appvoid commited on
Commit
eba3be6
·
verified ·
1 Parent(s): 93dfc3a

Qwen90M step 80000; complete resumable checkpoint

Browse files
Files changed (7) hide show
  1. complete.json +6 -6
  2. coverage.json +46 -46
  3. latest.json +1 -1
  4. model.safetensors +1 -1
  5. shard_scheduler.json +245 -245
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 75000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "b90a19c4b2d7929e861313731bcb5beb014b3e36b39789a0beda7568ce6a846a",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "5026bd4bb1b3b18f5fc726c01908ba56660bc127713bb70b791880fb27548eeb",
12
- "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
13
  "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f",
14
- "shard_scheduler.json": "a365a3ff77a3205ac0d3c410749b4fb4670618f5a7de9659698fe71360225cc6",
15
- "coverage.json": "2dd27baaf4f3a0abac91b8368d3639578f98d0308ab7fb49afd5e4690f165c2f",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 80000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "016fae2e58ad750ac0b4abde700bbdbea9199273ac137f7e182d0fe7874da0dc",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "63c9c8c99731fc20714a54a44377517fe0f1f3043beea6e7f6b8a0c370badb1e",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
  "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f",
14
+ "shard_scheduler.json": "ea08386ca957ff57a36e47f5f3cdc24c23475683922434b0de7b9c8abf2744b1",
15
+ "coverage.json": "42736094b47b18e0027534773acb49639d163b78174444895fdb7d300580166a",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 75000,
3
  "stats": {
4
- "climbmix/blocks": 2700000,
5
- "climbmix/targets": 4511757125,
6
- "climbmix/input_tokens": 4512042972,
7
- "ultra_qa/blocks": 1012500,
8
- "ultra_qa/targets": 1651705822,
9
- "ultra_qa/input_tokens": 1651709076,
10
- "cortex/blocks": 1012500,
11
- "cortex/targets": 38520571,
12
- "cortex/input_tokens": 1609425947,
13
- "ultra_style/blocks": 1012500,
14
- "ultra_style/targets": 1601103718,
15
- "ultra_style/input_tokens": 1601187526,
16
- "openmath/blocks": 675000,
17
- "openmath/targets": 1043094503,
18
- "openmath/input_tokens": 1043442386,
19
- "rewrite6/blocks": 337500,
20
- "rewrite6/targets": 553275354,
21
- "rewrite6/input_tokens": 553881803
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 369,
26
- "row": 3716,
27
  "start": {
28
- "file": 17,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 69,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 0,
37
- "documents": 1515140
38
  },
39
  "rewrite6": {
40
- "page": 10,
41
- "row": 1413,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 40960
47
  }
48
  },
49
  "epoch": 4,
50
  "offset": 2048,
51
- "documents": 863989
52
  },
53
  "openmath": {
54
- "page": 231,
55
- "row": 700,
56
  "start": {
57
  "file": 2,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 72,
61
- "row": 876
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 946876
67
  },
68
  "ultra_qa": {
69
- "page": 99,
70
- "row": 4047,
71
  "start": {
72
  "file": 0,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
- "row": 17989
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 409551
82
  },
83
  "ultra_style": {
84
- "page": 152,
85
- "row": 2035,
86
  "start": {
87
  "file": 0,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 7,
91
- "row": 27468
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 624627
97
  },
98
  "cortex": {
99
- "page": 19633,
100
- "row": 27,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 2513051
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 80000,
3
  "stats": {
4
+ "climbmix/blocks": 2880000,
5
+ "climbmix/targets": 4812534066,
6
+ "climbmix/input_tokens": 4812838946,
7
+ "ultra_qa/blocks": 1080000,
8
+ "ultra_qa/targets": 1761872361,
9
+ "ultra_qa/input_tokens": 1761875805,
10
+ "cortex/blocks": 1080000,
11
+ "cortex/targets": 41089673,
12
+ "cortex/input_tokens": 1716618096,
13
+ "ultra_style/blocks": 1080000,
14
+ "ultra_style/targets": 1707744164,
15
+ "ultra_style/input_tokens": 1707833604,
16
+ "openmath/blocks": 720000,
17
+ "openmath/targets": 1112712435,
18
+ "openmath/input_tokens": 1113083661,
19
+ "rewrite6/blocks": 360000,
20
+ "rewrite6/targets": 590150425,
21
+ "rewrite6/input_tokens": 590797964
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 394,
26
+ "row": 2302,
27
  "start": {
28
+ "file": 19,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 3,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 1616126
38
  },
39
  "rewrite6": {
40
+ "page": 24,
41
+ "row": 2442,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 98304
47
  }
48
  },
49
  "epoch": 4,
50
  "offset": 2048,
51
+ "documents": 922362
52
  },
53
  "openmath": {
54
+ "page": 246,
55
+ "row": 2574,
56
  "start": {
57
  "file": 2,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 134,
61
+ "row": 316
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 1010190
67
  },
68
  "ultra_qa": {
69
+ "page": 106,
70
+ "row": 2578,
71
  "start": {
72
  "file": 0,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
+ "row": 46661
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 436754
82
  },
83
  "ultra_style": {
84
+ "page": 162,
85
+ "row": 2876,
86
  "start": {
87
  "file": 0,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
  "row_group": 7,
91
+ "row": 68428
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 666428
97
  },
98
  "cortex": {
99
+ "page": 20940,
100
+ "row": 66,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 2680386
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 75000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "b90a19c4b2d7929e861313731bcb5beb014b3e36b39789a0beda7568ce6a846a", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "5026bd4bb1b3b18f5fc726c01908ba56660bc127713bb70b791880fb27548eeb", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "a365a3ff77a3205ac0d3c410749b4fb4670618f5a7de9659698fe71360225cc6", "coverage.json": "2dd27baaf4f3a0abac91b8368d3639578f98d0308ab7fb49afd5e4690f165c2f", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000075000"}
 
1
+ {"step": 80000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "016fae2e58ad750ac0b4abde700bbdbea9199273ac137f7e182d0fe7874da0dc", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "63c9c8c99731fc20714a54a44377517fe0f1f3043beea6e7f6b8a0c370badb1e", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "ea08386ca957ff57a36e47f5f3cdc24c23475683922434b0de7b9c8abf2744b1", "coverage.json": "42736094b47b18e0027534773acb49639d163b78174444895fdb7d300580166a", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000080000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b90a19c4b2d7929e861313731bcb5beb014b3e36b39789a0beda7568ce6a846a
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:016fae2e58ad750ac0b4abde700bbdbea9199273ac137f7e182d0fe7874da0dc
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": 74395.06001898335,
26
- "required_pages_per_second_upper_estimate": 0.00644837524988969
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": 9299.382502372919,
40
- "required_pages_per_second_upper_estimate": 0.003709662718355241
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": 18598.765004745837,
54
- "required_pages_per_second_upper_estimate": 0.004302496956763474
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": 27898.147507118756,
68
- "required_pages_per_second_upper_estimate": 0.0016988508129156528
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": 27898.147507118756,
82
- "required_pages_per_second_upper_estimate": 0.0021476766221130506
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": 21,
95
- "required_token_positions_per_second": 27898.147507118756,
96
- "required_pages_per_second_upper_estimate": 0.34640162294496635
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,412 +155,412 @@
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": 61,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
- "producer_batch_seconds": 0.13118641220319205,
162
- "producer_p95_seconds": 0.39197794900246663,
163
- "gpu_update_seconds": 0.9910335441786977,
164
- "producer_batches_per_second": 7.62274067264771,
165
- "gpu_batches_per_second": 1.009047580552617,
166
- "producer_headroom": 7.554391705169171,
167
  "sustainable": true,
168
- "buffer_seconds": 60.45304619490056,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
  "upload_pages_per_commit": 83,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 443816.2390829497,
174
- "upload_bytes_per_second": 5580829.247452082,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 95.2326333939206,
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": 1.1758935670004576,
184
- "bytes": 7725078,
185
- "pages": 38,
186
  "shards": 3
187
  },
188
  {
189
- "seconds": 1.0130331639993528,
190
- "bytes": 839053,
191
- "pages": 26,
192
- "shards": 1
193
- },
194
- {
195
- "seconds": 1.1233831270001247,
196
- "bytes": 6290212,
197
- "pages": 36,
198
  "shards": 2
199
  },
200
  {
201
- "seconds": 0.8791882089972205,
202
- "bytes": 6587099,
203
- "pages": 27,
204
  "shards": 3
205
  },
206
  {
207
- "seconds": 1.0528832150012022,
208
- "bytes": 11948549,
209
- "pages": 44,
210
- "shards": 3
211
  },
212
  {
213
- "seconds": 1.0517111139997724,
214
- "bytes": 6085549,
215
  "pages": 36,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 0.9597967700028676,
220
- "bytes": 2343290,
221
- "pages": 28,
222
  "shards": 2
223
  },
224
  {
225
- "seconds": 1.0771120079989487,
226
- "bytes": 10854426,
227
- "pages": 39,
228
  "shards": 3
229
  },
230
  {
231
- "seconds": 1.2815551900021092,
232
- "bytes": 7356201,
 
 
 
 
 
 
233
  "pages": 29,
234
  "shards": 3
235
  },
236
  {
237
- "seconds": 1.2021151590015506,
238
- "bytes": 960345,
239
- "pages": 28,
240
- "shards": 1
241
  },
242
  {
243
- "seconds": 3.8789173690020107,
244
- "bytes": 11941335,
245
- "pages": 36,
246
  "shards": 3
247
  },
248
  {
249
- "seconds": 1.0000528349992237,
250
- "bytes": 11258395,
251
- "pages": 42,
252
- "shards": 4
253
- },
254
- {
255
- "seconds": 0.9514743170002475,
256
- "bytes": 822769,
257
- "pages": 26,
258
- "shards": 1
259
  },
260
  {
261
- "seconds": 4.258936992002418,
262
- "bytes": 6104340,
263
- "pages": 36,
264
  "shards": 3
265
  },
266
  {
267
- "seconds": 1.0792534860011074,
268
- "bytes": 2442489,
269
- "pages": 31,
270
  "shards": 2
271
  },
272
  {
273
- "seconds": 1.0146635470009642,
274
- "bytes": 11041931,
275
  "pages": 34,
276
  "shards": 3
277
  },
278
  {
279
- "seconds": 1.0869366869992518,
280
- "bytes": 13432151,
281
- "pages": 40,
282
- "shards": 4
283
  },
284
  {
285
- "seconds": 0.7862899389983795,
286
- "bytes": 1239440,
287
- "pages": 35,
288
- "shards": 1
289
  },
290
  {
291
- "seconds": 0.8729745059972629,
292
- "bytes": 5859266,
293
- "pages": 29,
294
  "shards": 2
295
  },
296
  {
297
- "seconds": 1.0330235869987519,
298
- "bytes": 6506267,
299
- "pages": 40,
300
- "shards": 3
301
  },
302
  {
303
- "seconds": 1.471190406999085,
304
- "bytes": 6116938,
305
- "pages": 27,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 1.1064680679992307,
310
- "bytes": 7358899,
311
- "pages": 34,
312
- "shards": 3
313
  },
314
  {
315
- "seconds": 1.2355712200005655,
316
- "bytes": 6983067,
317
- "pages": 36,
318
- "shards": 2
319
  },
320
  {
321
- "seconds": 1.2635136329990928,
322
- "bytes": 9856957,
323
- "pages": 29,
324
  "shards": 3
325
  },
326
  {
327
- "seconds": 0.961533926998527,
328
- "bytes": 492258,
329
- "pages": 16,
330
- "shards": 1
331
  },
332
  {
333
- "seconds": 1.2022184809975442,
334
- "bytes": 7326926,
335
- "pages": 31,
336
  "shards": 3
337
  },
338
  {
339
- "seconds": 0.7280326340005558,
340
- "bytes": 1168675,
341
- "pages": 35,
342
  "shards": 1
343
  },
344
  {
345
- "seconds": 1.1539005899976473,
346
- "bytes": 5543356,
347
- "pages": 19,
348
- "shards": 2
349
- },
350
- {
351
- "seconds": 0.995166409000376,
352
- "bytes": 7263909,
353
  "pages": 39,
354
- "shards": 3
355
  },
356
  {
357
- "seconds": 1.1710582100022293,
358
- "bytes": 11639811,
359
- "pages": 30,
360
- "shards": 3
361
  },
362
  {
363
- "seconds": 1.5414868839980045,
364
- "bytes": 2598188,
365
  "pages": 37,
366
  "shards": 2
367
  },
368
  {
369
- "seconds": 1.4500177529989742,
370
- "bytes": 6186157,
371
- "pages": 35,
372
- "shards": 2
373
- },
374
- {
375
- "seconds": 1.1433413880004082,
376
- "bytes": 10345973,
377
- "pages": 35,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 0.963193997999042,
382
- "bytes": 2354048,
383
  "pages": 29,
 
 
 
 
 
 
 
 
 
 
 
 
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 0.9612779050003155,
388
- "bytes": 6227883,
389
  "pages": 36,
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 0.9593730830019922,
394
- "bytes": 6813129,
395
- "pages": 31,
396
- "shards": 2
397
  },
398
  {
399
- "seconds": 1.1213739670020004,
400
- "bytes": 11561927,
401
- "pages": 35,
402
- "shards": 4
403
  },
404
  {
405
- "seconds": 1.0164505040011136,
406
- "bytes": 6233089,
407
- "pages": 36,
408
  "shards": 2
409
  },
410
  {
411
- "seconds": 1.140030457998364,
412
- "bytes": 2525214,
413
- "pages": 34,
414
- "shards": 2
415
  },
416
  {
417
- "seconds": 0.8727539430001343,
418
- "bytes": 5779966,
419
- "pages": 32,
420
- "shards": 2
421
  },
422
  {
423
- "seconds": 1.2494100709991471,
424
- "bytes": 10781388,
425
- "pages": 37,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 0.9040151730005164,
430
- "bytes": 8396980,
431
  "pages": 35,
432
- "shards": 3
433
  },
434
  {
435
- "seconds": 1.1415953719988465,
436
- "bytes": 5808415,
437
- "pages": 28,
438
  "shards": 2
439
  },
440
  {
441
- "seconds": 1.0399577659991337,
442
- "bytes": 7078524,
443
- "pages": 40,
444
  "shards": 3
445
  },
446
  {
447
- "seconds": 0.9194039719986904,
448
- "bytes": 5850558,
449
- "pages": 28,
450
- "shards": 2
451
- },
452
- {
453
- "seconds": 0.8959461580016068,
454
- "bytes": 6085583,
455
- "pages": 36,
456
  "shards": 3
457
  },
458
  {
459
- "seconds": 0.9613766149996081,
460
- "bytes": 2539274,
461
- "pages": 33,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.3802653160018963,
466
- "bytes": 12066636,
467
- "pages": 34,
468
- "shards": 3
469
  },
470
  {
471
- "seconds": 0.828995023999596,
472
- "bytes": 4743589,
473
- "pages": 35,
474
- "shards": 2
475
  },
476
  {
477
- "seconds": 0.9685989429999609,
478
- "bytes": 7522789,
479
- "pages": 29,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 0.7894636139972135,
484
- "bytes": 5936931,
485
- "pages": 37,
486
- "shards": 2
487
  },
488
  {
489
- "seconds": 1.0263739810034167,
490
- "bytes": 6580240,
491
- "pages": 34,
492
  "shards": 3
493
  },
494
  {
495
- "seconds": 1.228623751005216,
496
- "bytes": 6179585,
497
  "pages": 37,
498
  "shards": 2
499
  },
500
  {
501
- "seconds": 1.3004778859976796,
502
- "bytes": 12330619,
503
- "pages": 35,
504
  "shards": 3
505
  },
506
  {
507
- "seconds": 1.576678247001837,
508
- "bytes": 2288077,
509
- "pages": 28,
510
  "shards": 2
511
  },
512
  {
513
- "seconds": 1.2293803360007587,
514
- "bytes": 10631332,
515
- "pages": 34,
 
 
 
 
 
 
516
  "shards": 3
517
  },
518
  {
519
- "seconds": 0.8141304599994328,
520
- "bytes": 2723440,
521
- "pages": 38,
522
- "shards": 2
523
  },
524
  {
525
- "seconds": 0.9625181489973329,
526
- "bytes": 5689945,
527
- "pages": 27,
528
  "shards": 2
529
  },
530
  {
531
- "seconds": 0.9489649809984257,
532
- "bytes": 5862208,
533
- "pages": 37,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 1.9020824750041356,
538
- "bytes": 12748087,
539
- "pages": 37,
540
- "shards": 4
541
  },
542
  {
543
- "seconds": 0.9002470639970852,
544
- "bytes": 6110509,
545
- "pages": 36,
546
  "shards": 2
547
  },
548
  {
549
- "seconds": 0.9653321419973508,
550
- "bytes": 880569,
551
- "pages": 26,
552
- "shards": 1
553
  },
554
  {
555
- "seconds": 1.074353076000989,
556
- "bytes": 7774309,
557
- "pages": 39,
558
  "shards": 3
559
  },
560
  {
561
- "seconds": 1.4574867379997158,
562
- "bytes": 9860401,
563
- "pages": 33,
 
 
 
 
 
 
564
  "shards": 3
565
  }
566
  ],
 
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": 75133.13345379168,
26
+ "required_pages_per_second_upper_estimate": 0.006512349584588858
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": 9391.64168172396,
40
+ "required_pages_per_second_upper_estimate": 0.003746466284396027
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": 18783.28336344792,
54
+ "required_pages_per_second_upper_estimate": 0.004345182031637001
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": 28174.925045171884,
68
+ "required_pages_per_second_upper_estimate": 0.0017157051128435733
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": 28174.925045171884,
82
+ "required_pages_per_second_upper_estimate": 0.0021689837231616427
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": 22,
95
+ "required_token_positions_per_second": 28174.925045171884,
96
+ "required_pages_per_second_upper_estimate": 0.34983827365275444
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": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
+ "producer_batch_seconds": 0.12603624823447035,
162
+ "producer_p95_seconds": 0.43727277699508704,
163
+ "gpu_update_seconds": 0.9812980852894166,
164
+ "producer_batches_per_second": 7.934225383634551,
165
+ "gpu_batches_per_second": 1.0190583422009507,
166
+ "producer_headroom": 7.785840177215272,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.84048128794383,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
  "upload_pages_per_commit": 83,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 287303.50039527926,
174
+ "upload_bytes_per_second": 6345416.31829655,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 61.64864310163291,
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": 0.9040151730005164,
184
+ "bytes": 8396980,
185
+ "pages": 35,
186
  "shards": 3
187
  },
188
  {
189
+ "seconds": 1.1415953719988465,
190
+ "bytes": 5808415,
191
+ "pages": 28,
 
 
 
 
 
 
192
  "shards": 2
193
  },
194
  {
195
+ "seconds": 1.0399577659991337,
196
+ "bytes": 7078524,
197
+ "pages": 40,
198
  "shards": 3
199
  },
200
  {
201
+ "seconds": 0.9194039719986904,
202
+ "bytes": 5850558,
203
+ "pages": 28,
204
+ "shards": 2
205
  },
206
  {
207
+ "seconds": 0.8959461580016068,
208
+ "bytes": 6085583,
209
  "pages": 36,
210
+ "shards": 3
211
  },
212
  {
213
+ "seconds": 0.9613766149996081,
214
+ "bytes": 2539274,
215
+ "pages": 33,
216
  "shards": 2
217
  },
218
  {
219
+ "seconds": 1.3802653160018963,
220
+ "bytes": 12066636,
221
+ "pages": 34,
222
  "shards": 3
223
  },
224
  {
225
+ "seconds": 0.828995023999596,
226
+ "bytes": 4743589,
227
+ "pages": 35,
228
+ "shards": 2
229
+ },
230
+ {
231
+ "seconds": 0.9685989429999609,
232
+ "bytes": 7522789,
233
  "pages": 29,
234
  "shards": 3
235
  },
236
  {
237
+ "seconds": 0.7894636139972135,
238
+ "bytes": 5936931,
239
+ "pages": 37,
240
+ "shards": 2
241
  },
242
  {
243
+ "seconds": 1.0263739810034167,
244
+ "bytes": 6580240,
245
+ "pages": 34,
246
  "shards": 3
247
  },
248
  {
249
+ "seconds": 1.228623751005216,
250
+ "bytes": 6179585,
251
+ "pages": 37,
252
+ "shards": 2
 
 
 
 
 
 
253
  },
254
  {
255
+ "seconds": 1.3004778859976796,
256
+ "bytes": 12330619,
257
+ "pages": 35,
258
  "shards": 3
259
  },
260
  {
261
+ "seconds": 1.576678247001837,
262
+ "bytes": 2288077,
263
+ "pages": 28,
264
  "shards": 2
265
  },
266
  {
267
+ "seconds": 1.2293803360007587,
268
+ "bytes": 10631332,
269
  "pages": 34,
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 0.8141304599994328,
274
+ "bytes": 2723440,
275
+ "pages": 38,
276
+ "shards": 2
277
  },
278
  {
279
+ "seconds": 0.9625181489973329,
280
+ "bytes": 5689945,
281
+ "pages": 27,
282
+ "shards": 2
283
  },
284
  {
285
+ "seconds": 0.9489649809984257,
286
+ "bytes": 5862208,
287
+ "pages": 37,
288
  "shards": 2
289
  },
290
  {
291
+ "seconds": 1.9020824750041356,
292
+ "bytes": 12748087,
293
+ "pages": 37,
294
+ "shards": 4
295
  },
296
  {
297
+ "seconds": 0.9002470639970852,
298
+ "bytes": 6110509,
299
+ "pages": 36,
300
  "shards": 2
301
  },
302
  {
303
+ "seconds": 0.9653321419973508,
304
+ "bytes": 880569,
305
+ "pages": 26,
306
+ "shards": 1
307
  },
308
  {
309
+ "seconds": 1.074353076000989,
310
+ "bytes": 7774309,
311
+ "pages": 39,
312
+ "shards": 3
313
  },
314
  {
315
+ "seconds": 1.4574867379997158,
316
+ "bytes": 9860401,
317
+ "pages": 33,
318
  "shards": 3
319
  },
320
  {
321
+ "seconds": 0.7756101689956267,
322
+ "bytes": 2491834,
323
+ "pages": 32,
324
+ "shards": 2
325
  },
326
  {
327
+ "seconds": 1.4068310860020574,
328
+ "bytes": 11759986,
329
+ "pages": 28,
330
  "shards": 3
331
  },
332
  {
333
+ "seconds": 0.806070794002153,
334
+ "bytes": 884841,
335
+ "pages": 27,
336
  "shards": 1
337
  },
338
  {
339
+ "seconds": 1.161329097005364,
340
+ "bytes": 12346007,
 
 
 
 
 
 
341
  "pages": 39,
342
+ "shards": 4
343
  },
344
  {
345
+ "seconds": 0.916572713998903,
346
+ "bytes": 942236,
347
+ "pages": 27,
348
+ "shards": 1
349
  },
350
  {
351
+ "seconds": 1.3328421769983834,
352
+ "bytes": 6157959,
353
  "pages": 37,
354
  "shards": 2
355
  },
356
  {
357
+ "seconds": 1.2523999580007512,
358
+ "bytes": 7610947,
359
+ "pages": 30,
 
 
 
 
 
 
360
  "shards": 3
361
  },
362
  {
363
+ "seconds": 0.7598877909986186,
364
+ "bytes": 961682,
365
  "pages": 29,
366
+ "shards": 1
367
+ },
368
+ {
369
+ "seconds": 1.3535268059931695,
370
+ "bytes": 16112402,
371
+ "pages": 42,
372
+ "shards": 4
373
+ },
374
+ {
375
+ "seconds": 0.9407444530006615,
376
+ "bytes": 2455446,
377
+ "pages": 32,
378
  "shards": 2
379
  },
380
  {
381
+ "seconds": 0.9131047679984476,
382
+ "bytes": 6268591,
383
  "pages": 36,
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 0.9507441360037774,
388
+ "bytes": 6652309,
389
+ "pages": 29,
390
+ "shards": 3
391
  },
392
  {
393
+ "seconds": 0.870253834000323,
394
+ "bytes": 6235033,
395
+ "pages": 38,
396
+ "shards": 3
397
  },
398
  {
399
+ "seconds": 0.9296957249971456,
400
+ "bytes": 6084778,
401
+ "pages": 29,
402
  "shards": 2
403
  },
404
  {
405
+ "seconds": 1.147810253998614,
406
+ "bytes": 8392070,
407
+ "pages": 39,
408
+ "shards": 3
409
  },
410
  {
411
+ "seconds": 1.153522004999104,
412
+ "bytes": 10226084,
413
+ "pages": 27,
414
+ "shards": 3
415
  },
416
  {
417
+ "seconds": 1.0265658620046452,
418
+ "bytes": 7737951,
419
+ "pages": 42,
420
  "shards": 3
421
  },
422
  {
423
+ "seconds": 0.7752080549980747,
424
+ "bytes": 1141313,
425
  "pages": 35,
426
+ "shards": 1
427
  },
428
  {
429
+ "seconds": 0.8455009660028736,
430
+ "bytes": 5959516,
431
+ "pages": 29,
432
  "shards": 2
433
  },
434
  {
435
+ "seconds": 1.0879218709960696,
436
+ "bytes": 6248481,
437
+ "pages": 29,
438
  "shards": 3
439
  },
440
  {
441
+ "seconds": 0.9940221099968767,
442
+ "bytes": 11924691,
443
+ "pages": 42,
 
 
 
 
 
 
444
  "shards": 3
445
  },
446
  {
447
+ "seconds": 0.9983105119972606,
448
+ "bytes": 6041328,
449
+ "pages": 29,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 0.9112700020050397,
454
+ "bytes": 2626384,
455
+ "pages": 37,
456
+ "shards": 2
457
  },
458
  {
459
+ "seconds": 1.077101428003516,
460
+ "bytes": 10954533,
461
+ "pages": 32,
462
+ "shards": 3
463
  },
464
  {
465
+ "seconds": 0.987721411001985,
466
+ "bytes": 7423511,
467
+ "pages": 39,
468
  "shards": 3
469
  },
470
  {
471
+ "seconds": 0.7559060139974463,
472
+ "bytes": 900379,
473
+ "pages": 27,
474
+ "shards": 1
475
  },
476
  {
477
+ "seconds": 1.208673165005166,
478
+ "bytes": 11672731,
479
+ "pages": 32,
480
  "shards": 3
481
  },
482
  {
483
+ "seconds": 0.8157727979996707,
484
+ "bytes": 2711312,
485
  "pages": 37,
486
  "shards": 2
487
  },
488
  {
489
+ "seconds": 1.0441342500052997,
490
+ "bytes": 9648084,
491
+ "pages": 31,
492
  "shards": 3
493
  },
494
  {
495
+ "seconds": 0.8669746300001862,
496
+ "bytes": 6281304,
497
+ "pages": 38,
498
  "shards": 2
499
  },
500
  {
501
+ "seconds": 1.0493857530018431,
502
+ "bytes": 2405359,
503
+ "pages": 29,
504
+ "shards": 2
505
+ },
506
+ {
507
+ "seconds": 1.4177541469980497,
508
+ "bytes": 10495082,
509
+ "pages": 41,
510
  "shards": 3
511
  },
512
  {
513
+ "seconds": 0.93050350699923,
514
+ "bytes": 8304633,
515
+ "pages": 32,
516
+ "shards": 3
517
  },
518
  {
519
+ "seconds": 1.291375706998224,
520
+ "bytes": 5813086,
521
+ "pages": 29,
522
  "shards": 2
523
  },
524
  {
525
+ "seconds": 1.007629627994902,
526
+ "bytes": 6351245,
527
+ "pages": 35,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.1004838699955144,
532
+ "bytes": 7253121,
533
+ "pages": 34,
534
+ "shards": 3
535
  },
536
  {
537
+ "seconds": 0.860785910001141,
538
+ "bytes": 6150922,
539
+ "pages": 38,
540
  "shards": 2
541
  },
542
  {
543
+ "seconds": 0.967503943000338,
544
+ "bytes": 2353719,
545
+ "pages": 29,
546
+ "shards": 2
547
  },
548
  {
549
+ "seconds": 1.1848933270011912,
550
+ "bytes": 11480176,
551
+ "pages": 30,
552
  "shards": 3
553
  },
554
  {
555
+ "seconds": 0.9306560840050224,
556
+ "bytes": 5914010,
557
+ "pages": 31,
558
+ "shards": 2
559
+ },
560
+ {
561
+ "seconds": 0.986767267997493,
562
+ "bytes": 6142895,
563
+ "pages": 38,
564
  "shards": 3
565
  }
566
  ],
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5026bd4bb1b3b18f5fc726c01908ba56660bc127713bb70b791880fb27548eeb
3
- size 721791615
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:63c9c8c99731fc20714a54a44377517fe0f1f3043beea6e7f6b8a0c370badb1e
3
+ size 721775039
training_config.json CHANGED
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 61,
50
- "prefetch_low_water": 30,
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": 62,
50
+ "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,