appvoid commited on
Commit
72513dc
·
verified ·
1 Parent(s): 7fae8ea

Qwen90M step 70000; 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 +239 -239
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 65000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "e33da1dad32ae2433103ec21715c2d542c98dc7e7caa2f82997e62b6fe8342d2",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "e3e3dd01665bea5cf593b73f4417e92f7e07fdb4e1949ec1cfe309333ab911ee",
12
- "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f",
14
- "shard_scheduler.json": "fb85ef21cda081906908c01fe4de71830aa788514ed6534050e7da541b10e3e2",
15
- "coverage.json": "f8f337b02bacae59e68d5dbc5eb4b6339c704cf53f376b91e8950e701871a03e",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 70000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "1c5054d917de2693f770bc99e564913ba6cc2edd42ed1e61bc5076094a4f4cc1",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "7ca837f5b08d05bbf41cfc5f430f7d7929aa5dfd920fa6c66696541123b2d2a1",
12
+ "training_config.json": "a9be8fd9bf1c8d33aae0c56259fc899809df0d3ff60285ef0a4c2ea73471632c",
13
  "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f",
14
+ "shard_scheduler.json": "eb1e6871a1097af2c4c674b06e837fee7074d58979d436eda419016044a46189",
15
+ "coverage.json": "1b3adc06f707cf80a7d00bfb046d62b635ee1b8cd68ef5217768a4182ba6132e",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 65000,
3
  "stats": {
4
- "climbmix/blocks": 2340000,
5
- "climbmix/targets": 3910377700,
6
- "climbmix/input_tokens": 3910625311,
7
- "ultra_qa/blocks": 877500,
8
- "ultra_qa/targets": 1431685811,
9
- "ultra_qa/input_tokens": 1431688622,
10
- "cortex/blocks": 877500,
11
- "cortex/targets": 33383362,
12
- "cortex/input_tokens": 1394824403,
13
- "ultra_style/blocks": 877500,
14
- "ultra_style/targets": 1387878431,
15
- "ultra_style/input_tokens": 1387950984,
16
- "openmath/blocks": 585000,
17
- "openmath/targets": 903936284,
18
- "openmath/input_tokens": 904237672,
19
- "rewrite6/blocks": 292500,
20
- "rewrite6/targets": 479420381,
21
- "rewrite6/input_tokens": 479946209
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 320,
26
- "row": 1789,
27
  "start": {
28
- "file": 15,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 39,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 2048,
37
- "documents": 1312509
38
  },
39
  "rewrite6": {
40
- "page": 32,
41
- "row": 1861,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 131072
47
  }
48
  },
49
  "epoch": 3,
50
  "offset": 0,
51
- "documents": 749145
52
  },
53
  "openmath": {
54
- "page": 200,
55
- "row": 1224,
56
  "start": {
57
- "file": 1,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 382,
61
- "row": 550
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 820424
67
  },
68
  "ultra_qa": {
69
- "page": 86,
70
- "row": 2631,
71
  "start": {
72
  "file": 0,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 5,
76
- "row": 29519
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 354887
82
  },
83
  "ultra_style": {
84
- "page": 132,
85
- "row": 470,
86
  "start": {
87
  "file": 0,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 6,
91
- "row": 30335
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 2048,
96
- "documents": 541142
97
  },
98
  "cortex": {
99
- "page": 17014,
100
- "row": 1,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 2177793
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 70000,
3
  "stats": {
4
+ "climbmix/blocks": 2520000,
5
+ "climbmix/targets": 4211257891,
6
+ "climbmix/input_tokens": 4211524900,
7
+ "ultra_qa/blocks": 945000,
8
+ "ultra_qa/targets": 1541761818,
9
+ "ultra_qa/input_tokens": 1541764859,
10
+ "cortex/blocks": 945000,
11
+ "cortex/targets": 35955650,
12
+ "cortex/input_tokens": 1502146727,
13
+ "ultra_style/blocks": 945000,
14
+ "ultra_style/targets": 1494609054,
15
+ "ultra_style/input_tokens": 1494687200,
16
+ "openmath/blocks": 630000,
17
+ "openmath/targets": 973552847,
18
+ "openmath/input_tokens": 973877515,
19
+ "rewrite6/blocks": 315000,
20
+ "rewrite6/targets": 516437914,
21
+ "rewrite6/input_tokens": 517004193
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 345,
26
+ "row": 1005,
27
  "start": {
28
+ "file": 16,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 56,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 10240,
37
+ "documents": 1414125
38
  },
39
  "rewrite6": {
40
+ "page": 46,
41
+ "row": 2052,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 188416
47
  }
48
  },
49
  "epoch": 3,
50
  "offset": 0,
51
+ "documents": 806680
52
  },
53
  "openmath": {
54
+ "page": 215,
55
+ "row": 3124,
56
  "start": {
57
+ "file": 2,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 7,
61
+ "row": 340
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 883764
67
  },
68
  "ultra_qa": {
69
+ "page": 93,
70
+ "row": 1286,
71
  "start": {
72
  "file": 0,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 5,
76
+ "row": 58191
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 382214
82
  },
83
  "ultra_style": {
84
+ "page": 142,
85
+ "row": 1102,
86
  "start": {
87
  "file": 0,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 6,
91
+ "row": 71295
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 0,
96
+ "documents": 582734
97
  },
98
  "cortex": {
99
+ "page": 18324,
100
+ "row": 41,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 2345513
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 65000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "e33da1dad32ae2433103ec21715c2d542c98dc7e7caa2f82997e62b6fe8342d2", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "e3e3dd01665bea5cf593b73f4417e92f7e07fdb4e1949ec1cfe309333ab911ee", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "fb85ef21cda081906908c01fe4de71830aa788514ed6534050e7da541b10e3e2", "coverage.json": "f8f337b02bacae59e68d5dbc5eb4b6339c704cf53f376b91e8950e701871a03e", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000065000"}
 
1
+ {"step": 70000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "1c5054d917de2693f770bc99e564913ba6cc2edd42ed1e61bc5076094a4f4cc1", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "7ca837f5b08d05bbf41cfc5f430f7d7929aa5dfd920fa6c66696541123b2d2a1", "training_config.json": "a9be8fd9bf1c8d33aae0c56259fc899809df0d3ff60285ef0a4c2ea73471632c", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "eb1e6871a1097af2c4c674b06e837fee7074d58979d436eda419016044a46189", "coverage.json": "1b3adc06f707cf80a7d00bfb046d62b635ee1b8cd68ef5217768a4182ba6132e", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000070000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e33da1dad32ae2433103ec21715c2d542c98dc7e7caa2f82997e62b6fe8342d2
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c5054d917de2693f770bc99e564913ba6cc2edd42ed1e61bc5076094a4f4cc1
3
  size 360622744
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 75748.15161257412,
26
- "required_pages_per_second_upper_estimate": 0.006565657799842846
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
36
  "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
- "required_token_positions_per_second": 9468.518951571765,
40
- "required_pages_per_second_upper_estimate": 0.0037771337767559298
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
50
  "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
- "required_token_positions_per_second": 18937.03790314353,
54
- "required_pages_per_second_upper_estimate": 0.0043807504384081025
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
64
  "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
- "required_token_positions_per_second": 28405.556854715294,
68
- "required_pages_per_second_upper_estimate": 0.0017297493800131785
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
78
  "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
- "required_token_positions_per_second": 28405.556854715294,
82
- "required_pages_per_second_upper_estimate": 0.0021867383982900044
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.042780299000014566,
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 22,
95
- "required_token_positions_per_second": 28405.556854715294,
96
- "required_pages_per_second_upper_estimate": 0.3527019488522703
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": 0.1434574083060457,
157
  "selected": {
158
- "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
- "producer_batch_seconds": 0.11656077062494319,
162
- "producer_p95_seconds": 0.29636224299974856,
163
- "gpu_update_seconds": 0.9733306810850448,
164
- "producer_batches_per_second": 8.579215757055119,
165
- "gpu_batches_per_second": 1.0274000598493669,
166
- "producer_headroom": 8.350413915990007,
167
  "sustainable": true,
168
- "buffer_seconds": 60.34650222727278,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
- "upload_pages_per_commit": 86,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 512873.7689111375,
174
- "upload_bytes_per_second": 4674834.667993678,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 110.05077171803086,
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": 1.406286874000216,
184
- "bytes": 18261180,
185
- "pages": 44,
186
- "shards": 5
187
- },
188
- {
189
- "seconds": 1.0190579750014876,
190
- "bytes": 5885831,
191
  "pages": 35,
192
  "shards": 2
193
  },
194
  {
195
- "seconds": 0.8101317319997179,
196
- "bytes": 713378,
197
- "pages": 25,
198
- "shards": 2
199
  },
200
  {
201
- "seconds": 1.03739196099923,
202
- "bytes": 7802740,
203
- "pages": 38,
204
- "shards": 3
205
  },
206
  {
207
- "seconds": 1.1620692619981128,
208
- "bytes": 6079962,
209
- "pages": 34,
210
- "shards": 2
211
  },
212
  {
213
- "seconds": 0.8016836730021168,
214
- "bytes": 4650466,
215
- "pages": 32,
216
  "shards": 2
217
  },
218
  {
219
- "seconds": 1.2261960830001044,
220
- "bytes": 13273679,
221
- "pages": 37,
222
- "shards": 4
223
  },
224
  {
225
- "seconds": 0.7825120570014406,
226
- "bytes": 1093241,
227
- "pages": 34,
228
- "shards": 1
229
  },
230
  {
231
- "seconds": 1.4853517330011528,
232
- "bytes": 11092996,
233
- "pages": 24,
234
- "shards": 4
235
  },
236
  {
237
- "seconds": 0.8748619340003643,
238
- "bytes": 6384371,
239
- "pages": 42,
240
- "shards": 2
241
  },
242
  {
243
- "seconds": 0.9224505880010838,
244
- "bytes": 1117951,
245
  "pages": 34,
246
  "shards": 1
247
  },
248
  {
249
- "seconds": 1.0108905939996475,
250
- "bytes": 7845615,
251
- "pages": 36,
252
  "shards": 3
253
  },
254
  {
255
- "seconds": 1.27624071000173,
256
- "bytes": 16075425,
257
- "pages": 35,
258
- "shards": 4
259
  },
260
  {
261
- "seconds": 0.8457192399982887,
262
- "bytes": 2628229,
263
  "pages": 36,
264
  "shards": 2
265
  },
266
  {
267
- "seconds": 0.9226963619985327,
268
- "bytes": 5611919,
269
- "pages": 27,
270
  "shards": 2
271
  },
272
  {
273
- "seconds": 0.7924021649996575,
274
- "bytes": 1122256,
275
- "pages": 34,
276
- "shards": 1
277
  },
278
  {
279
- "seconds": 1.1671633029982331,
280
- "bytes": 11363880,
281
- "pages": 40,
282
- "shards": 4
283
  },
284
  {
285
- "seconds": 0.9065282010014926,
286
- "bytes": 6177542,
287
- "pages": 28,
288
  "shards": 2
289
  },
290
  {
291
- "seconds": 1.0971184539994283,
292
- "bytes": 8482200,
293
- "pages": 38,
294
- "shards": 3
295
  },
296
  {
297
- "seconds": 1.1112909670009685,
298
- "bytes": 6011206,
299
- "pages": 35,
300
- "shards": 2
301
  },
302
  {
303
- "seconds": 0.9119941530007054,
304
- "bytes": 5747516,
305
- "pages": 29,
306
- "shards": 2
307
  },
308
  {
309
- "seconds": 1.016742787000112,
310
- "bytes": 7467282,
311
  "pages": 37,
312
  "shards": 3
313
  },
314
  {
315
- "seconds": 0.8858778819994768,
316
- "bytes": 5985432,
317
- "pages": 27,
318
  "shards": 2
319
  },
320
  {
321
- "seconds": 0.7830291239988583,
322
- "bytes": 1207474,
323
- "pages": 35,
324
  "shards": 1
325
  },
326
  {
327
- "seconds": 1.2103096939972602,
328
- "bytes": 12749523,
329
- "pages": 22,
330
- "shards": 4
331
  },
332
  {
333
- "seconds": 0.8322358400000667,
334
- "bytes": 4694947,
335
- "pages": 29,
336
- "shards": 2
337
  },
338
  {
339
- "seconds": 0.8593272979996982,
340
- "bytes": 6225685,
341
  "pages": 36,
342
  "shards": 2
343
  },
344
  {
345
- "seconds": 0.9578338470018934,
346
- "bytes": 2306768,
 
 
 
 
 
 
347
  "pages": 27,
348
  "shards": 2
349
  },
350
  {
351
- "seconds": 0.8912073430001328,
352
- "bytes": 6007286,
353
- "pages": 32,
354
- "shards": 2
 
 
 
 
 
 
355
  },
356
  {
357
- "seconds": 1.088943154001754,
358
- "bytes": 10590685,
359
  "pages": 36,
360
- "shards": 3
361
  },
362
  {
363
- "seconds": 0.9887308050019783,
364
- "bytes": 8510426,
365
- "pages": 40,
366
  "shards": 3
367
  },
368
  {
369
- "seconds": 1.0075569920009002,
370
- "bytes": 5909692,
371
- "pages": 30,
372
- "shards": 2
373
  },
374
  {
375
- "seconds": 0.7540812590013957,
376
- "bytes": 2508470,
377
- "pages": 31,
378
  "shards": 2
379
  },
380
  {
381
- "seconds": 1.3975003769992327,
382
- "bytes": 10375015,
383
- "pages": 34,
384
- "shards": 4
385
- },
386
- {
387
- "seconds": 1.0634488309988228,
388
- "bytes": 6386527,
389
  "pages": 36,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 14.660640815000079,
394
- "bytes": 2712693,
395
- "pages": 36,
396
  "shards": 2
397
  },
398
  {
399
- "seconds": 3.9525413480005227,
400
- "bytes": 15333188,
401
  "pages": 34,
402
  "shards": 4
403
  },
404
  {
405
- "seconds": 2.024789704999421,
406
- "bytes": 1170220,
407
- "pages": 34,
408
  "shards": 1
409
  },
410
  {
411
- "seconds": 1.2876270810011192,
412
- "bytes": 7494644,
413
- "pages": 37,
414
  "shards": 3
415
  },
416
  {
417
- "seconds": 0.8472818170012033,
418
- "bytes": 6269698,
419
- "pages": 35,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 0.9710190369987686,
424
- "bytes": 6743419,
425
- "pages": 30,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 1.0164730639990012,
430
- "bytes": 6038957,
431
- "pages": 35,
432
- "shards": 2
433
  },
434
  {
435
- "seconds": 0.9485057740021148,
436
- "bytes": 880447,
437
  "pages": 26,
438
  "shards": 1
439
  },
440
  {
441
- "seconds": 1.0207911869983946,
442
- "bytes": 13578331,
443
- "pages": 42,
444
- "shards": 4
445
  },
446
  {
447
- "seconds": 1.2704945900004532,
448
- "bytes": 9798343,
449
- "pages": 31,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 0.7868170459987596,
454
- "bytes": 2512813,
455
- "pages": 36,
456
- "shards": 2
457
  },
458
  {
459
- "seconds": 1.0122808699998131,
460
- "bytes": 6141256,
461
  "pages": 36,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 0.8550289739978325,
466
- "bytes": 5835711,
467
  "pages": 28,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.051623090999783,
472
- "bytes": 7071478,
473
- "pages": 33,
474
  "shards": 3
475
  },
476
  {
477
- "seconds": 1.1867464049973933,
478
- "bytes": 12128568,
479
- "pages": 40,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 0.8681122560010408,
484
- "bytes": 1227302,
485
- "pages": 34,
486
  "shards": 1
487
  },
488
  {
489
- "seconds": 1.205836907000048,
490
- "bytes": 7335325,
491
- "pages": 30,
492
  "shards": 3
493
  },
494
  {
495
- "seconds": 1.085818534000282,
496
- "bytes": 10112362,
497
- "pages": 32,
498
- "shards": 3
499
  },
500
  {
501
- "seconds": 0.927039637001144,
502
- "bytes": 2682847,
503
- "pages": 36,
504
- "shards": 2
505
  },
506
  {
507
- "seconds": 0.8612236650005798,
508
- "bytes": 6230072,
509
- "pages": 34,
510
- "shards": 2
511
  },
512
  {
513
- "seconds": 1.0601522930010105,
514
- "bytes": 12072218,
515
- "pages": 37,
516
- "shards": 3
517
  },
518
  {
519
- "seconds": 1.126622554998903,
520
- "bytes": 6643361,
521
  "pages": 34,
522
  "shards": 3
523
  },
524
  {
525
- "seconds": 0.8856843470020976,
526
- "bytes": 5828359,
527
- "pages": 27,
528
- "shards": 2
529
  },
530
  {
531
- "seconds": 1.0851010650003445,
532
- "bytes": 1256653,
533
- "pages": 34,
534
  "shards": 1
535
  },
536
  {
537
- "seconds": 1.3800626919983188,
538
- "bytes": 7827681,
539
- "pages": 39,
540
- "shards": 3
541
  },
542
  {
543
- "seconds": 1.5110167620005086,
544
- "bytes": 10255842,
545
- "pages": 33,
546
  "shards": 3
547
  },
548
  {
549
- "seconds": 8.627843940001185,
550
- "bytes": 8388512,
551
- "pages": 37,
 
 
 
 
 
 
552
  "shards": 3
553
  },
554
  {
555
- "seconds": 1.2330220080002618,
556
- "bytes": 6137294,
557
  "pages": 36,
558
  "shards": 2
559
- },
560
- {
561
- "seconds": 1.6848288789988146,
562
- "bytes": 823777,
563
- "pages": 25,
564
- "shards": 1
565
  }
566
  ],
567
  "cache_key": {
 
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 73286.81074761767,
26
+ "required_pages_per_second_upper_estimate": 0.006352315011879811
27
  },
28
  {
29
  "source": "rewrite6",
 
36
  "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
+ "required_token_positions_per_second": 9160.851343452208,
40
+ "required_pages_per_second_upper_estimate": 0.0036544005678363686
41
  },
42
  {
43
  "source": "openmath",
 
50
  "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
+ "required_token_positions_per_second": 18321.702686904417,
54
+ "required_pages_per_second_upper_estimate": 0.004238403465661115
55
  },
56
  {
57
  "source": "ultra_qa",
 
64
  "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
+ "required_token_positions_per_second": 27482.554030356623,
68
+ "required_pages_per_second_upper_estimate": 0.0016735433506312972
69
  },
70
  {
71
  "source": "ultra_style",
 
78
  "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
+ "required_token_positions_per_second": 27482.554030356623,
82
+ "required_pages_per_second_upper_estimate": 0.0021156830858355268
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.042780299000014566,
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 21,
95
+ "required_token_positions_per_second": 27482.554030356623,
96
+ "required_pages_per_second_upper_estimate": 0.3412413428654733
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": 0.1434574083060457,
157
  "selected": {
158
+ "prefetch_batches": 60,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
+ "producer_batch_seconds": 0.43211702229669413,
162
+ "producer_p95_seconds": 0.33367446499687503,
163
+ "gpu_update_seconds": 1.0060200361822496,
164
+ "producer_batches_per_second": 2.314187936140581,
165
+ "gpu_batches_per_second": 0.994015987787783,
166
+ "producer_headroom": 2.328119431248673,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.36120217093497,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 79,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 266264.8417269639,
174
+ "upload_bytes_per_second": 3764138.0954640657,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 57.13423670632064,
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": 1.0164730639990012,
184
+ "bytes": 6038957,
 
 
 
 
 
 
185
  "pages": 35,
186
  "shards": 2
187
  },
188
  {
189
+ "seconds": 0.9485057740021148,
190
+ "bytes": 880447,
191
+ "pages": 26,
192
+ "shards": 1
193
  },
194
  {
195
+ "seconds": 1.0207911869983946,
196
+ "bytes": 13578331,
197
+ "pages": 42,
198
+ "shards": 4
199
  },
200
  {
201
+ "seconds": 1.2704945900004532,
202
+ "bytes": 9798343,
203
+ "pages": 31,
204
+ "shards": 3
205
  },
206
  {
207
+ "seconds": 0.7868170459987596,
208
+ "bytes": 2512813,
209
+ "pages": 36,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.0122808699998131,
214
+ "bytes": 6141256,
215
+ "pages": 36,
216
+ "shards": 2
217
  },
218
  {
219
+ "seconds": 0.8550289739978325,
220
+ "bytes": 5835711,
221
+ "pages": 28,
222
+ "shards": 2
223
  },
224
  {
225
+ "seconds": 1.051623090999783,
226
+ "bytes": 7071478,
227
+ "pages": 33,
228
+ "shards": 3
229
  },
230
  {
231
+ "seconds": 1.1867464049973933,
232
+ "bytes": 12128568,
233
+ "pages": 40,
234
+ "shards": 3
235
  },
236
  {
237
+ "seconds": 0.8681122560010408,
238
+ "bytes": 1227302,
239
  "pages": 34,
240
  "shards": 1
241
  },
242
  {
243
+ "seconds": 1.205836907000048,
244
+ "bytes": 7335325,
245
+ "pages": 30,
246
  "shards": 3
247
  },
248
  {
249
+ "seconds": 1.085818534000282,
250
+ "bytes": 10112362,
251
+ "pages": 32,
252
+ "shards": 3
253
  },
254
  {
255
+ "seconds": 0.927039637001144,
256
+ "bytes": 2682847,
257
  "pages": 36,
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 0.8612236650005798,
262
+ "bytes": 6230072,
263
+ "pages": 34,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.0601522930010105,
268
+ "bytes": 12072218,
269
+ "pages": 37,
270
+ "shards": 3
271
  },
272
  {
273
+ "seconds": 1.126622554998903,
274
+ "bytes": 6643361,
275
+ "pages": 34,
276
+ "shards": 3
277
  },
278
  {
279
+ "seconds": 0.8856843470020976,
280
+ "bytes": 5828359,
281
+ "pages": 27,
282
  "shards": 2
283
  },
284
  {
285
+ "seconds": 1.0851010650003445,
286
+ "bytes": 1256653,
287
+ "pages": 34,
288
+ "shards": 1
289
  },
290
  {
291
+ "seconds": 1.3800626919983188,
292
+ "bytes": 7827681,
293
+ "pages": 39,
294
+ "shards": 3
295
  },
296
  {
297
+ "seconds": 1.5110167620005086,
298
+ "bytes": 10255842,
299
+ "pages": 33,
300
+ "shards": 3
301
  },
302
  {
303
+ "seconds": 8.627843940001185,
304
+ "bytes": 8388512,
305
  "pages": 37,
306
  "shards": 3
307
  },
308
  {
309
+ "seconds": 1.2330220080002618,
310
+ "bytes": 6137294,
311
+ "pages": 36,
312
  "shards": 2
313
  },
314
  {
315
+ "seconds": 1.6848288789988146,
316
+ "bytes": 823777,
317
+ "pages": 25,
318
  "shards": 1
319
  },
320
  {
321
+ "seconds": 1.0912059820002469,
322
+ "bytes": 11909059,
323
+ "pages": 36,
324
+ "shards": 5
325
  },
326
  {
327
+ "seconds": 1.1786047089990461,
328
+ "bytes": 318780,
329
+ "pages": 9,
330
+ "shards": 1
331
  },
332
  {
333
+ "seconds": 0.9711113170014869,
334
+ "bytes": 6125935,
335
  "pages": 36,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 1.0254750570020406,
340
+ "bytes": 8509608,
341
+ "pages": 38,
342
+ "shards": 3
343
+ },
344
+ {
345
+ "seconds": 15.209589787002187,
346
+ "bytes": 6010603,
347
  "pages": 27,
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 1.505151831999683,
352
+ "bytes": 12552564,
353
+ "pages": 42,
354
+ "shards": 4
355
+ },
356
+ {
357
+ "seconds": 13.275076857000386,
358
+ "bytes": 1045844,
359
+ "pages": 30,
360
+ "shards": 1
361
  },
362
  {
363
+ "seconds": 2.272034231998987,
364
+ "bytes": 5974861,
365
  "pages": 36,
366
+ "shards": 2
367
  },
368
  {
369
+ "seconds": 1.2107169869996142,
370
+ "bytes": 7639876,
371
+ "pages": 38,
372
  "shards": 3
373
  },
374
  {
375
+ "seconds": 1.2838386629991874,
376
+ "bytes": 10633493,
377
+ "pages": 36,
378
+ "shards": 3
379
  },
380
  {
381
+ "seconds": 1.5226552149979398,
382
+ "bytes": 6060968,
383
+ "pages": 28,
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 1.020731942000566,
388
+ "bytes": 2578449,
 
 
 
 
 
 
389
  "pages": 36,
390
  "shards": 2
391
  },
392
  {
393
+ "seconds": 1.7385529189996305,
394
+ "bytes": 5799908,
395
+ "pages": 35,
396
  "shards": 2
397
  },
398
  {
399
+ "seconds": 1.6018576979986392,
400
+ "bytes": 11859045,
401
  "pages": 34,
402
  "shards": 4
403
  },
404
  {
405
+ "seconds": 0.8893314249980904,
406
+ "bytes": 1064415,
407
+ "pages": 30,
408
  "shards": 1
409
  },
410
  {
411
+ "seconds": 1.0851846760015178,
412
+ "bytes": 11839223,
413
+ "pages": 39,
414
  "shards": 3
415
  },
416
  {
417
+ "seconds": 1.710380387998157,
418
+ "bytes": 2330946,
419
+ "pages": 28,
420
  "shards": 2
421
  },
422
  {
423
+ "seconds": 1.155976900998212,
424
+ "bytes": 10306222,
425
+ "pages": 38,
426
  "shards": 3
427
  },
428
  {
429
+ "seconds": 1.1758935670004576,
430
+ "bytes": 7725078,
431
+ "pages": 38,
432
+ "shards": 3
433
  },
434
  {
435
+ "seconds": 1.0130331639993528,
436
+ "bytes": 839053,
437
  "pages": 26,
438
  "shards": 1
439
  },
440
  {
441
+ "seconds": 1.1233831270001247,
442
+ "bytes": 6290212,
443
+ "pages": 36,
444
+ "shards": 2
445
  },
446
  {
447
+ "seconds": 0.8791882089972205,
448
+ "bytes": 6587099,
449
+ "pages": 27,
450
  "shards": 3
451
  },
452
  {
453
+ "seconds": 1.0528832150012022,
454
+ "bytes": 11948549,
455
+ "pages": 44,
456
+ "shards": 3
457
  },
458
  {
459
+ "seconds": 1.0517111139997724,
460
+ "bytes": 6085549,
461
  "pages": 36,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 0.9597967700028676,
466
+ "bytes": 2343290,
467
  "pages": 28,
468
  "shards": 2
469
  },
470
  {
471
+ "seconds": 1.0771120079989487,
472
+ "bytes": 10854426,
473
+ "pages": 39,
474
  "shards": 3
475
  },
476
  {
477
+ "seconds": 1.2815551900021092,
478
+ "bytes": 7356201,
479
+ "pages": 29,
480
  "shards": 3
481
  },
482
  {
483
+ "seconds": 1.2021151590015506,
484
+ "bytes": 960345,
485
+ "pages": 28,
486
  "shards": 1
487
  },
488
  {
489
+ "seconds": 3.8789173690020107,
490
+ "bytes": 11941335,
491
+ "pages": 36,
492
  "shards": 3
493
  },
494
  {
495
+ "seconds": 1.0000528349992237,
496
+ "bytes": 11258395,
497
+ "pages": 42,
498
+ "shards": 4
499
  },
500
  {
501
+ "seconds": 0.9514743170002475,
502
+ "bytes": 822769,
503
+ "pages": 26,
504
+ "shards": 1
505
  },
506
  {
507
+ "seconds": 4.258936992002418,
508
+ "bytes": 6104340,
509
+ "pages": 36,
510
+ "shards": 3
511
  },
512
  {
513
+ "seconds": 1.0792534860011074,
514
+ "bytes": 2442489,
515
+ "pages": 31,
516
+ "shards": 2
517
  },
518
  {
519
+ "seconds": 1.0146635470009642,
520
+ "bytes": 11041931,
521
  "pages": 34,
522
  "shards": 3
523
  },
524
  {
525
+ "seconds": 1.0869366869992518,
526
+ "bytes": 13432151,
527
+ "pages": 40,
528
+ "shards": 4
529
  },
530
  {
531
+ "seconds": 0.7862899389983795,
532
+ "bytes": 1239440,
533
+ "pages": 35,
534
  "shards": 1
535
  },
536
  {
537
+ "seconds": 0.8729745059972629,
538
+ "bytes": 5859266,
539
+ "pages": 29,
540
+ "shards": 2
541
  },
542
  {
543
+ "seconds": 1.0330235869987519,
544
+ "bytes": 6506267,
545
+ "pages": 40,
546
  "shards": 3
547
  },
548
  {
549
+ "seconds": 1.471190406999085,
550
+ "bytes": 6116938,
551
+ "pages": 27,
552
+ "shards": 2
553
+ },
554
+ {
555
+ "seconds": 1.1064680679992307,
556
+ "bytes": 7358899,
557
+ "pages": 34,
558
  "shards": 3
559
  },
560
  {
561
+ "seconds": 1.2355712200005655,
562
+ "bytes": 6983067,
563
  "pages": 36,
564
  "shards": 2
 
 
 
 
 
 
565
  }
566
  ],
567
  "cache_key": {
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e3e3dd01665bea5cf593b73f4417e92f7e07fdb4e1949ec1cfe309333ab911ee
3
- size 721787647
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ca837f5b08d05bbf41cfc5f430f7d7929aa5dfd920fa6c66696541123b2d2a1
3
+ size 721775935
training_config.json CHANGED
@@ -46,8 +46,8 @@
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,
53
  "force_rebenchmark": false,
 
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
+ "prefetch_batches": 60,
50
+ "prefetch_low_water": 30,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,