 12690000,
11
+ "cortex/targets": 482534935,
12
+ "cortex/input_tokens": 20171789678,
13
+ "ultra_style/blocks": 12690000,
14
+ "ultra_style/targets": 20070737053,
15
+ "ultra_style/input_tokens": 20071789036,
16
+ "openmath/blocks": 8460000,
17
+ "openmath/targets": 13072208273,
18
+ "openmath/input_tokens": 13076563771,
19
+ "rewrite6/blocks": 4230000,
20
+ "rewrite6/targets": 6934478168,
21
+ "rewrite6/input_tokens": 6942080796
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 4634,
26
+ "row": 2400,
27
  "start": {
28
+ "file": 224,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 26,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 18983264
38
  },
39
  "rewrite6": {
40
+ "page": 36,
41
+ "row": 2703,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 147456
47
  }
48
  },
49
  "epoch": 52,
50
  "offset": 0,
51
+ "documents": 10831167
52
  },
53
  "openmath": {
54
+ "page": 2896,
55
+ "row": 2245,
56
  "start": {
57
  "file": 27,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 72,
61
+ "row": 470
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 11864261
67
  },
68
  "ultra_qa": {
69
+ "page": 1252,
70
+ "row": 1383,
71
  "start": {
72
  "file": 9,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
+ "row": 64095
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 5129575
82
  },
83
  "ultra_style": {
84
+ "page": 1911,
85
+ "row": 1420,
86
  "start": {
87
  "file": 11,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 3,
91
+ "row": 37668
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 7828876
97
  },
98
  "cortex": {
99
+ "page": 246106,
100
+ "row": 57,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 31501625
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 935000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "32c5ec46c4d26de24f43e5e5ac475183bd0d5a9168b597e1a4995f06cda64d9b", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "c55cec8e778846c453320337a4022807364361a6bb535fd24e51ca5d5f695ab5", "training_config.json": "aafacc05aaf9ea874dcfa3d02457799efaf5525cf7b211a4208541f6e989efee", "performance.json": "a7562f2130d78c752d5494512bcbf6d4c62627eee4ff6943c08f7615044b2de4", "shard_scheduler.json": "9525e5f4fb5139b10175901a17eb2c1ac2df509c25c3edcfbc3108aaabd5b46e", "coverage.json": "ea1e9229096e365e0bf25587fc7c00524d8f83b2754ab83b1c551cc02c49d653", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000935000"}
 
1
+ {"step": 940000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "d09a6f6d15b58b397b5cb80bf0442935d3486b863565dc4a3b414d4d29102f73", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "379eeb0d627e11d6aeb818be90d2beb7ad27d2bcf192919683768be57619bc1b", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "a7562f2130d78c752d5494512bcbf6d4c62627eee4ff6943c08f7615044b2de4", "shard_scheduler.json": "e7f0f1e8787fc3b9f72fcf8f200650769ecf8ab43afd791855595c601ac93a27", "coverage.json": "3d6b4fdad9262d21f1491946476e23ba4ec13ac73d7d5a5a0af7a9f9856d8a06", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000940000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:32c5ec46c4d26de24f43e5e5ac475183bd0d5a9168b597e1a4995f06cda64d9b
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d09a6f6d15b58b397b5cb80bf0442935d3486b863565dc4a3b414d4d29102f73
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": 77571.26574998416,
26
- "required_pages_per_second_upper_estimate": 0.005366560154718245
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": 9696.40821874802,
40
- "required_pages_per_second_upper_estimate": 0.004059683637160962
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": 19392.81643749604,
54
- "required_pages_per_second_upper_estimate": 0.0040390122409692635
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": 29089.22465624406,
68
- "required_pages_per_second_upper_estimate": 0.0017438546014675901
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": 29089.22465624406,
82
- "required_pages_per_second_upper_estimate": 0.0022615592670458835
83
  },
84
  {
85
  "source": "cortex",
@@ -92,8 +92,8 @@
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": 29089.22465624406,
96
- "required_pages_per_second_upper_estimate": 0.37312534031431177
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,168 +155,30 @@
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": 64,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.05626329523441598,
162
- "producer_p95_seconds": 0.1819149599996308,
163
- "gpu_update_seconds": 0.9504550336670903,
164
- "producer_batches_per_second": 17.773576820084738,
165
- "gpu_batches_per_second": 1.052127627902346,
166
- "producer_headroom": 16.892985554918255,
167
  "sustainable": true,
168
- "buffer_seconds": 60.82912215469378,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 88,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 324038.0284101658,
174
- "upload_bytes_per_second": 4252787.429657175,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 69.53101767758112,
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": 1.8764951709999877,
184
- "bytes": 52123,
185
- "pages": 1,
186
- "shards": 1
187
- },
188
- {
189
- "seconds": 1.2044020090006597,
190
- "bytes": 7888698,
191
- "pages": 37,
192
- "shards": 3
193
- },
194
- {
195
- "seconds": 1.2931400809993647,
196
- "bytes": 6805061,
197
- "pages": 31,
198
- "shards": 2
199
- },
200
- {
201
- "seconds": 1.581821284000398,
202
- "bytes": 6273764,
203
- "pages": 36,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 0.9842661719994794,
208
- "bytes": 2388376,
209
- "pages": 30,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 4.260054792999654,
214
- "bytes": 10427634,
215
- "pages": 39,
216
- "shards": 3
217
- },
218
- {
219
- "seconds": 1.4442174710002291,
220
- "bytes": 7300139,
221
- "pages": 36,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 1.0513844460001565,
226
- "bytes": 1119612,
227
- "pages": 35,
228
- "shards": 1
229
- },
230
- {
231
- "seconds": 2.561698224999418,
232
- "bytes": 16238211,
233
- "pages": 36,
234
- "shards": 4
235
- },
236
- {
237
- "seconds": 2.406933221000145,
238
- "bytes": 7260996,
239
- "pages": 29,
240
- "shards": 3
241
- },
242
- {
243
- "seconds": 1.1487480899995717,
244
- "bytes": 1204399,
245
- "pages": 35,
246
- "shards": 1
247
- },
248
- {
249
- "seconds": 1.2970412000004217,
250
- "bytes": 7490143,
251
- "pages": 38,
252
- "shards": 3
253
- },
254
- {
255
- "seconds": 4.696426351999435,
256
- "bytes": 5652066,
257
- "pages": 31,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.1291608219999034,
262
- "bytes": 6292028,
263
- "pages": 36,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 1.6897458180001195,
268
- "bytes": 13896820,
269
- "pages": 38,
270
- "shards": 4
271
- },
272
- {
273
- "seconds": 1.1221129010000368,
274
- "bytes": 948347,
275
- "pages": 27,
276
- "shards": 1
277
- },
278
- {
279
- "seconds": 1.7077558099999806,
280
- "bytes": 9679861,
281
- "pages": 34,
282
- "shards": 3
283
- },
284
- {
285
- "seconds": 1.4209248310007752,
286
- "bytes": 7838674,
287
- "pages": 39,
288
- "shards": 3
289
- },
290
- {
291
- "seconds": 1.0323733719997108,
292
- "bytes": 1179407,
293
- "pages": 35,
294
- "shards": 1
295
- },
296
- {
297
- "seconds": 1.2691897260001497,
298
- "bytes": 7296014,
299
- "pages": 30,
300
- "shards": 3
301
- },
302
- {
303
- "seconds": 1.574737728999935,
304
- "bytes": 10770405,
305
- "pages": 43,
306
- "shards": 3
307
- },
308
- {
309
- "seconds": 1.9674206870004127,
310
- "bytes": 5752054,
311
- "pages": 28,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.3296988840002086,
316
- "bytes": 7650823,
317
- "pages": 38,
318
- "shards": 3
319
- },
320
  {
321
  "seconds": 1.2172214809997968,
322
  "bytes": 6191100,
@@ -460,6 +322,246 @@
460
  "bytes": 6201672,
461
  "pages": 37,
462
  "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
463
  }
464
  ],
465
  "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": 77337.79198135664,
26
+ "required_pages_per_second_upper_estimate": 0.005350407897670814
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": 9667.22399766958,
40
+ "required_pages_per_second_upper_estimate": 0.004047464813231256
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": 19334.44799533916,
54
+ "required_pages_per_second_upper_estimate": 0.004026855633747318
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": 29001.67199300874,
68
+ "required_pages_per_second_upper_estimate": 0.0017386059529918088
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": 29001.67199300874,
82
+ "required_pages_per_second_upper_estimate": 0.0022547524325828034
83
  },
84
  {
85
  "source": "cortex",
 
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": 29001.67199300874,
96
+ "required_pages_per_second_upper_estimate": 0.37200230875705464
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": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.2906923827109864,
162
+ "producer_p95_seconds": 0.20526774199970532,
163
+ "gpu_update_seconds": 0.9533243464950897,
164
+ "producer_batches_per_second": 3.440062621091193,
165
+ "gpu_batches_per_second": 1.0489609372471331,
166
+ "producer_headroom": 3.2794954501539464,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.05943382919065,
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": 328751.58465276496,
174
+ "upload_bytes_per_second": 3596231.5696296524,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 70.54243712126933,
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": 1.2172214809997968,
184
  "bytes": 6191100,
 
322
  "bytes": 6201672,
323
  "pages": 37,
324
  "shards": 2
325
+ },
326
+ {
327
+ "seconds": 1.2297464679995755,
328
+ "bytes": 644185,
329
+ "pages": 19,
330
+ "shards": 1
331
+ },
332
+ {
333
+ "seconds": 1.3486207620007917,
334
+ "bytes": 11139901,
335
+ "pages": 32,
336
+ "shards": 4
337
+ },
338
+ {
339
+ "seconds": 0.9246388850006042,
340
+ "bytes": 1100937,
341
+ "pages": 34,
342
+ "shards": 1
343
+ },
344
+ {
345
+ "seconds": 1.532587591000265,
346
+ "bytes": 13240452,
347
+ "pages": 31,
348
+ "shards": 4
349
+ },
350
+ {
351
+ "seconds": 1.8508207049999328,
352
+ "bytes": 10873151,
353
+ "pages": 40,
354
+ "shards": 3
355
+ },
356
+ {
357
+ "seconds": 0.9900608629995986,
358
+ "bytes": 1198965,
359
+ "pages": 34,
360
+ "shards": 1
361
+ },
362
+ {
363
+ "seconds": 1.5980903330000729,
364
+ "bytes": 7181818,
365
+ "pages": 30,
366
+ "shards": 3
367
+ },
368
+ {
369
+ "seconds": 0.9470139499990182,
370
+ "bytes": 1239114,
371
+ "pages": 35,
372
+ "shards": 2
373
+ },
374
+ {
375
+ "seconds": 2.040032099999735,
376
+ "bytes": 7666715,
377
+ "pages": 33,
378
+ "shards": 3
379
+ },
380
+ {
381
+ "seconds": 34.010384213999714,
382
+ "bytes": 15924428,
383
+ "pages": 39,
384
+ "shards": 4
385
+ },
386
+ {
387
+ "seconds": 1.0855836930004443,
388
+ "bytes": 1537634,
389
+ "pages": 45,
390
+ "shards": 1
391
+ },
392
+ {
393
+ "seconds": 1.2260879469995416,
394
+ "bytes": 7238361,
395
+ "pages": 30,
396
+ "shards": 3
397
+ },
398
+ {
399
+ "seconds": 1.2277038499996706,
400
+ "bytes": 10100946,
401
+ "pages": 32,
402
+ "shards": 3
403
+ },
404
+ {
405
+ "seconds": 0.9283436340010667,
406
+ "bytes": 2568999,
407
+ "pages": 37,
408
+ "shards": 2
409
+ },
410
+ {
411
+ "seconds": 1.7634025150000525,
412
+ "bytes": 6258608,
413
+ "pages": 37,
414
+ "shards": 2
415
+ },
416
+ {
417
+ "seconds": 1.2659589719987707,
418
+ "bytes": 11586393,
419
+ "pages": 31,
420
+ "shards": 3
421
+ },
422
+ {
423
+ "seconds": 1.253272633000961,
424
+ "bytes": 6708450,
425
+ "pages": 42,
426
+ "shards": 3
427
+ },
428
+ {
429
+ "seconds": 1.5571684259994072,
430
+ "bytes": 5848121,
431
+ "pages": 28,
432
+ "shards": 2
433
+ },
434
+ {
435
+ "seconds": 1.1272584719990846,
436
+ "bytes": 2509117,
437
+ "pages": 36,
438
+ "shards": 2
439
+ },
440
+ {
441
+ "seconds": 1.3743308060002164,
442
+ "bytes": 6082381,
443
+ "pages": 31,
444
+ "shards": 2
445
+ },
446
+ {
447
+ "seconds": 1.292268752000382,
448
+ "bytes": 11000537,
449
+ "pages": 38,
450
+ "shards": 3
451
+ },
452
+ {
453
+ "seconds": 1.4258081890002359,
454
+ "bytes": 8234938,
455
+ "pages": 31,
456
+ "shards": 3
457
+ },
458
+ {
459
+ "seconds": 1.4250404260001233,
460
+ "bytes": 6072497,
461
+ "pages": 36,
462
+ "shards": 2
463
+ },
464
+ {
465
+ "seconds": 1.7697906350003905,
466
+ "bytes": 6241903,
467
+ "pages": 36,
468
+ "shards": 2
469
+ },
470
+ {
471
+ "seconds": 1.7708508949999668,
472
+ "bytes": 6504956,
473
+ "pages": 35,
474
+ "shards": 3
475
+ },
476
+ {
477
+ "seconds": 1.1802944610008126,
478
+ "bytes": 6153792,
479
+ "pages": 36,
480
+ "shards": 2
481
+ },
482
+ {
483
+ "seconds": 1.060665737999443,
484
+ "bytes": 2571654,
485
+ "pages": 33,
486
+ "shards": 2
487
+ },
488
+ {
489
+ "seconds": 1.9573821910016704,
490
+ "bytes": 16491610,
491
+ "pages": 35,
492
+ "shards": 4
493
+ },
494
+ {
495
+ "seconds": 1.2731761789982556,
496
+ "bytes": 6636284,
497
+ "pages": 37,
498
+ "shards": 2
499
+ },
500
+ {
501
+ "seconds": 1.2044296969997959,
502
+ "bytes": 2640995,
503
+ "pages": 37,
504
+ "shards": 2
505
+ },
506
+ {
507
+ "seconds": 1.348795689000326,
508
+ "bytes": 5804474,
509
+ "pages": 28,
510
+ "shards": 2
511
+ },
512
+ {
513
+ "seconds": 1.1589008739993005,
514
+ "bytes": 2402568,
515
+ "pages": 31,
516
+ "shards": 2
517
+ },
518
+ {
519
+ "seconds": 1.2708631089990376,
520
+ "bytes": 9724430,
521
+ "pages": 38,
522
+ "shards": 3
523
+ },
524
+ {
525
+ "seconds": 1.795558247000372,
526
+ "bytes": 12019830,
527
+ "pages": 36,
528
+ "shards": 3
529
+ },
530
+ {
531
+ "seconds": 0.9979082219997508,
532
+ "bytes": 2716683,
533
+ "pages": 37,
534
+ "shards": 2
535
+ },
536
+ {
537
+ "seconds": 1.3286046470002475,
538
+ "bytes": 11093643,
539
+ "pages": 34,
540
+ "shards": 3
541
+ },
542
+ {
543
+ "seconds": 1.2714691889996175,
544
+ "bytes": 7462349,
545
+ "pages": 39,
546
+ "shards": 3
547
+ },
548
+ {
549
+ "seconds": 1.0150521480009047,
550
+ "bytes": 842077,
551
+ "pages": 26,
552
+ "shards": 2
553
+ },
554
+ {
555
+ "seconds": 1.1759212130000378,
556
+ "bytes": 6164782,
557
+ "pages": 33,
558
+ "shards": 2
559
+ },
560
+ {
561
+ "seconds": 1.4110855310009356,
562
+ "bytes": 8184458,
563
+ "pages": 31,
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:c55cec8e778846c453320337a4022807364361a6bb535fd24e51ca5d5f695ab5
3
- size 721765887
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:379eeb0d627e11d6aeb818be90d2beb7ad27d2bcf192919683768be57619bc1b
3
+ size 721778751
training_config.json CHANGED
@@ -53,8 +53,8 @@
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
- "prefetch_batches": 64,
57
- "prefetch_low_water": 32,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
60
  "force_rebenchmark": false,
 
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,
60
  "force_rebenchmark": false,