appvoid commited on
Commit
0bbcbaa
·
verified ·
1 Parent(s): beb90ed

Qwen90M step 483230; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +48 -48
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +181 -181
  6. trainer.pt +2 -2
  7. training_config.json +1 -1
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 480000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "e59bfaa3ad90f7bdfba61500283a1881f70fb7357e98f55e965e5c5450a7199a",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "db9a911bd1f9aa38ba6720dbd8e0a226e9b574ebcb1cd5814f64817fae1e107d",
12
- "training_config.json": "42b690efbefcf434fbcb690b27d17a2af5e3b65c3c6aab71efa1d83d1c6a6905",
13
  "performance.json": "58f8b6f79b191aa11b2bedb4594ca2897b6fdad2419ea7c9210688ac8022e7cb",
14
- "shard_scheduler.json": "99dff589984cd9ecd0f8113aa15c684b2d16a8376d52ae86de45ab20d0a7c84e",
15
- "coverage.json": "15b555705988c6d4ac287412f0fe38ff9364ae15871a90a339f590525bb34a89",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 483230,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "3091ee2c20e9e7b92f2fbd1e0f7e26d27ee1f69d13a57fb3cec81e93a601cebe",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "dce3605b8fce93509920de867adac7256115c7fda251e6af444df4ef1a81445a",
12
+ "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6",
13
  "performance.json": "58f8b6f79b191aa11b2bedb4594ca2897b6fdad2419ea7c9210688ac8022e7cb",
14
+ "shard_scheduler.json": "39a4c183a128fcd08cafa3c3e7c2cdf0b813ba4d0d8bcde2764b7a2fb143e4f4",
15
+ "coverage.json": "04878bda51568cbea45cc34a1b0c8774e481860c8887b6161ce50e0119dd075f",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 480000,
3
  "stats": {
4
- "climbmix/blocks": 17280000,
5
- "climbmix/targets": 28884017959,
6
- "climbmix/input_tokens": 28885848126,
7
- "ultra_qa/blocks": 6480000,
8
- "ultra_qa/targets": 10570973360,
9
- "ultra_qa/input_tokens": 10570993707,
10
- "cortex/blocks": 6480000,
11
- "cortex/targets": 246461343,
12
- "cortex/input_tokens": 10301013526,
13
- "ultra_style/blocks": 6480000,
14
- "ultra_style/targets": 10249752091,
15
- "ultra_style/input_tokens": 10250288878,
16
- "openmath/blocks": 4320000,
17
- "openmath/targets": 6675226184,
18
- "openmath/input_tokens": 6677450412,
19
- "rewrite6/blocks": 2160000,
20
- "rewrite6/targets": 3541074887,
21
- "rewrite6/input_tokens": 3544957180
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 2366,
26
- "row": 3875,
27
  "start": {
28
- "file": 114,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 41,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 9695011
38
  },
39
  "rewrite6": {
40
- "page": 46,
41
- "row": 1940,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 188416
47
  }
48
  },
49
- "epoch": 26,
50
- "offset": 2048,
51
- "documents": 5530860
52
  },
53
  "openmath": {
54
- "page": 1479,
55
- "row": 475,
56
  "start": {
57
  "file": 13,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 381,
61
- "row": 534
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 6058459
67
  },
68
  "ultra_qa": {
69
- "page": 639,
70
- "row": 2127,
71
  "start": {
72
  "file": 5,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
- "row": 19215
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 4096,
81
- "documents": 2619471
82
  },
83
  "ultra_style": {
84
- "page": 975,
85
- "row": 2846,
86
  "start": {
87
  "file": 5,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 6,
91
- "row": 59733
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 4096,
96
- "documents": 3996446
97
  },
98
  "cortex": {
99
- "page": 125681,
100
  "row": 34,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 16087202
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 483230,
3
  "stats": {
4
+ "climbmix/blocks": 17396280,
5
+ "climbmix/targets": 29078088464,
6
+ "climbmix/input_tokens": 29079931166,
7
+ "ultra_qa/blocks": 6523605,
8
+ "ultra_qa/targets": 10642172522,
9
+ "ultra_qa/input_tokens": 10642193016,
10
+ "cortex/blocks": 6523605,
11
+ "cortex/targets": 248114285,
12
+ "cortex/input_tokens": 10370213106,
13
+ "ultra_style/blocks": 6523605,
14
+ "ultra_style/targets": 10318245277,
15
+ "ultra_style/input_tokens": 10318785974,
16
+ "openmath/blocks": 4349070,
17
+ "openmath/targets": 6720156497,
18
+ "openmath/input_tokens": 6722395717,
19
+ "rewrite6/blocks": 2174535,
20
+ "rewrite6/targets": 3564874771,
21
+ "rewrite6/input_tokens": 3568783073
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 2383,
26
+ "row": 126,
27
  "start": {
28
+ "file": 115,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 26,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 9760894
38
  },
39
  "rewrite6": {
40
+ "page": 5,
41
+ "row": 1536,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 20480
47
  }
48
  },
49
+ "epoch": 27,
50
+ "offset": 0,
51
+ "documents": 5567924
52
  },
53
  "openmath": {
54
+ "page": 1489,
55
+ "row": 259,
56
  "start": {
57
  "file": 13,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 422,
61
+ "row": 494
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 6099203
67
  },
68
  "ultra_qa": {
69
+ "page": 643,
70
+ "row": 3346,
71
  "start": {
72
  "file": 5,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
+ "row": 35599
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 0,
81
+ "documents": 2637074
82
  },
83
  "ultra_style": {
84
+ "page": 982,
85
+ "row": 1968,
86
  "start": {
87
  "file": 5,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 7,
91
+ "row": 1917
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 4024240
97
  },
98
  "cortex": {
99
+ "page": 126525,
100
  "row": 34,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 16195234
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 480000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "e59bfaa3ad90f7bdfba61500283a1881f70fb7357e98f55e965e5c5450a7199a", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "db9a911bd1f9aa38ba6720dbd8e0a226e9b574ebcb1cd5814f64817fae1e107d", "training_config.json": "42b690efbefcf434fbcb690b27d17a2af5e3b65c3c6aab71efa1d83d1c6a6905", "performance.json": "58f8b6f79b191aa11b2bedb4594ca2897b6fdad2419ea7c9210688ac8022e7cb", "shard_scheduler.json": "99dff589984cd9ecd0f8113aa15c684b2d16a8376d52ae86de45ab20d0a7c84e", "coverage.json": "15b555705988c6d4ac287412f0fe38ff9364ae15871a90a339f590525bb34a89", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000480000"}
 
1
+ {"step": 483230, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "3091ee2c20e9e7b92f2fbd1e0f7e26d27ee1f69d13a57fb3cec81e93a601cebe", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "dce3605b8fce93509920de867adac7256115c7fda251e6af444df4ef1a81445a", "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6", "performance.json": "58f8b6f79b191aa11b2bedb4594ca2897b6fdad2419ea7c9210688ac8022e7cb", "shard_scheduler.json": "39a4c183a128fcd08cafa3c3e7c2cdf0b813ba4d0d8bcde2764b7a2fb143e4f4", "coverage.json": "04878bda51568cbea45cc34a1b0c8774e481860c8887b6161ce50e0119dd075f", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000483230"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e59bfaa3ad90f7bdfba61500283a1881f70fb7357e98f55e965e5c5450a7199a
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3091ee2c20e9e7b92f2fbd1e0f7e26d27ee1f69d13a57fb3cec81e93a601cebe
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": 75948.0229487074,
26
- "required_pages_per_second_upper_estimate": 0.005254260451283705
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": 9493.502868588424,
40
- "required_pages_per_second_upper_estimate": 0.003974731404194673
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": 18987.00573717685,
54
- "required_pages_per_second_upper_estimate": 0.003954492574040618
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": 28480.508605765273,
68
- "required_pages_per_second_upper_estimate": 0.0017073630036969789
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": 28480.508605765273,
82
- "required_pages_per_second_upper_estimate": 0.002214234271580101
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": 22,
95
- "required_token_positions_per_second": 28480.508605765273,
96
- "required_pages_per_second_upper_estimate": 0.36531738440714295
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,186 +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": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 704566,
161
- "producer_batch_seconds": 0.14875617208599579,
162
- "producer_p95_seconds": 0.35381016099563567,
163
- "gpu_update_seconds": 0.970769180519594,
164
- "producer_batches_per_second": 6.722410142564714,
165
- "gpu_batches_per_second": 1.030110988345098,
166
- "producer_headroom": 6.525908585214155,
167
  "sustainable": true,
168
- "buffer_seconds": 60.18768919221483,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 80,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 352449.9817351,
174
- "upload_bytes_per_second": 5306492.160540054,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 75.62756146468877,
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": 1.3809676959972421,
184
- "bytes": 13424902,
185
- "pages": 35,
186
- "shards": 4
187
- },
188
- {
189
- "seconds": 0.9754727400031697,
190
- "bytes": 994247,
191
- "pages": 29,
192
- "shards": 1
193
- },
194
- {
195
- "seconds": 1.2377208289981354,
196
- "bytes": 11067407,
197
- "pages": 30,
198
- "shards": 3
199
- },
200
- {
201
- "seconds": 1.5320398900003056,
202
- "bytes": 7616229,
203
- "pages": 35,
204
- "shards": 3
205
- },
206
- {
207
- "seconds": 0.994875273998332,
208
- "bytes": 1060442,
209
- "pages": 31,
210
- "shards": 1
211
- },
212
- {
213
- "seconds": 1.0037741340020148,
214
- "bytes": 6334013,
215
- "pages": 36,
216
- "shards": 2
217
- },
218
- {
219
- "seconds": 1.4788876900020114,
220
- "bytes": 12280922,
221
- "pages": 40,
222
- "shards": 4
223
- },
224
- {
225
- "seconds": 1.3189756050014694,
226
- "bytes": 5844934,
227
- "pages": 28,
228
- "shards": 2
229
- },
230
- {
231
- "seconds": 1.142625862001296,
232
- "bytes": 7700249,
233
- "pages": 41,
234
- "shards": 3
235
- },
236
- {
237
- "seconds": 0.8653922449957463,
238
- "bytes": 888116,
239
- "pages": 27,
240
- "shards": 1
241
- },
242
- {
243
- "seconds": 1.3053653089955333,
244
- "bytes": 10336652,
245
- "pages": 41,
246
- "shards": 3
247
- },
248
- {
249
- "seconds": 1.1407873469943297,
250
- "bytes": 7506775,
251
- "pages": 31,
252
- "shards": 3
253
- },
254
- {
255
- "seconds": 1.0540632260017446,
256
- "bytes": 6792818,
257
- "pages": 30,
258
- "shards": 3
259
- },
260
- {
261
- "seconds": 0.9928698779986007,
262
- "bytes": 5905666,
263
- "pages": 30,
264
- "shards": 2
265
- },
266
- {
267
- "seconds": 2.588622392999241,
268
- "bytes": 7185173,
269
- "pages": 40,
270
- "shards": 3
271
- },
272
- {
273
- "seconds": 1.1344652970001334,
274
- "bytes": 5709671,
275
- "pages": 28,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 1.2757203309956822,
280
- "bytes": 7569977,
281
- "pages": 40,
282
- "shards": 3
283
- },
284
- {
285
- "seconds": 0.8296898909975425,
286
- "bytes": 1129072,
287
- "pages": 35,
288
- "shards": 1
289
- },
290
- {
291
- "seconds": 1.0205191320055746,
292
- "bytes": 9233133,
293
- "pages": 13,
294
- "shards": 3
295
- },
296
- {
297
- "seconds": 1.1241545290031354,
298
- "bytes": 8856234,
299
- "pages": 49,
300
- "shards": 3
301
- },
302
- {
303
- "seconds": 1.2785397469997406,
304
- "bytes": 5725754,
305
- "pages": 29,
306
- "shards": 2
307
- },
308
- {
309
- "seconds": 1.0718349349990604,
310
- "bytes": 7504941,
311
- "pages": 39,
312
- "shards": 3
313
- },
314
- {
315
- "seconds": 1.2024066430021776,
316
- "bytes": 5964235,
317
- "pages": 32,
318
- "shards": 2
319
- },
320
- {
321
- "seconds": 1.2837438990027294,
322
- "bytes": 5816020,
323
- "pages": 32,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 1.0380413320017396,
328
- "bytes": 232525,
329
- "pages": 6,
330
- "shards": 1
331
- },
332
- {
333
- "seconds": 1.743612624995876,
334
- "bytes": 7907888,
335
- "pages": 23,
336
- "shards": 3
337
- },
338
  {
339
  "seconds": 2.2645728169954964,
340
  "bytes": 6235357,
@@ -562,6 +406,162 @@
562
  "bytes": 904800,
563
  "pages": 27,
564
  "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 76395.03562198069,
26
+ "required_pages_per_second_upper_estimate": 0.005285185825233053
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": 9549.379452747586,
40
+ "required_pages_per_second_upper_estimate": 0.003998125763146351
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": 19098.75890549517,
54
+ "required_pages_per_second_upper_estimate": 0.0039777678118720905
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": 28648.138358242755,
68
+ "required_pages_per_second_upper_estimate": 0.0017174121514021922
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": 28648.138358242755,
82
+ "required_pages_per_second_upper_estimate": 0.0022272667475098673
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": 23,
95
+ "required_token_positions_per_second": 28648.138358242755,
96
+ "required_pages_per_second_upper_estimate": 0.3674675588851189
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": 704566,
161
+ "producer_batch_seconds": 0.13329735316432334,
162
+ "producer_p95_seconds": 0.36278332399524515,
163
+ "gpu_update_seconds": 0.9650888882992639,
164
+ "producer_batches_per_second": 7.50202443080203,
165
+ "gpu_batches_per_second": 1.0361739857582015,
166
+ "producer_headroom": 7.240120417916649,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.800599962853624,
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": 298978.01718284166,
174
+ "upload_bytes_per_second": 5442207.661463906,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 64.15372263540208,
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": 2.2645728169954964,
184
  "bytes": 6235357,
 
406
  "bytes": 904800,
407
  "pages": 27,
408
  "shards": 1
409
+ },
410
+ {
411
+ "seconds": 1.1419830579980044,
412
+ "bytes": 7590743,
413
+ "pages": 31,
414
+ "shards": 3
415
+ },
416
+ {
417
+ "seconds": 0.8363852010006667,
418
+ "bytes": 1241882,
419
+ "pages": 35,
420
+ "shards": 1
421
+ },
422
+ {
423
+ "seconds": 1.234218167999643,
424
+ "bytes": 11088956,
425
+ "pages": 37,
426
+ "shards": 4
427
+ },
428
+ {
429
+ "seconds": 1.0702078030008124,
430
+ "bytes": 12031494,
431
+ "pages": 37,
432
+ "shards": 3
433
+ },
434
+ {
435
+ "seconds": 0.9152153160030139,
436
+ "bytes": 886815,
437
+ "pages": 27,
438
+ "shards": 1
439
+ },
440
+ {
441
+ "seconds": 1.3266999849947751,
442
+ "bytes": 11836619,
443
+ "pages": 37,
444
+ "shards": 4
445
+ },
446
+ {
447
+ "seconds": 0.9408924880044651,
448
+ "bytes": 1241676,
449
+ "pages": 37,
450
+ "shards": 1
451
+ },
452
+ {
453
+ "seconds": 1.40072601299471,
454
+ "bytes": 6187128,
455
+ "pages": 29,
456
+ "shards": 2
457
+ },
458
+ {
459
+ "seconds": 1.231672369001899,
460
+ "bytes": 7550882,
461
+ "pages": 41,
462
+ "shards": 3
463
+ },
464
+ {
465
+ "seconds": 1.3564282489969628,
466
+ "bytes": 10548156,
467
+ "pages": 30,
468
+ "shards": 3
469
+ },
470
+ {
471
+ "seconds": 1.1907792190031614,
472
+ "bytes": 7251298,
473
+ "pages": 32,
474
+ "shards": 4
475
+ },
476
+ {
477
+ "seconds": 0.9704950960003771,
478
+ "bytes": 1288101,
479
+ "pages": 36,
480
+ "shards": 1
481
+ },
482
+ {
483
+ "seconds": 1.2859423129993957,
484
+ "bytes": 5955041,
485
+ "pages": 29,
486
+ "shards": 2
487
+ },
488
+ {
489
+ "seconds": 1.6710909179964801,
490
+ "bytes": 11523305,
491
+ "pages": 38,
492
+ "shards": 4
493
+ },
494
+ {
495
+ "seconds": 0.8526758169973618,
496
+ "bytes": 1207490,
497
+ "pages": 36,
498
+ "shards": 1
499
+ },
500
+ {
501
+ "seconds": 1.3300546199970995,
502
+ "bytes": 11866142,
503
+ "pages": 30,
504
+ "shards": 3
505
+ },
506
+ {
507
+ "seconds": 0.8256019309992553,
508
+ "bytes": 2326710,
509
+ "pages": 30,
510
+ "shards": 2
511
+ },
512
+ {
513
+ "seconds": 1.1752165859943489,
514
+ "bytes": 9508137,
515
+ "pages": 30,
516
+ "shards": 3
517
+ },
518
+ {
519
+ "seconds": 1.008012272999622,
520
+ "bytes": 7606530,
521
+ "pages": 40,
522
+ "shards": 3
523
+ },
524
+ {
525
+ "seconds": 0.8981697260023793,
526
+ "bytes": 1206792,
527
+ "pages": 36,
528
+ "shards": 1
529
+ },
530
+ {
531
+ "seconds": 1.5740248830043129,
532
+ "bytes": 5500237,
533
+ "pages": 20,
534
+ "shards": 2
535
+ },
536
+ {
537
+ "seconds": 1.1595400419973885,
538
+ "bytes": 7406603,
539
+ "pages": 34,
540
+ "shards": 3
541
+ },
542
+ {
543
+ "seconds": 1.1209028549928917,
544
+ "bytes": 11633071,
545
+ "pages": 38,
546
+ "shards": 3
547
+ },
548
+ {
549
+ "seconds": 1.5646861489949515,
550
+ "bytes": 6165459,
551
+ "pages": 36,
552
+ "shards": 2
553
+ },
554
+ {
555
+ "seconds": 0.9572175139983301,
556
+ "bytes": 2339484,
557
+ "pages": 32,
558
+ "shards": 2
559
+ },
560
+ {
561
+ "seconds": 1.3525367339971126,
562
+ "bytes": 9731842,
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:db9a911bd1f9aa38ba6720dbd8e0a226e9b574ebcb1cd5814f64817fae1e107d
3
- size 721770047
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dce3605b8fce93509920de867adac7256115c7fda251e6af444df4ef1a81445a
3
+ size 721773887
training_config.json CHANGED
@@ -46,7 +46,7 @@
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,
 
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
+ "prefetch_batches": 63,
50
  "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,