appvoid commited on
Commit
94cfd93
·
verified ·
1 Parent(s): b3e70f5

Qwen90M step 545000; 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 +247 -247
  6. trainer.pt +2 -2
  7. training_config.json +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 540000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "b38e10d38ed7ada06841ff8982da7d676daf791bf507a6cf07a68d71c4a4078d",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "043647c8c26e8fed077bb197406f27a2662b7f2ed8742ee4065fa206a57818d0",
12
- "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6",
13
  "performance.json": "3070ebbdf9b2b0fd6a0a237819043aa2c2a35c74a97edfa223e5c324471af4af",
14
- "shard_scheduler.json": "de12fe13ce1f77cd03863bd480ed4c92361aad4e582dff7ec3d7a242555cb0c8",
15
- "coverage.json": "c6a7d6cc92b6319e9b013aeaaaf6b0e9a72597bf84b77a74dadf487e898d80d9",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 545000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "a7beda1249d2d8fd1d0789e4fbf74cf1401e6a10abd39cdf9f1651fdfbec6e4b",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "abf7465cb001fce37e180b53d58dac3a443134a6c6fe31d634de0b34d4e174e7",
12
+ "training_config.json": "eb51b7316ffa2242848fb50069dd8c6e0ecf28534fa6202f12a9fcfdcf6b16eb",
13
  "performance.json": "3070ebbdf9b2b0fd6a0a237819043aa2c2a35c74a97edfa223e5c324471af4af",
14
+ "shard_scheduler.json": "9362a296b2832b4b5007ca3ec1a25ba07f94fdb1146f86932cdf2269d63a6f4c",
15
+ "coverage.json": "682a00d4779d7723fea633cd98c833ad848bbb3df21fa445fa4095cd4ab9203f",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 540000,
3
  "stats": {
4
- "climbmix/blocks": 19440000,
5
- "climbmix/targets": 32495587833,
6
- "climbmix/input_tokens": 32497646947,
7
- "ultra_qa/blocks": 7290000,
8
- "ultra_qa/targets": 11892698357,
9
- "ultra_qa/input_tokens": 11892721312,
10
- "cortex/blocks": 7290000,
11
- "cortex/targets": 277235924,
12
- "cortex/input_tokens": 11588137812,
13
- "ultra_style/blocks": 7290000,
14
- "ultra_style/targets": 11529724861,
15
- "ultra_style/input_tokens": 11530329511,
16
- "openmath/blocks": 4860000,
17
- "openmath/targets": 7509510059,
18
- "openmath/input_tokens": 7512011367,
19
- "rewrite6/blocks": 2430000,
20
- "rewrite6/targets": 3983637088,
21
- "rewrite6/input_tokens": 3988004561
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 2662,
26
- "row": 3055,
27
  "start": {
28
- "file": 128,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 68,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 0,
37
- "documents": 10906607
38
  },
39
  "rewrite6": {
40
- "page": 14,
41
- "row": 2627,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 57344
47
  }
48
  },
49
  "epoch": 30,
50
- "offset": 2048,
51
- "documents": 6222091
52
  },
53
  "openmath": {
54
- "page": 1663,
55
- "row": 2767,
56
  "start": {
57
  "file": 15,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 261,
61
- "row": 898
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 6814415
67
  },
68
  "ultra_qa": {
69
- "page": 719,
70
- "row": 1872,
71
  "start": {
72
  "file": 5,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 5,
76
- "row": 24371
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 2048,
81
- "documents": 2946896
82
  },
83
  "ultra_style": {
84
- "page": 1098,
85
- "row": 616,
86
  "start": {
87
  "file": 6,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 4,
91
- "row": 47086
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 4096,
96
- "documents": 4498024
97
  },
98
  "cortex": {
99
- "page": 141384,
100
- "row": 37,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 18097189
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 545000,
3
  "stats": {
4
+ "climbmix/blocks": 19620000,
5
+ "climbmix/targets": 32797131226,
6
+ "climbmix/input_tokens": 32799209443,
7
+ "ultra_qa/blocks": 7357500,
8
+ "ultra_qa/targets": 12002847915,
9
+ "ultra_qa/input_tokens": 12002871095,
10
+ "cortex/blocks": 7357500,
11
+ "cortex/targets": 279804176,
12
+ "cortex/input_tokens": 11695457470,
13
+ "ultra_style/blocks": 7357500,
14
+ "ultra_style/targets": 11636722703,
15
+ "ultra_style/input_tokens": 11637332832,
16
+ "openmath/blocks": 4905000,
17
+ "openmath/targets": 7578975444,
18
+ "openmath/input_tokens": 7581499711,
19
+ "rewrite6/blocks": 2452500,
20
+ "rewrite6/targets": 4020473427,
21
+ "rewrite6/input_tokens": 4024881624
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 2687,
26
+ "row": 1151,
27
  "start": {
28
+ "file": 130,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 3,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 11007103
38
  },
39
  "rewrite6": {
40
+ "page": 28,
41
+ "row": 3287,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 114688
47
  }
48
  },
49
  "epoch": 30,
50
+ "offset": 0,
51
+ "documents": 6280095
52
  },
53
  "openmath": {
54
+ "page": 1679,
55
+ "row": 155,
56
  "start": {
57
  "file": 15,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 327,
61
+ "row": 434
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 6877339
67
  },
68
  "ultra_qa": {
69
+ "page": 726,
70
+ "row": 497,
71
  "start": {
72
  "file": 5,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 5,
76
+ "row": 53043
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 2974193
82
  },
83
  "ultra_style": {
84
+ "page": 1108,
85
+ "row": 980,
86
  "start": {
87
  "file": 6,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 5,
91
+ "row": 2462
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 4539348
97
  },
98
  "cortex": {
99
+ "page": 142692,
100
+ "row": 98,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 18264674
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 540000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "b38e10d38ed7ada06841ff8982da7d676daf791bf507a6cf07a68d71c4a4078d", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "043647c8c26e8fed077bb197406f27a2662b7f2ed8742ee4065fa206a57818d0", "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6", "performance.json": "3070ebbdf9b2b0fd6a0a237819043aa2c2a35c74a97edfa223e5c324471af4af", "shard_scheduler.json": "de12fe13ce1f77cd03863bd480ed4c92361aad4e582dff7ec3d7a242555cb0c8", "coverage.json": "c6a7d6cc92b6319e9b013aeaaaf6b0e9a72597bf84b77a74dadf487e898d80d9", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000540000"}
 
1
+ {"step": 545000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "a7beda1249d2d8fd1d0789e4fbf74cf1401e6a10abd39cdf9f1651fdfbec6e4b", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "abf7465cb001fce37e180b53d58dac3a443134a6c6fe31d634de0b34d4e174e7", "training_config.json": "eb51b7316ffa2242848fb50069dd8c6e0ecf28534fa6202f12a9fcfdcf6b16eb", "performance.json": "3070ebbdf9b2b0fd6a0a237819043aa2c2a35c74a97edfa223e5c324471af4af", "shard_scheduler.json": "9362a296b2832b4b5007ca3ec1a25ba07f94fdb1146f86932cdf2269d63a6f4c", "coverage.json": "682a00d4779d7723fea633cd98c833ad848bbb3df21fa445fa4095cd4ab9203f", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000545000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b38e10d38ed7ada06841ff8982da7d676daf791bf507a6cf07a68d71c4a4078d
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a7beda1249d2d8fd1d0789e4fbf74cf1401e6a10abd39cdf9f1651fdfbec6e4b
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": 76448.66075247723,
26
- "required_pages_per_second_upper_estimate": 0.005288895736188249
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": 9556.082594059653,
40
- "required_pages_per_second_upper_estimate": 0.004000932228436205
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": 19112.165188119307,
54
- "required_pages_per_second_upper_estimate": 0.00398055998699525
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": 28668.24778217896,
68
- "required_pages_per_second_upper_estimate": 0.0017186176806618552
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": 28668.24778217896,
82
- "required_pages_per_second_upper_estimate": 0.0022288301667688993
83
  },
84
  {
85
  "source": "cortex",
@@ -92,8 +92,8 @@
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": 28668.24778217896,
96
- "required_pages_per_second_upper_estimate": 0.3677255009835554
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": 704566,
161
- "producer_batch_seconds": 0.3263542902969334,
162
- "producer_p95_seconds": 0.2078722389996983,
163
- "gpu_update_seconds": 0.9644119239539579,
164
- "producer_batches_per_second": 3.064154600480815,
165
- "gpu_batches_per_second": 1.0369013231401534,
166
- "producer_headroom": 2.9551072335420745,
167
  "sustainable": true,
168
- "buffer_seconds": 60.757951209099346,
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": 393197.65477415506,
174
- "upload_bytes_per_second": 5782840.946028825,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 84.37106354158868,
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": 1.4530193070004316,
184
- "bytes": 7690650,
185
- "pages": 38,
186
  "shards": 3
187
  },
188
  {
189
- "seconds": 1.0416680760008603,
190
- "bytes": 5628205,
191
- "pages": 24,
192
  "shards": 2
193
  },
194
  {
195
- "seconds": 1.0915457699993567,
196
- "bytes": 13214903,
197
- "pages": 39,
198
- "shards": 4
199
- },
200
- {
201
- "seconds": 1.4333472569996957,
202
- "bytes": 6210705,
203
- "pages": 36,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 1.232096203000765,
208
- "bytes": 1201273,
209
- "pages": 35,
210
- "shards": 1
211
- },
212
- {
213
- "seconds": 1.2199994860002334,
214
- "bytes": 7355179,
215
- "pages": 26,
216
  "shards": 3
217
  },
218
  {
219
- "seconds": 1.1187948159986263,
220
- "bytes": 10435269,
221
- "pages": 45,
222
  "shards": 3
223
  },
224
  {
225
- "seconds": 0.8605066399995849,
226
- "bytes": 757336,
227
- "pages": 24,
228
- "shards": 1
229
  },
230
  {
231
- "seconds": 1.4440002779992938,
232
- "bytes": 13477323,
233
- "pages": 45,
234
- "shards": 4
235
  },
236
  {
237
- "seconds": 1.0930595740010176,
238
- "bytes": 6005049,
239
- "pages": 30,
240
  "shards": 2
241
  },
242
  {
243
- "seconds": 1.1409170160004578,
244
- "bytes": 7751245,
245
- "pages": 38,
 
 
 
 
 
 
246
  "shards": 3
247
  },
248
  {
249
- "seconds": 1.385997141000189,
250
- "bytes": 5854774,
 
 
 
 
 
 
251
  "pages": 28,
252
  "shards": 2
253
  },
254
  {
255
- "seconds": 0.8962630369987892,
256
- "bytes": 1124581,
257
- "pages": 35,
258
- "shards": 1
259
  },
260
  {
261
- "seconds": 1.1323460120001982,
262
- "bytes": 7094094,
263
- "pages": 25,
264
- "shards": 3
265
  },
266
  {
267
- "seconds": 1.3368760009998368,
268
- "bytes": 10726853,
269
  "pages": 39,
270
  "shards": 3
271
  },
272
  {
273
- "seconds": 0.9546346269999049,
274
- "bytes": 5999756,
275
- "pages": 29,
276
  "shards": 2
277
  },
278
  {
279
- "seconds": 1.1835868019988993,
280
- "bytes": 7840439,
281
- "pages": 38,
282
  "shards": 3
283
  },
284
  {
285
- "seconds": 1.00542208799925,
286
- "bytes": 5360697,
287
- "pages": 33,
288
  "shards": 2
289
  },
290
  {
291
- "seconds": 1.2967056479992607,
292
- "bytes": 7612762,
293
- "pages": 38,
294
- "shards": 3
295
  },
296
  {
297
- "seconds": 0.9550784620005288,
298
- "bytes": 883731,
 
 
 
 
 
 
299
  "pages": 26,
300
  "shards": 1
301
  },
302
  {
303
- "seconds": 1.1562814480002999,
304
- "bytes": 11922123,
305
- "pages": 40,
306
  "shards": 3
307
  },
308
  {
309
- "seconds": 1.2267546230013977,
310
- "bytes": 11778233,
311
- "pages": 33,
312
- "shards": 4
313
  },
314
  {
315
- "seconds": 0.9305929850015673,
316
- "bytes": 1085853,
317
- "pages": 35,
318
  "shards": 2
319
  },
320
  {
321
- "seconds": 1.4772485220000817,
322
- "bytes": 7565247,
323
- "pages": 36,
324
- "shards": 3
325
  },
326
  {
327
- "seconds": 1.035932449000029,
328
- "bytes": 5952411,
329
- "pages": 30,
330
  "shards": 2
331
  },
332
  {
333
- "seconds": 0.7514245589991333,
334
- "bytes": 257606,
335
- "pages": 8,
336
- "shards": 1
337
  },
338
  {
339
- "seconds": 0.9524619530002383,
340
- "bytes": 4638620,
341
- "pages": 26,
342
- "shards": 2
343
  },
344
  {
345
- "seconds": 1.3739831450002384,
346
- "bytes": 13357834,
347
- "pages": 40,
348
- "shards": 4
349
  },
350
  {
351
- "seconds": 1.1055001610002364,
352
- "bytes": 6014129,
353
  "pages": 36,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 0.9146980179993989,
358
- "bytes": 2516554,
359
- "pages": 34,
360
  "shards": 2
361
  },
362
  {
363
- "seconds": 1.1637235170001077,
364
- "bytes": 10485207,
365
- "pages": 34,
366
- "shards": 3
367
- },
368
- {
369
- "seconds": 1.0128666689997772,
370
- "bytes": 6455974,
371
  "pages": 36,
372
  "shards": 2
373
  },
374
  {
375
- "seconds": 1.0892136950005806,
376
- "bytes": 8273419,
377
- "pages": 32,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 1.3893709189997026,
382
- "bytes": 6184789,
383
- "pages": 36,
384
  "shards": 2
385
  },
386
  {
387
- "seconds": 1.226723518999279,
388
- "bytes": 5211595,
389
- "pages": 30,
 
 
 
 
 
 
390
  "shards": 2
391
  },
392
  {
393
- "seconds": 1.4460434569991776,
394
- "bytes": 7613200,
395
- "pages": 39,
396
  "shards": 3
397
  },
398
  {
399
- "seconds": 1.303638638000848,
400
- "bytes": 6027470,
401
- "pages": 37,
402
  "shards": 2
403
  },
404
  {
405
- "seconds": 1.9445103569996718,
406
- "bytes": 2479005,
407
- "pages": 31,
408
  "shards": 2
409
  },
410
  {
411
- "seconds": 1.5074067269997613,
412
- "bytes": 15537281,
413
- "pages": 35,
414
- "shards": 4
415
- },
416
- {
417
- "seconds": 0.9092938880003203,
418
- "bytes": 1113077,
419
- "pages": 34,
420
- "shards": 1
421
  },
422
  {
423
- "seconds": 1.0412252780006384,
424
- "bytes": 7665353,
425
- "pages": 39,
426
  "shards": 3
427
  },
428
  {
429
- "seconds": 1.1747847249989718,
430
- "bytes": 10716254,
431
- "pages": 34,
432
  "shards": 3
433
  },
434
  {
435
- "seconds": 1.0711626960001013,
436
- "bytes": 2631436,
437
- "pages": 37,
438
- "shards": 2
439
- },
440
- {
441
- "seconds": 1.137612402000741,
442
- "bytes": 5722821,
443
- "pages": 28,
444
  "shards": 2
445
  },
446
  {
447
- "seconds": 1.4303736129986646,
448
- "bytes": 11919616,
449
- "pages": 42,
450
  "shards": 3
451
  },
452
  {
453
- "seconds": 1.0499967649993778,
454
- "bytes": 7310677,
455
- "pages": 35,
456
  "shards": 3
457
  },
458
  {
459
- "seconds": 1.0670534790006059,
460
- "bytes": 5802765,
461
- "pages": 28,
462
- "shards": 2
463
  },
464
  {
465
- "seconds": 0.9111343770000531,
466
- "bytes": 2649643,
467
- "pages": 36,
468
- "shards": 2
469
  },
470
  {
471
- "seconds": 1.2196053769985156,
472
- "bytes": 6057143,
473
- "pages": 37,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.2347321990000637,
478
- "bytes": 9643472,
479
- "pages": 35,
480
  "shards": 3
481
  },
482
  {
483
- "seconds": 1.1648727040010272,
484
- "bytes": 8469052,
485
  "pages": 35,
486
- "shards": 3
487
- },
488
- {
489
- "seconds": 1.1223673130007228,
490
- "bytes": 6068911,
491
- "pages": 36,
492
  "shards": 2
493
  },
494
  {
495
- "seconds": 1.1270621679996111,
496
- "bytes": 5842224,
497
  "pages": 28,
498
- "shards": 2
499
  },
500
  {
501
- "seconds": 1.3927226420000807,
502
- "bytes": 7163586,
503
- "pages": 36,
504
- "shards": 4
505
  },
506
  {
507
- "seconds": 1.230830958000297,
508
- "bytes": 6020552,
509
  "pages": 37,
510
- "shards": 2
511
- },
512
- {
513
- "seconds": 1.2729082169989852,
514
- "bytes": 7700586,
515
- "pages": 39,
516
- "shards": 3
517
  },
518
  {
519
- "seconds": 0.9891643420014589,
520
- "bytes": 6883017,
521
  "pages": 33,
522
- "shards": 2
523
  },
524
  {
525
- "seconds": 1.2779713299987634,
526
- "bytes": 9455856,
527
  "pages": 30,
528
  "shards": 3
529
  },
530
  {
531
- "seconds": 1.159703276000073,
532
- "bytes": 2570411,
 
 
 
 
 
 
 
 
 
 
 
 
533
  "pages": 35,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 1.2064628600001015,
538
- "bytes": 6189523,
539
- "pages": 36,
540
  "shards": 2
541
  },
542
  {
543
- "seconds": 1.4404870660000597,
544
- "bytes": 11907122,
545
- "pages": 40,
546
- "shards": 4
547
  },
548
  {
549
- "seconds": 0.842472636999446,
550
- "bytes": 857193,
551
- "pages": 26,
552
- "shards": 1
553
  },
554
  {
555
- "seconds": 1.0425618379995285,
556
- "bytes": 11991809,
557
- "pages": 35,
558
- "shards": 3
559
  },
560
  {
561
- "seconds": 1.543527111000003,
562
- "bytes": 7770797,
563
- "pages": 39,
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": 77724.66568028349,
26
+ "required_pages_per_second_upper_estimate": 0.00537717271783323
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": 9715.583210035436,
40
+ "required_pages_per_second_upper_estimate": 0.004067711805593652
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": 19431.166420070873,
54
+ "required_pages_per_second_upper_estimate": 0.00404699953098255
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": 29146.74963010631,
68
+ "required_pages_per_second_upper_estimate": 0.0017473031358150829
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": 29146.74963010631,
82
+ "required_pages_per_second_upper_estimate": 0.002266031580738054
83
  },
84
  {
85
  "source": "cortex",
 
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": 29146.74963010631,
96
+ "required_pages_per_second_upper_estimate": 0.37386320891351205
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": 64,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 704566,
161
+ "producer_batch_seconds": 0.10329342065628566,
162
+ "producer_p95_seconds": 0.2151632269997208,
163
+ "gpu_update_seconds": 0.9485791846731952,
164
+ "producer_batches_per_second": 9.681158718981271,
165
+ "gpu_batches_per_second": 1.0542082476166923,
166
+ "producer_headroom": 9.18334564434305,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.709067819084495,
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": 297407.2945755357,
174
+ "upload_bytes_per_second": 5461795.829792473,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 63.81668212842515,
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": 1.1747847249989718,
184
+ "bytes": 10716254,
185
+ "pages": 34,
186
  "shards": 3
187
  },
188
  {
189
+ "seconds": 1.0711626960001013,
190
+ "bytes": 2631436,
191
+ "pages": 37,
192
  "shards": 2
193
  },
194
  {
195
+ "seconds": 1.137612402000741,
196
+ "bytes": 5722821,
197
+ "pages": 28,
 
 
 
 
 
 
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.4303736129986646,
202
+ "bytes": 11919616,
203
+ "pages": 42,
 
 
 
 
 
 
204
  "shards": 3
205
  },
206
  {
207
+ "seconds": 1.0499967649993778,
208
+ "bytes": 7310677,
209
+ "pages": 35,
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.0670534790006059,
214
+ "bytes": 5802765,
215
+ "pages": 28,
216
+ "shards": 2
217
  },
218
  {
219
+ "seconds": 0.9111343770000531,
220
+ "bytes": 2649643,
221
+ "pages": 36,
222
+ "shards": 2
223
  },
224
  {
225
+ "seconds": 1.2196053769985156,
226
+ "bytes": 6057143,
227
+ "pages": 37,
228
  "shards": 2
229
  },
230
  {
231
+ "seconds": 1.2347321990000637,
232
+ "bytes": 9643472,
233
+ "pages": 35,
234
+ "shards": 3
235
+ },
236
+ {
237
+ "seconds": 1.1648727040010272,
238
+ "bytes": 8469052,
239
+ "pages": 35,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 1.1223673130007228,
244
+ "bytes": 6068911,
245
+ "pages": 36,
246
+ "shards": 2
247
+ },
248
+ {
249
+ "seconds": 1.1270621679996111,
250
+ "bytes": 5842224,
251
  "pages": 28,
252
  "shards": 2
253
  },
254
  {
255
+ "seconds": 1.3927226420000807,
256
+ "bytes": 7163586,
257
+ "pages": 36,
258
+ "shards": 4
259
  },
260
  {
261
+ "seconds": 1.230830958000297,
262
+ "bytes": 6020552,
263
+ "pages": 37,
264
+ "shards": 2
265
  },
266
  {
267
+ "seconds": 1.2729082169989852,
268
+ "bytes": 7700586,
269
  "pages": 39,
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 0.9891643420014589,
274
+ "bytes": 6883017,
275
+ "pages": 33,
276
  "shards": 2
277
  },
278
  {
279
+ "seconds": 1.2779713299987634,
280
+ "bytes": 9455856,
281
+ "pages": 30,
282
  "shards": 3
283
  },
284
  {
285
+ "seconds": 1.159703276000073,
286
+ "bytes": 2570411,
287
+ "pages": 35,
288
  "shards": 2
289
  },
290
  {
291
+ "seconds": 1.2064628600001015,
292
+ "bytes": 6189523,
293
+ "pages": 36,
294
+ "shards": 2
295
  },
296
  {
297
+ "seconds": 1.4404870660000597,
298
+ "bytes": 11907122,
299
+ "pages": 40,
300
+ "shards": 4
301
+ },
302
+ {
303
+ "seconds": 0.842472636999446,
304
+ "bytes": 857193,
305
  "pages": 26,
306
  "shards": 1
307
  },
308
  {
309
+ "seconds": 1.0425618379995285,
310
+ "bytes": 11991809,
311
+ "pages": 35,
312
  "shards": 3
313
  },
314
  {
315
+ "seconds": 1.543527111000003,
316
+ "bytes": 7770797,
317
+ "pages": 39,
318
+ "shards": 3
319
  },
320
  {
321
+ "seconds": 1.0592705330000172,
322
+ "bytes": 5143840,
323
+ "pages": 36,
324
  "shards": 2
325
  },
326
  {
327
+ "seconds": 0.9495372719993611,
328
+ "bytes": 5544847,
329
+ "pages": 18,
330
+ "shards": 2
331
  },
332
  {
333
+ "seconds": 1.068828384999506,
334
+ "bytes": 2455135,
335
+ "pages": 29,
336
  "shards": 2
337
  },
338
  {
339
+ "seconds": 1.2464688429990929,
340
+ "bytes": 6118587,
341
+ "pages": 36,
342
+ "shards": 2
343
  },
344
  {
345
+ "seconds": 0.8108600380001008,
346
+ "bytes": 790179,
347
+ "pages": 24,
348
+ "shards": 1
349
  },
350
  {
351
+ "seconds": 2.05223027700049,
352
+ "bytes": 18642145,
353
+ "pages": 45,
354
+ "shards": 5
355
  },
356
  {
357
+ "seconds": 1.3427877739995893,
358
+ "bytes": 6044460,
359
  "pages": 36,
360
  "shards": 2
361
  },
362
  {
363
+ "seconds": 1.2501696180006547,
364
+ "bytes": 2425262,
365
+ "pages": 29,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.180065475000447,
370
+ "bytes": 6094205,
 
 
 
 
 
 
371
  "pages": 36,
372
  "shards": 2
373
  },
374
  {
375
+ "seconds": 1.2719860980032536,
376
+ "bytes": 10158142,
377
+ "pages": 41,
378
  "shards": 3
379
  },
380
  {
381
+ "seconds": 0.8902074819998234,
382
+ "bytes": 2207572,
383
+ "pages": 24,
384
  "shards": 2
385
  },
386
  {
387
+ "seconds": 1.290630097999383,
388
+ "bytes": 12177821,
389
+ "pages": 42,
390
+ "shards": 3
391
+ },
392
+ {
393
+ "seconds": 3.9100393570006418,
394
+ "bytes": 2234803,
395
+ "pages": 29,
396
  "shards": 2
397
  },
398
  {
399
+ "seconds": 0.9967152519966476,
400
+ "bytes": 11168227,
401
+ "pages": 38,
402
  "shards": 3
403
  },
404
  {
405
+ "seconds": 1.0345216240020818,
406
+ "bytes": 6241760,
407
+ "pages": 36,
408
  "shards": 2
409
  },
410
  {
411
+ "seconds": 1.262870265996753,
412
+ "bytes": 2380360,
413
+ "pages": 28,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 0.9234551270019438,
418
+ "bytes": 5872826,
419
+ "pages": 30,
420
+ "shards": 2
 
 
 
 
 
 
421
  },
422
  {
423
+ "seconds": 1.268712075001531,
424
+ "bytes": 10759229,
425
+ "pages": 40,
426
  "shards": 3
427
  },
428
  {
429
+ "seconds": 1.2720383849991777,
430
+ "bytes": 7578314,
431
+ "pages": 39,
432
  "shards": 3
433
  },
434
  {
435
+ "seconds": 1.001895314999274,
436
+ "bytes": 6152237,
437
+ "pages": 36,
 
 
 
 
 
 
438
  "shards": 2
439
  },
440
  {
441
+ "seconds": 0.8757686259996262,
442
+ "bytes": 2112854,
443
+ "pages": 23,
444
  "shards": 3
445
  },
446
  {
447
+ "seconds": 1.275124282001343,
448
+ "bytes": 10806597,
449
+ "pages": 46,
450
  "shards": 3
451
  },
452
  {
453
+ "seconds": 0.989914507001231,
454
+ "bytes": 672610,
455
+ "pages": 18,
456
+ "shards": 1
457
  },
458
  {
459
+ "seconds": 1.4889224110011128,
460
+ "bytes": 13780761,
461
+ "pages": 49,
462
+ "shards": 4
463
  },
464
  {
465
+ "seconds": 0.993486828996538,
466
+ "bytes": 5717549,
467
+ "pages": 19,
468
  "shards": 2
469
  },
470
  {
471
+ "seconds": 1.089157542999601,
472
+ "bytes": 7603521,
473
+ "pages": 46,
474
  "shards": 3
475
  },
476
  {
477
+ "seconds": 1.0148978149991308,
478
+ "bytes": 6571585,
479
  "pages": 35,
 
 
 
 
 
 
480
  "shards": 2
481
  },
482
  {
483
+ "seconds": 1.1422953030014469,
484
+ "bytes": 877918,
485
  "pages": 28,
486
+ "shards": 1
487
  },
488
  {
489
+ "seconds": 1.1798037480002677,
490
+ "bytes": 7431767,
491
+ "pages": 29,
492
+ "shards": 3
493
  },
494
  {
495
+ "seconds": 1.270509606001724,
496
+ "bytes": 15692090,
497
  "pages": 37,
498
+ "shards": 4
 
 
 
 
 
 
499
  },
500
  {
501
+ "seconds": 1.250901432998944,
502
+ "bytes": 1087271,
503
  "pages": 33,
504
+ "shards": 1
505
  },
506
  {
507
+ "seconds": 1.5110034520002955,
508
+ "bytes": 7289248,
509
  "pages": 30,
510
  "shards": 3
511
  },
512
  {
513
+ "seconds": 1.0708622599995579,
514
+ "bytes": 1033241,
515
+ "pages": 32,
516
+ "shards": 1
517
+ },
518
+ {
519
+ "seconds": 1.2405112279993773,
520
+ "bytes": 12607883,
521
+ "pages": 39,
522
+ "shards": 4
523
+ },
524
+ {
525
+ "seconds": 1.1709360159984499,
526
+ "bytes": 6311701,
527
  "pages": 35,
528
  "shards": 2
529
  },
530
  {
531
+ "seconds": 1.1361314889982168,
532
+ "bytes": 6780795,
533
+ "pages": 32,
534
  "shards": 2
535
  },
536
  {
537
+ "seconds": 1.0505489870010933,
538
+ "bytes": 7688939,
539
+ "pages": 38,
540
+ "shards": 3
541
  },
542
  {
543
+ "seconds": 1.4415530749975005,
544
+ "bytes": 9558241,
545
+ "pages": 36,
546
+ "shards": 3
547
  },
548
  {
549
+ "seconds": 2.3017471879975346,
550
+ "bytes": 2579509,
551
+ "pages": 37,
552
+ "shards": 2
553
  },
554
  {
555
+ "seconds": 1.0516134489989781,
556
+ "bytes": 6165714,
557
+ "pages": 27,
558
+ "shards": 2
559
+ },
560
+ {
561
+ "seconds": 0.908222386999114,
562
+ "bytes": 5648395,
563
+ "pages": 35,
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 sha256:043647c8c26e8fed077bb197406f27a2662b7f2ed8742ee4065fa206a57818d0
3
- size 721757375
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:abf7465cb001fce37e180b53d58dac3a443134a6c6fe31d634de0b34d4e174e7
3
+ size 721774207
training_config.json CHANGED
@@ -46,8 +46,8 @@
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
- "prefetch_batches": 63,
50
- "prefetch_low_water": 31,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,
 
46
  "cpu_threads": 4,
47
  "page_rows": 4096,
48
  "cortex_page_rows": 128,
49
+ "prefetch_batches": 64,
50
+ "prefetch_low_water": 32,
51
  "memory_pages": 12,
52
  "cache_gb": 20,
53
  "force_rebenchmark": false,