appvoid commited on
Commit
a98ad73
·
verified ·
1 Parent(s): 832a343

Qwen90M step 485000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
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"
 
1
  {
2
+ "step": 485000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "7fd364f1b254abb0a762a84c0a7c714c534994ffdbdf3dc30a0b7cae01ffeb6e",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "2e31b934cad9f27085ea79b3fd3db6d8ab83a24db1f9125cd66705d726ad26e8",
12
+ "training_config.json": "42b690efbefcf434fbcb690b27d17a2af5e3b65c3c6aab71efa1d83d1c6a6905",
13
+ "performance.json": "e2fbfc542f997748539ec27cc61ae6365d39122628b292c00db059dd801c9082",
14
+ "shard_scheduler.json": "b8e4654ade0997d04db51bcd77ba9d9a995dfd09656a2adb9fbb8efc28315f5f",
15
+ "coverage.json": "32c5b1549b70f2106c3b9addeed59093c0fe5ba7d9a8bd64ba52215c1518731c",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
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
  }
 
1
  {
2
+ "step": 485000,
3
  "stats": {
4
+ "climbmix/blocks": 17460000,
5
+ "climbmix/targets": 29184768142,
6
+ "climbmix/input_tokens": 29186617700,
7
+ "ultra_qa/blocks": 6547500,
8
+ "ultra_qa/targets": 10681167812,
9
+ "ultra_qa/input_tokens": 10681188377,
10
+ "cortex/blocks": 6547500,
11
+ "cortex/targets": 249030011,
12
+ "cortex/input_tokens": 10408163590,
13
+ "ultra_style/blocks": 6547500,
14
+ "ultra_style/targets": 10356067453,
15
+ "ultra_style/input_tokens": 10356610174,
16
+ "openmath/blocks": 4365000,
17
+ "openmath/targets": 6744753340,
18
+ "openmath/input_tokens": 6747000681,
19
+ "rewrite6/blocks": 2182500,
20
+ "rewrite6/targets": 3577914913,
21
+ "rewrite6/input_tokens": 3581837409
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 2391,
26
+ "row": 3140,
27
  "start": {
28
  "file": 115,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 58,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 9796676
38
  },
39
  "rewrite6": {
40
+ "page": 10,
41
+ "row": 1343,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 40960
47
  }
48
  },
49
  "epoch": 27,
50
  "offset": 0,
51
+ "documents": 5588211
52
  },
53
  "openmath": {
54
+ "page": 1494,
55
+ "row": 2026,
56
  "start": {
57
+ "file": 14,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 6,
61
+ "row": 324
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 6121450
67
  },
68
  "ultra_qa": {
69
+ "page": 646,
70
+ "row": 785,
71
  "start": {
72
  "file": 5,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 0,
76
+ "row": 47887
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 2048,
81
+ "documents": 2646801
82
  },
83
  "ultra_style": {
84
+ "page": 986,
85
+ "row": 411,
86
  "start": {
87
  "file": 5,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 7,
91
+ "row": 18301
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 4039067
97
  },
98
  "cortex": {
99
+ "page": 126988,
100
+ "row": 6,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 16254470
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
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"}
 
1
+ {"step": 485000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "7fd364f1b254abb0a762a84c0a7c714c534994ffdbdf3dc30a0b7cae01ffeb6e", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "2e31b934cad9f27085ea79b3fd3db6d8ab83a24db1f9125cd66705d726ad26e8", "training_config.json": "42b690efbefcf434fbcb690b27d17a2af5e3b65c3c6aab71efa1d83d1c6a6905", "performance.json": "e2fbfc542f997748539ec27cc61ae6365d39122628b292c00db059dd801c9082", "shard_scheduler.json": "b8e4654ade0997d04db51bcd77ba9d9a995dfd09656a2adb9fbb8efc28315f5f", "coverage.json": "32c5b1549b70f2106c3b9addeed59093c0fe5ba7d9a8bd64ba52215c1518731c", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000485000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3091ee2c20e9e7b92f2fbd1e0f7e26d27ee1f69d13a57fb3cec81e93a601cebe
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7fd364f1b254abb0a762a84c0a7c714c534994ffdbdf3dc30a0b7cae01ffeb6e
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9383984099999907,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9383984099999907,
56
- "input_tokens_per_second": 196419.77014858948
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9398671289999925,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9398671289999925,
56
+ "input_tokens_per_second": 196112.82734838733
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": 76395.03562198069,
26
- "required_pages_per_second_upper_estimate": 0.005285185825233053
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": 9549.379452747586,
40
- "required_pages_per_second_upper_estimate": 0.003998125763146351
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": 19098.75890549517,
54
- "required_pages_per_second_upper_estimate": 0.0039777678118720905
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": 28648.138358242755,
68
- "required_pages_per_second_upper_estimate": 0.0017174121514021922
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": 28648.138358242755,
82
- "required_pages_per_second_upper_estimate": 0.0022272667475098673
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": 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.",
@@ -155,412 +155,100 @@
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,
185
- "pages": 44,
186
- "shards": 2
187
- },
188
- {
189
- "seconds": 1.3341411860019434,
190
- "bytes": 11745696,
191
- "pages": 39,
192
- "shards": 4
193
- },
194
- {
195
- "seconds": 3.1945914269963396,
196
- "bytes": 906980,
197
- "pages": 27,
198
  "shards": 1
199
  },
200
  {
201
- "seconds": 1.100945904996479,
202
- "bytes": 6366458,
203
  "pages": 37,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 1.1544355079968227,
208
- "bytes": 7474922,
209
- "pages": 32,
210
- "shards": 3
211
- },
212
- {
213
- "seconds": 1.0820681480036,
214
- "bytes": 10987528,
215
- "pages": 39,
216
  "shards": 3
217
  },
218
  {
219
- "seconds": 1.2853161590028321,
220
- "bytes": 6397328,
221
- "pages": 37,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 1.1332088359995396,
226
- "bytes": 7397305,
227
- "pages": 32,
228
- "shards": 3
229
- },
230
- {
231
- "seconds": 0.7730173130039475,
232
- "bytes": 1053057,
233
- "pages": 32,
234
- "shards": 1
235
- },
236
- {
237
- "seconds": 1.3466375830030302,
238
- "bytes": 11763121,
239
- "pages": 37,
240
- "shards": 4
241
  },
242
  {
243
- "seconds": 1.4802930930018192,
244
- "bytes": 1138429,
245
  "pages": 32,
246
- "shards": 1
247
- },
248
- {
249
- "seconds": 1.8897758500024793,
250
- "bytes": 11833830,
251
- "pages": 33,
252
  "shards": 3
253
  },
254
  {
255
- "seconds": 1.1506007540010614,
256
- "bytes": 2440099,
257
- "pages": 31,
258
  "shards": 2
259
  },
260
  {
261
- "seconds": 1.5953800970019074,
262
- "bytes": 10985431,
263
- "pages": 43,
264
- "shards": 3
265
- },
266
- {
267
- "seconds": 1.5438401100036572,
268
- "bytes": 7453011,
269
  "pages": 31,
270
  "shards": 3
271
  },
272
  {
273
- "seconds": 1.0392796939995605,
274
- "bytes": 1282728,
275
  "pages": 37,
276
- "shards": 1
277
- },
278
- {
279
- "seconds": 1.3358671570022125,
280
- "bytes": 5881131,
281
- "pages": 29,
282
  "shards": 2
283
  },
284
  {
285
- "seconds": 1.5660001680007554,
286
- "bytes": 16980575,
287
- "pages": 44,
288
- "shards": 6
289
- },
290
- {
291
- "seconds": 0.8579309720007586,
292
- "bytes": 927462,
293
- "pages": 27,
294
- "shards": 1
295
- },
296
- {
297
- "seconds": 1.3555858160034404,
298
- "bytes": 7535281,
299
- "pages": 40,
300
- "shards": 3
301
- },
302
- {
303
- "seconds": 1.1513379759999225,
304
- "bytes": 10862357,
305
- "pages": 31,
306
- "shards": 3
307
- },
308
- {
309
- "seconds": 0.9964048779947916,
310
- "bytes": 1334873,
311
- "pages": 36,
312
- "shards": 1
313
- },
314
- {
315
- "seconds": 1.2506730089953635,
316
- "bytes": 7336892,
317
- "pages": 32,
318
- "shards": 3
319
- },
320
- {
321
- "seconds": 1.3737510540013318,
322
- "bytes": 6860686,
323
- "pages": 33,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 1.2590301210002508,
328
- "bytes": 10720957,
329
- "pages": 40,
330
- "shards": 3
331
- },
332
- {
333
- "seconds": 1.187173960999644,
334
- "bytes": 7325771,
335
- "pages": 31,
336
- "shards": 3
337
- },
338
- {
339
- "seconds": 0.8256457960014814,
340
- "bytes": 1213741,
341
  "pages": 36,
342
- "shards": 1
343
- },
344
- {
345
- "seconds": 1.0744473489976372,
346
- "bytes": 7318826,
347
- "pages": 31,
348
- "shards": 3
349
- },
350
- {
351
- "seconds": 1.3007696439963183,
352
- "bytes": 9895316,
353
- "pages": 41,
354
- "shards": 3
355
- },
356
- {
357
- "seconds": 1.3112097839984926,
358
- "bytes": 6821672,
359
- "pages": 34,
360
- "shards": 2
361
- },
362
- {
363
- "seconds": 1.0360545559960883,
364
- "bytes": 7415725,
365
- "pages": 31,
366
- "shards": 3
367
- },
368
- {
369
- "seconds": 0.9522623859957093,
370
- "bytes": 963026,
371
- "pages": 30,
372
- "shards": 1
373
- },
374
- {
375
- "seconds": 1.1533947639982216,
376
- "bytes": 11916585,
377
- "pages": 41,
378
- "shards": 4
379
- },
380
- {
381
- "seconds": 1.1312003819984966,
382
- "bytes": 5813649,
383
- "pages": 29,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 1.0730638719978742,
388
- "bytes": 1187099,
389
- "pages": 36,
390
- "shards": 1
391
- },
392
- {
393
- "seconds": 1.1143769740010612,
394
- "bytes": 13033863,
395
- "pages": 32,
396
- "shards": 4
397
- },
398
- {
399
- "seconds": 1.1261608329950832,
400
- "bytes": 10613783,
401
- "pages": 39,
402
- "shards": 3
403
- },
404
- {
405
- "seconds": 1.0124226469997666,
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
  ],
@@ -606,7 +294,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 2.3715912330000037,
610
- 0.09919368099997428
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": 75745.19625580676,
26
+ "required_pages_per_second_upper_estimate": 0.005240228430046073
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": 9468.149531975845,
40
+ "required_pages_per_second_upper_estimate": 0.003964116491592859
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": 18936.29906395169,
54
+ "required_pages_per_second_upper_estimate": 0.003943931711232715
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": 28404.448595927533,
68
+ "required_pages_per_second_upper_estimate": 0.0017028033222441167
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": 28404.448595927533,
82
+ "required_pages_per_second_upper_estimate": 0.0022083209403678437
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": 28404.448595927533,
96
+ "required_pages_per_second_upper_estimate": 0.3643417682678202
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": 704566,
161
+ "producer_batch_seconds": 0.16442591646878846,
162
+ "producer_p95_seconds": 0.13880198200013183,
163
+ "gpu_update_seconds": 0.973368657611048,
164
+ "producer_batches_per_second": 6.081766314435117,
165
+ "gpu_batches_per_second": 1.0273599752577955,
166
+ "producer_headroom": 5.919800713385801,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.348856771884975,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 69,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 424533.6472953534,
174
+ "upload_bytes_per_second": 5696740.445231295,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 91.09503807206585,
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": 0.9475190029999112,
184
+ "bytes": 21104,
185
+ "pages": 1,
 
 
 
 
 
 
 
 
 
 
 
 
186
  "shards": 1
187
  },
188
  {
189
+ "seconds": 1.029576297999938,
190
+ "bytes": 6528629,
191
  "pages": 37,
 
 
 
 
 
 
 
 
 
 
 
 
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 1.0352755850000221,
196
+ "bytes": 6365146,
197
+ "pages": 35,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.1308769879999545,
202
+ "bytes": 6215001,
203
+ "pages": 35,
204
+ "shards": 2
 
 
 
 
 
 
 
 
 
 
 
 
205
  },
206
  {
207
+ "seconds": 1.016927755999859,
208
+ "bytes": 7382400,
209
  "pages": 32,
 
 
 
 
 
 
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.1139478459999737,
214
+ "bytes": 6306186,
215
+ "pages": 35,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.1660128299999997,
220
+ "bytes": 8072363,
 
 
 
 
 
 
221
  "pages": 31,
222
  "shards": 3
223
  },
224
  {
225
+ "seconds": 0.9177546119999533,
226
+ "bytes": 6210206,
227
  "pages": 37,
 
 
 
 
 
 
228
  "shards": 2
229
  },
230
  {
231
+ "seconds": 1.1674345439998888,
232
+ "bytes": 5990051,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
233
  "pages": 36,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.0493045279999933,
238
+ "bytes": 6091029,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
239
  "pages": 31,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 1.0872325649997947,
244
+ "bytes": 6195111,
245
  "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.1556296060000477,
250
+ "bytes": 7640700,
251
+ "pages": 37,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
252
  "shards": 3
253
  }
254
  ],
 
294
  },
295
  "cache_reused": true,
296
  "resume_probe_seconds": [
297
+ 2.220795152999983,
298
+ 0.01521137299999964
299
  ]
300
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:dce3605b8fce93509920de867adac7256115c7fda251e6af444df4ef1a81445a
3
- size 721773887
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e31b934cad9f27085ea79b3fd3db6d8ab83a24db1f9125cd66705d726ad26e8
3
+ size 721772095
training_config.json CHANGED
@@ -46,7 +46,7 @@
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,
 
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,