appvoid commited on
Commit
4bccc84
·
verified ·
1 Parent(s): a303de3

Qwen90M step 625000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +49 -49
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +249 -249
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 620000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "5e1c8846d93cc736a471c45613d78007b35b8e94c1285a2904ac37057a5c3356",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "9933156547ab18b657069aa60eb52e3b669e25faa068a6484883abec26f0f1a7",
12
- "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb",
13
  "performance.json": "539995c1e4203016bfc6259363445c7c25dfb51604d74cd18d2cc7ad51bfb71e",
14
- "shard_scheduler.json": "b2d31325d348832dedfe0eeb8c5de7f5606e001abaa7474b129c45c2b3901479",
15
- "coverage.json": "bb539a2055773d2d68de8915bf0ebe63f0b3e8e118e9bf2703f2ab7066ea0769",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"
 
1
  {
2
+ "step": 625000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "e79b91ffdb8a4225e499aae34fbdee578329f2c0cbb6e3ff0bd680339364d489",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "08c9cb9f6f17586214603f83e8a6ed392e1f2c67500da7bfa88fe516501b5bdf",
12
+ "training_config.json": "99d98cec9c0814e63e7b7a5836277d5c336d3c2fb33fa69da403f3610c902c99",
13
  "performance.json": "539995c1e4203016bfc6259363445c7c25dfb51604d74cd18d2cc7ad51bfb71e",
14
+ "shard_scheduler.json": "e43f0cf9936e7a3794ba03cd0134b307fbf0346bb197cf8898bd22441a49a669",
15
+ "coverage.json": "06fc3d77f8d5cf5a9663605fb6f19290a742c2d87610dbe416069837444aa522",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 620000,
3
  "stats": {
4
- "climbmix/blocks": 22320000,
5
- "climbmix/targets": 37312355817,
6
- "climbmix/input_tokens": 37314720200,
7
- "ultra_qa/blocks": 8370000,
8
- "ultra_qa/targets": 13655080912,
9
- "ultra_qa/input_tokens": 13655107298,
10
- "cortex/blocks": 8370000,
11
- "cortex/targets": 318257872,
12
- "cortex/input_tokens": 13304792682,
13
- "ultra_style/blocks": 8370000,
14
- "ultra_style/targets": 13238688750,
15
- "ultra_style/input_tokens": 13239383690,
16
- "openmath/blocks": 5580000,
17
- "openmath/targets": 8622121150,
18
- "openmath/input_tokens": 8624993272,
19
- "rewrite6/blocks": 2790000,
20
- "rewrite6/targets": 4573810627,
21
- "rewrite6/input_tokens": 4578825235
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3056,
26
- "row": 2715,
27
  "start": {
28
- "file": 147,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 75,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 12520091
38
  },
39
  "rewrite6": {
40
- "page": 39,
41
- "row": 610,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 159744
47
  }
48
  },
49
- "epoch": 34,
50
  "offset": 0,
51
- "documents": 7144090
52
  },
53
  "openmath": {
54
- "page": 1910,
55
- "row": 970,
56
  "start": {
57
- "file": 17,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 400,
61
- "row": 310
62
  }
63
  },
64
  "epoch": 0,
65
- "offset": 2048,
66
- "documents": 7824330
67
  },
68
  "ultra_qa": {
69
- "page": 826,
70
- "row": 27,
71
  "start": {
72
  "file": 6,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 4,
76
- "row": 7499
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 3383323
82
  },
83
  "ultra_style": {
84
- "page": 1260,
85
- "row": 3840,
86
  "start": {
87
  "file": 7,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
- "row": 25036
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 5164800
97
  },
98
  "cortex": {
99
- "page": 162331,
100
- "row": 125,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 20778493
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 625000,
3
  "stats": {
4
+ "climbmix/blocks": 22500000,
5
+ "climbmix/targets": 37613482229,
6
+ "climbmix/input_tokens": 37615865584,
7
+ "ultra_qa/blocks": 8437500,
8
+ "ultra_qa/targets": 13765166892,
9
+ "ultra_qa/input_tokens": 13765193488,
10
+ "cortex/blocks": 8437500,
11
+ "cortex/targets": 320825520,
12
+ "cortex/input_tokens": 13412066305,
13
+ "ultra_style/blocks": 8437500,
14
+ "ultra_style/targets": 13345548414,
15
+ "ultra_style/input_tokens": 13346249095,
16
+ "openmath/blocks": 5625000,
17
+ "openmath/targets": 8691637642,
18
+ "openmath/input_tokens": 8694532988,
19
+ "rewrite6/blocks": 2812500,
20
+ "rewrite6/targets": 4610720475,
21
+ "rewrite6/input_tokens": 4615775369
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3081,
26
+ "row": 1020,
27
  "start": {
28
+ "file": 149,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 12,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 24576,
37
+ "documents": 12620796
38
  },
39
  "rewrite6": {
40
+ "page": 3,
41
+ "row": 45,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 12288
47
  }
48
  },
49
+ "epoch": 35,
50
  "offset": 0,
51
+ "documents": 7201473
52
  },
53
  "openmath": {
54
+ "page": 1925,
55
+ "row": 2658,
56
  "start": {
57
+ "file": 18,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 25,
61
+ "row": 100
62
  }
63
  },
64
  "epoch": 0,
65
+ "offset": 0,
66
+ "documents": 7887458
67
  },
68
  "ultra_qa": {
69
+ "page": 832,
70
+ "row": 2800,
71
  "start": {
72
  "file": 6,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 4,
76
+ "row": 32075
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 3410672
82
  },
83
  "ultra_style": {
84
+ "page": 1271,
85
+ "row": 697,
86
  "start": {
87
  "file": 7,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 4,
91
+ "row": 70092
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 5206713
97
  },
98
  "cortex": {
99
+ "page": 163641,
100
+ "row": 38,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 20946086
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 620000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "5e1c8846d93cc736a471c45613d78007b35b8e94c1285a2904ac37057a5c3356", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "9933156547ab18b657069aa60eb52e3b669e25faa068a6484883abec26f0f1a7", "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb", "performance.json": "539995c1e4203016bfc6259363445c7c25dfb51604d74cd18d2cc7ad51bfb71e", "shard_scheduler.json": "b2d31325d348832dedfe0eeb8c5de7f5606e001abaa7474b129c45c2b3901479", "coverage.json": "bb539a2055773d2d68de8915bf0ebe63f0b3e8e118e9bf2703f2ab7066ea0769", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000620000"}
 
1
+ {"step": 625000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "e79b91ffdb8a4225e499aae34fbdee578329f2c0cbb6e3ff0bd680339364d489", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "08c9cb9f6f17586214603f83e8a6ed392e1f2c67500da7bfa88fe516501b5bdf", "training_config.json": "99d98cec9c0814e63e7b7a5836277d5c336d3c2fb33fa69da403f3610c902c99", "performance.json": "539995c1e4203016bfc6259363445c7c25dfb51604d74cd18d2cc7ad51bfb71e", "shard_scheduler.json": "e43f0cf9936e7a3794ba03cd0134b307fbf0346bb197cf8898bd22441a49a669", "coverage.json": "06fc3d77f8d5cf5a9663605fb6f19290a742c2d87610dbe416069837444aa522", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000625000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5e1c8846d93cc736a471c45613d78007b35b8e94c1285a2904ac37057a5c3356
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e79b91ffdb8a4225e499aae34fbdee578329f2c0cbb6e3ff0bd680339364d489
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": 76942.96702956906,
26
- "required_pages_per_second_upper_estimate": 0.005323092991386044
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": 9617.870878696132,
40
- "required_pages_per_second_upper_estimate": 0.0040268016929273925
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": 19235.741757392265,
54
- "required_pages_per_second_upper_estimate": 0.004006297727441522
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": 28853.612636088397,
68
- "required_pages_per_second_upper_estimate": 0.0017297300206179822
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": 28853.612636088397,
82
- "required_pages_per_second_upper_estimate": 0.0022432414688265327
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": 28853.612636088397,
96
- "required_pages_per_second_upper_estimate": 0.37010316230023216
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": 0.18642643738270692,
162
- "producer_p95_seconds": 0.23165708300075494,
163
- "gpu_update_seconds": 0.9582162327021578,
164
- "producer_batches_per_second": 5.364046076507607,
165
- "gpu_batches_per_second": 1.0436057811085213,
166
- "producer_headroom": 5.139916023471909,
167
  "sustainable": true,
168
- "buffer_seconds": 60.367622660235945,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 81,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 480694.27939603303,
174
- "upload_bytes_per_second": 5198415.335503838,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 103.14580236826652,
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": 1.2262302330000239,
184
- "bytes": 12002858,
185
  "pages": 41,
186
  "shards": 4
187
  },
188
  {
189
- "seconds": 0.976691177000248,
190
- "bytes": 6244802,
191
- "pages": 36,
192
  "shards": 2
193
  },
194
  {
195
- "seconds": 1.5793915419999394,
196
- "bytes": 781594,
197
- "pages": 26,
198
- "shards": 1
199
- },
200
- {
201
- "seconds": 1.4397608880008193,
202
- "bytes": 13440043,
203
- "pages": 39,
204
- "shards": 4
205
- },
206
- {
207
- "seconds": 1.215214173998902,
208
- "bytes": 10529948,
209
- "pages": 39,
210
- "shards": 3
211
- },
212
- {
213
- "seconds": 1.3451746110004024,
214
- "bytes": 2254637,
215
  "pages": 28,
216
  "shards": 2
217
  },
218
  {
219
- "seconds": 1.1764542460005032,
220
- "bytes": 6191157,
221
  "pages": 36,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 1.0847412589992018,
226
- "bytes": 1126127,
227
- "pages": 34,
228
- "shards": 1
229
  },
230
  {
231
- "seconds": 1.2304532270009076,
232
- "bytes": 11321849,
233
- "pages": 33,
234
- "shards": 4
 
 
 
 
 
 
235
  },
236
  {
237
- "seconds": 1.4404141169998184,
238
- "bytes": 12120623,
239
  "pages": 39,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 0.8672846369991021,
244
- "bytes": 2691618,
245
- "pages": 37,
246
  "shards": 2
247
  },
248
  {
249
- "seconds": 0.9980244999987917,
250
- "bytes": 5788373,
251
- "pages": 22,
252
  "shards": 2
253
  },
254
  {
255
- "seconds": 3.3476856620000035,
256
- "bytes": 5904691,
257
- "pages": 43,
258
  "shards": 2
259
  },
260
  {
261
- "seconds": 1.2678156700003456,
262
- "bytes": 7637773,
 
 
 
 
 
 
263
  "pages": 38,
264
  "shards": 3
265
  },
266
  {
267
- "seconds": 1.1486175610007194,
268
- "bytes": 5966085,
 
 
 
 
 
 
269
  "pages": 28,
270
  "shards": 2
271
  },
272
  {
273
- "seconds": 1.0020495059998211,
274
- "bytes": 6673858,
275
- "pages": 32,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.1948771370007307,
280
- "bytes": 12132833,
281
- "pages": 38,
282
- "shards": 4
283
  },
284
  {
285
- "seconds": 1.2424271460004093,
286
- "bytes": 1081337,
287
- "pages": 34,
288
- "shards": 1
289
  },
290
  {
291
- "seconds": 1.3408090630000515,
292
- "bytes": 7658166,
293
- "pages": 39,
 
 
 
 
 
 
294
  "shards": 3
295
  },
296
  {
297
- "seconds": 0.8867853440006002,
298
- "bytes": 5785267,
299
- "pages": 27,
300
  "shards": 2
301
  },
302
  {
303
- "seconds": 1.081288327000948,
304
- "bytes": 5787210,
305
- "pages": 36,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 1.1759565290012688,
310
- "bytes": 13132586,
311
- "pages": 31,
312
- "shards": 4
313
  },
314
  {
315
- "seconds": 1.004212905998429,
316
- "bytes": 6157452,
317
- "pages": 38,
318
  "shards": 3
319
  },
320
  {
321
- "seconds": 1.3965238540004066,
322
- "bytes": 817171,
323
- "pages": 26,
324
  "shards": 1
325
  },
326
  {
327
- "seconds": 1.115737056999933,
328
- "bytes": 11496202,
329
- "pages": 35,
330
- "shards": 4
331
  },
332
  {
333
- "seconds": 1.037322108000808,
334
- "bytes": 1047345,
335
- "pages": 34,
336
- "shards": 1
337
  },
338
  {
339
- "seconds": 1.0591198959991743,
340
- "bytes": 7189815,
341
- "pages": 31,
342
- "shards": 3
343
  },
344
  {
345
- "seconds": 1.272207456000615,
346
- "bytes": 11979936,
347
- "pages": 41,
348
  "shards": 3
349
  },
350
  {
351
- "seconds": 1.1375752950007154,
352
- "bytes": 4905532,
353
- "pages": 33,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.533432411000831,
358
- "bytes": 7357007,
359
- "pages": 30,
360
  "shards": 3
361
  },
362
  {
363
- "seconds": 1.1559075860004668,
364
- "bytes": 1149523,
 
 
 
 
 
 
365
  "pages": 34,
366
- "shards": 1
367
  },
368
  {
369
- "seconds": 1.0760069800016936,
370
- "bytes": 5947734,
371
- "pages": 28,
372
  "shards": 2
373
  },
374
  {
375
- "seconds": 1.2793344790006813,
376
- "bytes": 12370460,
377
- "pages": 40,
378
- "shards": 4
379
  },
380
  {
381
- "seconds": 0.8282303630003298,
382
- "bytes": 647721,
383
- "pages": 20,
384
  "shards": 1
385
  },
386
  {
387
- "seconds": 1.3377682740010641,
388
- "bytes": 13564102,
389
- "pages": 48,
390
- "shards": 4
391
  },
392
  {
393
- "seconds": 1.359001171998898,
394
- "bytes": 1213849,
395
- "pages": 32,
396
  "shards": 1
397
  },
398
  {
399
- "seconds": 1.1420412400002533,
400
- "bytes": 10074433,
401
- "pages": 34,
402
- "shards": 3
403
  },
404
  {
405
- "seconds": 1.157611688000543,
406
- "bytes": 7621104,
407
- "pages": 39,
408
  "shards": 3
409
  },
410
  {
411
- "seconds": 1.0118182780006464,
412
- "bytes": 802933,
413
- "pages": 26,
414
- "shards": 1
415
  },
416
  {
417
- "seconds": 1.0449377190016094,
418
- "bytes": 6509260,
419
- "pages": 28,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.2692650000008143,
424
- "bytes": 13237909,
425
- "pages": 41,
426
  "shards": 4
427
  },
428
  {
429
- "seconds": 1.122997641999973,
430
- "bytes": 6774100,
431
- "pages": 37,
432
- "shards": 2
433
  },
434
  {
435
- "seconds": 2.5239626219990896,
436
- "bytes": 2432130,
437
- "pages": 28,
438
- "shards": 2
439
  },
440
  {
441
- "seconds": 1.0105871389987442,
442
- "bytes": 6118542,
 
 
 
 
 
 
443
  "pages": 36,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 1.3655256990005,
448
- "bytes": 10029580,
449
- "pages": 38,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 0.8411326719997305,
454
- "bytes": 2272552,
455
- "pages": 26,
456
- "shards": 2
457
- },
458
- {
459
- "seconds": 1.4776745150011266,
460
- "bytes": 12319008,
461
- "pages": 37,
462
  "shards": 3
463
  },
464
  {
465
- "seconds": 1.1878520110003592,
466
- "bytes": 6586003,
467
  "pages": 39,
468
  "shards": 3
469
  },
470
  {
471
- "seconds": 4.034771163000187,
472
- "bytes": 5791973,
473
- "pages": 31,
474
- "shards": 2
475
- },
476
- {
477
- "seconds": 1.1227300080008717,
478
- "bytes": 6451625,
479
- "pages": 38,
480
- "shards": 2
481
- },
482
- {
483
- "seconds": 1.4268143489989598,
484
- "bytes": 2671684,
485
- "pages": 37,
486
- "shards": 2
487
  },
488
  {
489
- "seconds": 1.154910495999502,
490
- "bytes": 10565027,
491
- "pages": 23,
492
  "shards": 3
493
  },
494
  {
495
- "seconds": 1.3083442860006471,
496
- "bytes": 8430845,
497
  "pages": 38,
498
  "shards": 3
499
  },
500
  {
501
- "seconds": 1.0249396090002847,
502
- "bytes": 5941897,
503
- "pages": 36,
504
- "shards": 3
505
  },
506
  {
507
- "seconds": 1.2488192199998593,
508
- "bytes": 5819043,
509
  "pages": 28,
510
- "shards": 2
511
- },
512
- {
513
- "seconds": 0.903354562999084,
514
- "bytes": 2640360,
515
- "pages": 37,
516
- "shards": 2
517
- },
518
- {
519
- "seconds": 1.1839327219986444,
520
- "bytes": 10186345,
521
- "pages": 33,
522
  "shards": 3
523
  },
524
  {
525
- "seconds": 3.6534681639986957,
526
- "bytes": 12129367,
527
- "pages": 40,
528
  "shards": 3
529
  },
530
  {
531
- "seconds": 0.9705111450002732,
532
- "bytes": 2305053,
533
- "pages": 28,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 1.3458375579994026,
538
- "bytes": 11053239,
539
- "pages": 36,
540
  "shards": 3
541
  },
542
  {
543
- "seconds": 1.1655933139991248,
544
- "bytes": 2619678,
545
  "pages": 35,
546
- "shards": 2
547
  },
548
  {
549
- "seconds": 1.0188674239998363,
550
- "bytes": 6227206,
551
- "pages": 37,
552
- "shards": 2
553
  },
554
  {
555
- "seconds": 0.9700920809991658,
556
- "bytes": 6347868,
557
- "pages": 36,
558
- "shards": 2
559
  },
560
  {
561
- "seconds": 1.00576553800056,
562
- "bytes": 8295904,
563
- "pages": 33,
 
 
 
 
 
 
564
  "shards": 3
 
 
 
 
 
 
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": 74700.81649595029,
26
+ "required_pages_per_second_upper_estimate": 0.005167975814964294
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": 9337.602061993786,
40
+ "required_pages_per_second_upper_estimate": 0.003909458992052544
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": 18675.204123987573,
54
+ "required_pages_per_second_upper_estimate": 0.003889552520774789
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": 28012.806185981357,
68
+ "required_pages_per_second_upper_estimate": 0.001679324957771178
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": 28012.806185981357,
82
+ "required_pages_per_second_upper_estimate": 0.002177872465647434
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": 28012.806185981357,
96
+ "required_pages_per_second_upper_estimate": 0.35931819994588776
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": 61,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.43565610739844374,
162
+ "producer_p95_seconds": 0.2395410280005308,
163
+ "gpu_update_seconds": 0.9869771638171716,
164
+ "producer_batches_per_second": 2.2953884566696017,
165
+ "gpu_batches_per_second": 1.0131946681850896,
166
+ "producer_headroom": 2.265495988822438,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.205606992847464,
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": 322686.5127973047,
174
+ "upload_bytes_per_second": 5089683.248589828,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 69.2410138887344,
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": 1.2692650000008143,
184
+ "bytes": 13237909,
185
  "pages": 41,
186
  "shards": 4
187
  },
188
  {
189
+ "seconds": 1.122997641999973,
190
+ "bytes": 6774100,
191
+ "pages": 37,
192
  "shards": 2
193
  },
194
  {
195
+ "seconds": 2.5239626219990896,
196
+ "bytes": 2432130,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
197
  "pages": 28,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.0105871389987442,
202
+ "bytes": 6118542,
203
  "pages": 36,
204
  "shards": 2
205
  },
206
  {
207
+ "seconds": 1.3655256990005,
208
+ "bytes": 10029580,
209
+ "pages": 38,
210
+ "shards": 3
211
  },
212
  {
213
+ "seconds": 0.8411326719997305,
214
+ "bytes": 2272552,
215
+ "pages": 26,
216
+ "shards": 2
217
+ },
218
+ {
219
+ "seconds": 1.4776745150011266,
220
+ "bytes": 12319008,
221
+ "pages": 37,
222
+ "shards": 3
223
  },
224
  {
225
+ "seconds": 1.1878520110003592,
226
+ "bytes": 6586003,
227
  "pages": 39,
228
  "shards": 3
229
  },
230
  {
231
+ "seconds": 4.034771163000187,
232
+ "bytes": 5791973,
233
+ "pages": 31,
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.1227300080008717,
238
+ "bytes": 6451625,
239
+ "pages": 38,
240
  "shards": 2
241
  },
242
  {
243
+ "seconds": 1.4268143489989598,
244
+ "bytes": 2671684,
245
+ "pages": 37,
246
  "shards": 2
247
  },
248
  {
249
+ "seconds": 1.154910495999502,
250
+ "bytes": 10565027,
251
+ "pages": 23,
252
+ "shards": 3
253
+ },
254
+ {
255
+ "seconds": 1.3083442860006471,
256
+ "bytes": 8430845,
257
  "pages": 38,
258
  "shards": 3
259
  },
260
  {
261
+ "seconds": 1.0249396090002847,
262
+ "bytes": 5941897,
263
+ "pages": 36,
264
+ "shards": 3
265
+ },
266
+ {
267
+ "seconds": 1.2488192199998593,
268
+ "bytes": 5819043,
269
  "pages": 28,
270
  "shards": 2
271
  },
272
  {
273
+ "seconds": 0.903354562999084,
274
+ "bytes": 2640360,
275
+ "pages": 37,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 1.1839327219986444,
280
+ "bytes": 10186345,
281
+ "pages": 33,
282
+ "shards": 3
283
  },
284
  {
285
+ "seconds": 3.6534681639986957,
286
+ "bytes": 12129367,
287
+ "pages": 40,
288
+ "shards": 3
289
  },
290
  {
291
+ "seconds": 0.9705111450002732,
292
+ "bytes": 2305053,
293
+ "pages": 28,
294
+ "shards": 2
295
+ },
296
+ {
297
+ "seconds": 1.3458375579994026,
298
+ "bytes": 11053239,
299
+ "pages": 36,
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 1.1655933139991248,
304
+ "bytes": 2619678,
305
+ "pages": 35,
306
  "shards": 2
307
  },
308
  {
309
+ "seconds": 1.0188674239998363,
310
+ "bytes": 6227206,
311
+ "pages": 37,
312
  "shards": 2
313
  },
314
  {
315
+ "seconds": 0.9700920809991658,
316
+ "bytes": 6347868,
317
+ "pages": 36,
318
+ "shards": 2
319
  },
320
  {
321
+ "seconds": 1.00576553800056,
322
+ "bytes": 8295904,
323
+ "pages": 33,
324
  "shards": 3
325
  },
326
  {
327
+ "seconds": 0.88294980000137,
328
+ "bytes": 116127,
329
+ "pages": 4,
330
  "shards": 1
331
  },
332
  {
333
+ "seconds": 1.2832525270005135,
334
+ "bytes": 10171575,
335
+ "pages": 37,
336
+ "shards": 3
337
  },
338
  {
339
+ "seconds": 1.4380360830000427,
340
+ "bytes": 7912633,
341
+ "pages": 39,
342
+ "shards": 3
343
  },
344
  {
345
+ "seconds": 0.8056872329998441,
346
+ "bytes": 861106,
347
+ "pages": 26,
348
+ "shards": 1
349
  },
350
  {
351
+ "seconds": 1.1610822420007025,
352
+ "bytes": 10322718,
353
+ "pages": 35,
354
  "shards": 3
355
  },
356
  {
357
+ "seconds": 1.0416573500006052,
358
+ "bytes": 2564522,
359
+ "pages": 36,
360
  "shards": 2
361
  },
362
  {
363
+ "seconds": 1.327567568001541,
364
+ "bytes": 11737942,
365
+ "pages": 34,
366
  "shards": 3
367
  },
368
  {
369
+ "seconds": 1.1044216970003617,
370
+ "bytes": 7366106,
371
+ "pages": 38,
372
+ "shards": 3
373
+ },
374
+ {
375
+ "seconds": 1.023783693999576,
376
+ "bytes": 5395680,
377
  "pages": 34,
378
+ "shards": 2
379
  },
380
  {
381
+ "seconds": 1.439521827000135,
382
+ "bytes": 6360408,
383
+ "pages": 36,
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 1.058846297000855,
388
+ "bytes": 7513643,
389
+ "pages": 31,
390
+ "shards": 3
391
  },
392
  {
393
+ "seconds": 0.8542627569986507,
394
+ "bytes": 1239253,
395
+ "pages": 35,
396
  "shards": 1
397
  },
398
  {
399
+ "seconds": 1.5474993310017453,
400
+ "bytes": 17697706,
401
+ "pages": 35,
402
+ "shards": 5
403
  },
404
  {
405
+ "seconds": 0.7860386789980112,
406
+ "bytes": 1154184,
407
+ "pages": 35,
408
  "shards": 1
409
  },
410
  {
411
+ "seconds": 0.9510616730003676,
412
+ "bytes": 6332480,
413
+ "pages": 36,
414
+ "shards": 2
415
  },
416
  {
417
+ "seconds": 1.1097759929980384,
418
+ "bytes": 7296844,
419
+ "pages": 30,
420
  "shards": 3
421
  },
422
  {
423
+ "seconds": 1.8535761809980613,
424
+ "bytes": 4986805,
425
+ "pages": 35,
426
+ "shards": 2
427
  },
428
  {
429
+ "seconds": 1.1148564830000396,
430
+ "bytes": 6125180,
431
+ "pages": 36,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.2460403479999513,
436
+ "bytes": 13328563,
437
+ "pages": 37,
438
  "shards": 4
439
  },
440
  {
441
+ "seconds": 1.0476313660001324,
442
+ "bytes": 1088716,
443
+ "pages": 31,
444
+ "shards": 1
445
  },
446
  {
447
+ "seconds": 1.4290280249988427,
448
+ "bytes": 12076168,
449
+ "pages": 35,
450
+ "shards": 5
451
  },
452
  {
453
+ "seconds": 1.0162399760010885,
454
+ "bytes": 1245993,
455
+ "pages": 35,
456
+ "shards": 1
457
+ },
458
+ {
459
+ "seconds": 1.0449379370002134,
460
+ "bytes": 6389820,
461
  "pages": 36,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 1.0748847499999101,
466
+ "bytes": 7299524,
467
+ "pages": 30,
468
  "shards": 3
469
  },
470
  {
471
+ "seconds": 3.3975770770011877,
472
+ "bytes": 10836444,
473
+ "pages": 35,
 
 
 
 
 
 
474
  "shards": 3
475
  },
476
  {
477
+ "seconds": 1.1278280619990255,
478
+ "bytes": 7877906,
479
  "pages": 39,
480
  "shards": 3
481
  },
482
  {
483
+ "seconds": 1.147284323000349,
484
+ "bytes": 1234396,
485
+ "pages": 34,
486
+ "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
487
  },
488
  {
489
+ "seconds": 1.2498522079986287,
490
+ "bytes": 10921992,
491
+ "pages": 31,
492
  "shards": 3
493
  },
494
  {
495
+ "seconds": 1.2837640949983324,
496
+ "bytes": 7864429,
497
  "pages": 38,
498
  "shards": 3
499
  },
500
  {
501
+ "seconds": 0.8372707349990378,
502
+ "bytes": 1184239,
503
+ "pages": 35,
504
+ "shards": 1
505
  },
506
  {
507
+ "seconds": 1.1600354100010009,
508
+ "bytes": 11531600,
509
  "pages": 28,
 
 
 
 
 
 
 
 
 
 
 
 
510
  "shards": 3
511
  },
512
  {
513
+ "seconds": 1.4267704899975797,
514
+ "bytes": 7513990,
515
+ "pages": 38,
516
  "shards": 3
517
  },
518
  {
519
+ "seconds": 1.0001321140007349,
520
+ "bytes": 5054764,
521
+ "pages": 38,
522
  "shards": 2
523
  },
524
  {
525
+ "seconds": 1.5159087600004568,
526
+ "bytes": 7203777,
527
+ "pages": 29,
528
  "shards": 3
529
  },
530
  {
531
+ "seconds": 1.1342419799984782,
532
+ "bytes": 1205083,
533
  "pages": 35,
534
+ "shards": 1
535
  },
536
  {
537
+ "seconds": 1.2808730480028316,
538
+ "bytes": 10383198,
539
+ "pages": 27,
540
+ "shards": 3
541
  },
542
  {
543
+ "seconds": 1.2442956449995108,
544
+ "bytes": 13618701,
545
+ "pages": 45,
546
+ "shards": 4
547
  },
548
  {
549
+ "seconds": 0.9636815740013844,
550
+ "bytes": 1058213,
551
+ "pages": 34,
552
+ "shards": 1
553
+ },
554
+ {
555
+ "seconds": 1.1714537059997383,
556
+ "bytes": 7311656,
557
+ "pages": 31,
558
  "shards": 3
559
+ },
560
+ {
561
+ "seconds": 2.4602762009999424,
562
+ "bytes": 844768,
563
+ "pages": 28,
564
+ "shards": 1
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9933156547ab18b657069aa60eb52e3b669e25faa068a6484883abec26f0f1a7
3
- size 721780991
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:08c9cb9f6f17586214603f83e8a6ed392e1f2c67500da7bfa88fe516501b5bdf
3
+ size 721760319
training_config.json CHANGED
@@ -53,8 +53,8 @@
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,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 61,
57
+ "prefetch_low_water": 30,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
60
  "force_rebenchmark": false,