appvoid commited on
Commit
7b338cf
·
verified ·
1 Parent(s): 483b5b5

Qwen90M step 745000; 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 +247 -247
  6. trainer.pt +2 -2
  7. training_config.json +1 -1
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 740000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "965f2a1b8062278f91a7a579c0f9910dd996acdaaa15c0589449d5be638afbb9",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "7a080253df8bf544a1d7795cc94dbd9bed18537ff0c92c322418f47ac23af393",
12
- "training_config.json": "a5b628e58d936dba526336e6ed1a3b16a703b7373430cc4aca66220c358ed037",
13
  "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379",
14
- "shard_scheduler.json": "7f8712a7f9538f373e624550e1f5a9ce3a8e415d6a140e719f8256eddac3ad48",
15
- "coverage.json": "7585df2fd9672883a1500e76331661f838d543263097041bee9efaf33517c187",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63"
 
1
  {
2
+ "step": 745000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "0d7ce863025d2619d914fb70f1a1e4a1068b7ce5609365ceb4f22dff6ccfaf85",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "d44898a301bc80cd4b55a3e048afe4856a384a2f161cf4889c3ae4e68e12da5a",
12
+ "training_config.json": "75248af0a321738b405c3d51d1d55c7040dbc0785759d95f8dbdf6cd071a6687",
13
  "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379",
14
+ "shard_scheduler.json": "b7c0bd5b9394f7c75d2d94d6f2fc1c2fae65583ba811d4b0353f995a152acb08",
15
+ "coverage.json": "798d5521406ab366e17be4a5abaa9337e274ccadae7107b1abfdcd9cbdebec16",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 740000,
3
  "stats": {
4
- "climbmix/blocks": 26640000,
5
- "climbmix/targets": 44534380574,
6
- "climbmix/input_tokens": 44537202395,
7
- "ultra_qa/blocks": 9990000,
8
- "ultra_qa/targets": 16297612094,
9
- "ultra_qa/input_tokens": 16297643521,
10
- "cortex/blocks": 9990000,
11
- "cortex/targets": 379836958,
12
- "cortex/input_tokens": 15879487851,
13
- "ultra_style/blocks": 9990000,
14
- "ultra_style/targets": 15801366047,
15
- "ultra_style/input_tokens": 15802194940,
16
- "openmath/blocks": 6660000,
17
- "openmath/targets": 10291092474,
18
- "openmath/input_tokens": 10294521411,
19
- "rewrite6/blocks": 3330000,
20
- "rewrite6/targets": 5459078580,
21
- "rewrite6/input_tokens": 5465063978
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3648,
26
- "row": 803,
27
  "start": {
28
- "file": 176,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 48,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 14943011
38
  },
39
  "rewrite6": {
40
- "page": 25,
41
- "row": 3061,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 102400
47
  }
48
  },
49
  "epoch": 41,
50
- "offset": 2048,
51
- "documents": 8527025
52
  },
53
  "openmath": {
54
- "page": 2280,
55
- "row": 890,
56
  "start": {
57
  "file": 21,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 169,
61
- "row": 230
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 9339770
67
  },
68
  "ultra_qa": {
69
- "page": 985,
70
- "row": 3725,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
- "row": 9755
77
  }
78
  },
79
  "epoch": 0,
80
- "offset": 0,
81
- "documents": 4038285
82
  },
83
  "ultra_style": {
84
- "page": 1504,
85
- "row": 3284,
86
  "start": {
87
- "file": 8,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 7,
91
- "row": 84737
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
- "documents": 6163668
97
  },
98
  "cortex": {
99
- "page": 193744,
100
- "row": 26,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 24799258
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 745000,
3
  "stats": {
4
+ "climbmix/blocks": 26820000,
5
+ "climbmix/targets": 44835016967,
6
+ "climbmix/input_tokens": 44837857904,
7
+ "ultra_qa/blocks": 10057500,
8
+ "ultra_qa/targets": 16407642063,
9
+ "ultra_qa/input_tokens": 16407673700,
10
+ "cortex/blocks": 10057500,
11
+ "cortex/targets": 382400656,
12
+ "cortex/input_tokens": 15986736473,
13
+ "ultra_style/blocks": 10057500,
14
+ "ultra_style/targets": 15908518038,
15
+ "ultra_style/input_tokens": 15909352153,
16
+ "openmath/blocks": 6705000,
17
+ "openmath/targets": 10360628323,
18
+ "openmath/input_tokens": 10364080425,
19
+ "rewrite6/blocks": 3352500,
20
+ "rewrite6/targets": 5495963673,
21
+ "rewrite6/input_tokens": 5501989340
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3672,
26
+ "row": 3991,
27
  "start": {
28
+ "file": 177,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 62,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 15044503
38
  },
39
  "rewrite6": {
40
+ "page": 39,
41
+ "row": 3164,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 159744
47
  }
48
  },
49
  "epoch": 41,
50
+ "offset": 0,
51
+ "documents": 8584472
52
  },
53
  "openmath": {
54
+ "page": 2295,
55
+ "row": 2630,
56
  "start": {
57
  "file": 21,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 230,
61
+ "row": 670
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 9402950
67
  },
68
  "ultra_qa": {
69
+ "page": 992,
70
+ "row": 2453,
71
  "start": {
72
  "file": 7,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 6,
76
+ "row": 38427
77
  }
78
  },
79
  "epoch": 0,
80
+ "offset": 4096,
81
+ "documents": 4065685
82
  },
83
  "ultra_style": {
84
+ "page": 1514,
85
+ "row": 3015,
86
  "start": {
87
+ "file": 9,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 0,
91
+ "row": 36870
92
  }
93
  },
94
  "epoch": 0,
95
  "offset": 2048,
96
+ "documents": 6204359
97
  },
98
  "cortex": {
99
+ "page": 195052,
100
+ "row": 49,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 24966705
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 740000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "965f2a1b8062278f91a7a579c0f9910dd996acdaaa15c0589449d5be638afbb9", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "7a080253df8bf544a1d7795cc94dbd9bed18537ff0c92c322418f47ac23af393", "training_config.json": "a5b628e58d936dba526336e6ed1a3b16a703b7373430cc4aca66220c358ed037", "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379", "shard_scheduler.json": "7f8712a7f9538f373e624550e1f5a9ce3a8e415d6a140e719f8256eddac3ad48", "coverage.json": "7585df2fd9672883a1500e76331661f838d543263097041bee9efaf33517c187", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63", "folder": "step-000000740000"}
 
1
+ {"step": 745000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "0d7ce863025d2619d914fb70f1a1e4a1068b7ce5609365ceb4f22dff6ccfaf85", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "d44898a301bc80cd4b55a3e048afe4856a384a2f161cf4889c3ae4e68e12da5a", "training_config.json": "75248af0a321738b405c3d51d1d55c7040dbc0785759d95f8dbdf6cd071a6687", "performance.json": "48e3aa7938b07d5e08810d60bc208cc2cb9133c4c39a24151758b79306675379", "shard_scheduler.json": "b7c0bd5b9394f7c75d2d94d6f2fc1c2fae65583ba811d4b0353f995a152acb08", "coverage.json": "798d5521406ab366e17be4a5abaa9337e274ccadae7107b1abfdcd9cbdebec16", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "13510bfc04d8fed598fd0d09c1f778fbaebf19cefdd81f23f5ef3b909ef94a63", "folder": "step-000000745000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:965f2a1b8062278f91a7a579c0f9910dd996acdaaa15c0589449d5be638afbb9
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d7ce863025d2619d914fb70f1a1e4a1068b7ce5609365ceb4f22dff6ccfaf85
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": 75740.10139994962,
26
- "required_pages_per_second_upper_estimate": 0.005239875956096181
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": 9467.512674993703,
40
- "required_pages_per_second_upper_estimate": 0.003963849852873521
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": 18935.025349987405,
54
- "required_pages_per_second_upper_estimate": 0.0039436664302040505
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": 28402.538024981106,
68
- "required_pages_per_second_upper_estimate": 0.0017026887864331514
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": 28402.538024981106,
82
- "required_pages_per_second_upper_estimate": 0.002208172401880472
83
  },
84
  {
85
  "source": "cortex",
@@ -91,9 +91,9 @@
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
- "estimated_pages_for_buffer": 22,
95
- "required_token_positions_per_second": 28402.538024981106,
96
- "required_pages_per_second_upper_estimate": 0.3643172615151307
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": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.11142892917197855,
162
- "producer_p95_seconds": 0.2649414719999186,
163
- "gpu_update_seconds": 0.973434133797569,
164
- "producer_batches_per_second": 8.974330162112638,
165
- "gpu_batches_per_second": 1.0272908718526153,
166
- "producer_headroom": 8.735919307769514,
167
  "sustainable": true,
168
- "buffer_seconds": 60.35291629544928,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
  "upload_pages_per_commit": 82,
172
  "max_pending_pages": 4096,
173
- "generation_bytes_per_second": 287306.2932266875,
174
- "upload_bytes_per_second": 5098154.501941984,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 61.64924237823934,
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": 1.2874604959997669,
184
- "bytes": 2530940,
185
- "pages": 33,
186
- "shards": 2
187
  },
188
  {
189
- "seconds": 1.3693337599997903,
190
- "bytes": 10798536,
191
- "pages": 37,
192
  "shards": 3
193
  },
194
  {
195
- "seconds": 1.092897812000956,
196
- "bytes": 6537422,
197
- "pages": 28,
198
- "shards": 2
 
 
 
 
 
 
199
  },
200
  {
201
- "seconds": 0.9582660920004855,
202
- "bytes": 2603502,
203
  "pages": 36,
204
  "shards": 2
205
  },
206
  {
207
- "seconds": 1.5727497110001423,
208
- "bytes": 11757823,
209
- "pages": 31,
210
- "shards": 3
211
  },
212
  {
213
- "seconds": 1.1941701559990179,
214
- "bytes": 5221736,
215
- "pages": 34,
216
  "shards": 2
217
  },
218
  {
219
- "seconds": 1.6056268569991516,
220
- "bytes": 7578967,
221
- "pages": 37,
222
- "shards": 3
223
- },
224
- {
225
- "seconds": 1.20693444599965,
226
- "bytes": 6453330,
227
- "pages": 35,
228
  "shards": 2
229
  },
230
  {
231
- "seconds": 0.9651811909989192,
232
- "bytes": 2682508,
233
- "pages": 37,
234
  "shards": 2
235
  },
236
  {
237
- "seconds": 1.4213833419998991,
238
- "bytes": 10754179,
239
- "pages": 34,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 1.5129684479998105,
244
- "bytes": 6722253,
245
  "pages": 30,
246
  "shards": 3
247
  },
248
  {
249
- "seconds": 1.165229547999843,
250
- "bytes": 7984438,
251
- "pages": 38,
252
- "shards": 3
253
  },
254
  {
255
- "seconds": 1.001051560999258,
256
- "bytes": 939443,
257
- "pages": 30,
258
- "shards": 1
259
  },
260
  {
261
- "seconds": 1.4753997340012575,
262
- "bytes": 11016252,
263
- "pages": 34,
264
- "shards": 4
265
  },
266
  {
267
- "seconds": 1.6974678700007644,
268
- "bytes": 6266676,
269
- "pages": 36,
270
  "shards": 2
271
  },
272
  {
273
- "seconds": 1.6492898739998054,
274
- "bytes": 1094803,
275
- "pages": 34,
276
  "shards": 1
277
  },
278
  {
279
- "seconds": 1.8326161909990333,
280
- "bytes": 13393499,
281
- "pages": 34,
282
- "shards": 4
283
  },
284
  {
285
- "seconds": 1.4041516550005326,
286
- "bytes": 11034703,
287
- "pages": 37,
288
  "shards": 3
289
  },
290
  {
291
- "seconds": 1.1081106500005262,
292
- "bytes": 1056276,
293
- "pages": 28,
294
- "shards": 1
295
  },
296
  {
297
- "seconds": 1.211192590999417,
298
- "bytes": 7370118,
299
- "pages": 35,
300
  "shards": 3
301
  },
302
  {
303
- "seconds": 1.246054737999657,
304
- "bytes": 1155172,
305
- "pages": 34,
306
- "shards": 1
307
  },
308
  {
309
- "seconds": 1.6413428979994933,
310
- "bytes": 11205821,
311
- "pages": 33,
312
- "shards": 4
313
  },
314
  {
315
- "seconds": 1.3393069079993438,
316
- "bytes": 11931034,
317
- "pages": 38,
318
- "shards": 3
319
  },
320
  {
321
- "seconds": 0.9887657250001212,
322
- "bytes": 1035853,
323
- "pages": 34,
324
  "shards": 1
325
  },
326
  {
327
- "seconds": 1.3170902919991931,
328
- "bytes": 7309091,
329
- "pages": 30,
330
  "shards": 3
331
  },
332
  {
333
- "seconds": 1.3663103399994725,
334
- "bytes": 5168575,
335
- "pages": 32,
336
- "shards": 2
337
- },
338
- {
339
- "seconds": 1.1543622389999655,
340
- "bytes": 6027007,
341
- "pages": 31,
342
- "shards": 2
343
  },
344
  {
345
- "seconds": 1.39643492399955,
346
- "bytes": 7665740,
347
  "pages": 38,
348
  "shards": 3
349
  },
350
  {
351
- "seconds": 1.1344610849992023,
352
- "bytes": 6680190,
353
- "pages": 29,
354
  "shards": 2
355
  },
356
  {
357
- "seconds": 1.5331677430003765,
358
- "bytes": 11639669,
359
- "pages": 33,
360
- "shards": 4
361
  },
362
  {
363
- "seconds": 0.9482507379998424,
364
- "bytes": 1158257,
365
- "pages": 35,
366
- "shards": 1
367
  },
368
  {
369
- "seconds": 1.2962373409991415,
370
- "bytes": 6282079,
371
  "pages": 36,
372
  "shards": 2
373
  },
374
  {
375
- "seconds": 1.4015076579999004,
376
- "bytes": 7431338,
377
- "pages": 31,
378
  "shards": 3
379
  },
380
  {
381
- "seconds": 0.9902176969990251,
382
- "bytes": 5564439,
383
- "pages": 37,
384
- "shards": 2
385
  },
386
  {
387
- "seconds": 1.667742324003484,
388
- "bytes": 13340513,
389
- "pages": 34,
390
- "shards": 4
391
  },
392
  {
393
- "seconds": 1.2163241669986746,
394
- "bytes": 1010699,
395
- "pages": 28,
396
- "shards": 1
397
  },
398
  {
399
- "seconds": 1.3674872539995704,
400
- "bytes": 6041473,
401
  "pages": 37,
402
  "shards": 2
403
  },
404
  {
405
- "seconds": 1.3613106869997864,
406
- "bytes": 11006649,
407
- "pages": 33,
408
- "shards": 4
409
  },
410
  {
411
- "seconds": 0.8713928039978782,
412
- "bytes": 1169048,
413
- "pages": 34,
414
- "shards": 1
415
  },
416
  {
417
- "seconds": 1.0586420059989905,
418
- "bytes": 6189631,
419
- "pages": 35,
420
  "shards": 2
421
  },
422
  {
423
- "seconds": 1.3973069029998442,
424
- "bytes": 13145960,
425
- "pages": 34,
426
- "shards": 4
427
  },
428
  {
429
- "seconds": 1.3652075969985162,
430
- "bytes": 5700613,
431
  "pages": 36,
432
- "shards": 3
433
  },
434
  {
435
- "seconds": 1.7661054199998034,
436
- "bytes": 7454533,
437
- "pages": 30,
 
 
 
 
 
 
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 0.9842067560020951,
442
- "bytes": 1088907,
443
- "pages": 33,
444
- "shards": 1
445
  },
446
  {
447
- "seconds": 1.1375966550003795,
448
- "bytes": 6241738,
449
- "pages": 36,
450
  "shards": 2
451
  },
452
  {
453
- "seconds": 1.329601728997659,
454
- "bytes": 12006565,
455
- "pages": 36,
456
- "shards": 4
457
  },
458
  {
459
- "seconds": 1.637713142998109,
460
- "bytes": 6812710,
461
- "pages": 30,
462
  "shards": 2
463
  },
464
  {
465
- "seconds": 1.216705439001089,
466
- "bytes": 6221386,
467
- "pages": 36,
468
  "shards": 2
469
  },
470
  {
471
- "seconds": 1.2352482690002944,
472
- "bytes": 2570616,
473
- "pages": 35,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 1.360712293000688,
478
- "bytes": 10168781,
479
- "pages": 33,
480
- "shards": 3
481
  },
482
  {
483
- "seconds": 1.3043567909990088,
484
- "bytes": 7359704,
485
- "pages": 30,
486
- "shards": 3
487
  },
488
  {
489
- "seconds": 1.5744073349997052,
490
- "bytes": 1306429,
491
- "pages": 35,
492
- "shards": 1
493
  },
494
  {
495
- "seconds": 1.5331175120008993,
496
- "bytes": 11738494,
497
- "pages": 32,
498
- "shards": 3
499
  },
500
  {
501
- "seconds": 1.0237298360007117,
502
- "bytes": 7075351,
503
- "pages": 32,
504
  "shards": 3
505
  },
506
  {
507
- "seconds": 1.4313278709996666,
508
- "bytes": 6586683,
509
- "pages": 37,
510
- "shards": 2
511
  },
512
  {
513
- "seconds": 1.1789775520010153,
514
- "bytes": 937197,
515
- "pages": 30,
516
- "shards": 1
517
  },
518
  {
519
- "seconds": 1.0937288770001032,
520
- "bytes": 7693834,
521
- "pages": 37,
522
- "shards": 3
523
  },
524
  {
525
- "seconds": 1.3860260590008693,
526
- "bytes": 10598757,
527
- "pages": 35,
528
  "shards": 3
529
  },
530
  {
531
- "seconds": 1.0763569690025179,
532
- "bytes": 2341572,
533
- "pages": 29,
534
  "shards": 2
535
  },
536
  {
537
- "seconds": 1.3760699319973355,
538
- "bytes": 11906705,
539
- "pages": 41,
540
- "shards": 3
541
  },
542
  {
543
- "seconds": 1.2023094430005585,
544
- "bytes": 5878390,
545
  "pages": 28,
546
  "shards": 2
547
  },
548
  {
549
- "seconds": 0.9879657829987991,
550
- "bytes": 6822345,
551
- "pages": 34,
552
- "shards": 3
553
  },
554
  {
555
- "seconds": 1.0887722430015856,
556
- "bytes": 6046508,
557
  "pages": 36,
558
  "shards": 2
559
  },
560
  {
561
- "seconds": 0.9664684920026048,
562
- "bytes": 1011694,
563
- "pages": 29,
564
- "shards": 1
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": 76481.61050310897,
26
+ "required_pages_per_second_upper_estimate": 0.005291175276390908
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": 9560.201312888621,
40
+ "required_pages_per_second_upper_estimate": 0.004002656650001265
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": 19120.402625777242,
54
+ "required_pages_per_second_upper_estimate": 0.003982275628023559
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": 28680.603938665863,
68
+ "required_pages_per_second_upper_estimate": 0.0017193584133764865
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": 28680.603938665863,
82
+ "required_pages_per_second_upper_estimate": 0.002229790803586767
83
  },
84
  {
85
  "source": "cortex",
 
91
  "read_seconds": 0.04061180899998362,
92
  "tokens_per_second": 910315.184299418,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
+ "estimated_pages_for_buffer": 23,
95
+ "required_token_positions_per_second": 28680.603938665863,
96
+ "required_pages_per_second_upper_estimate": 0.3678839924919622
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": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.07352219521106917,
162
+ "producer_p95_seconds": 0.255219880000368,
163
+ "gpu_update_seconds": 0.9639964367251781,
164
+ "producer_batches_per_second": 13.601334904775047,
165
+ "gpu_batches_per_second": 1.0373482327353105,
166
+ "producer_headroom": 13.111638382908936,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.73177551368622,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
  "upload_pages_per_commit": 82,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 392318.91666500096,
174
+ "upload_bytes_per_second": 5364724.667990984,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 84.18250679934046,
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": 1.3973069029998442,
184
+ "bytes": 13145960,
185
+ "pages": 34,
186
+ "shards": 4
187
  },
188
  {
189
+ "seconds": 1.3652075969985162,
190
+ "bytes": 5700613,
191
+ "pages": 36,
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 1.7661054199998034,
196
+ "bytes": 7454533,
197
+ "pages": 30,
198
+ "shards": 3
199
+ },
200
+ {
201
+ "seconds": 0.9842067560020951,
202
+ "bytes": 1088907,
203
+ "pages": 33,
204
+ "shards": 1
205
  },
206
  {
207
+ "seconds": 1.1375966550003795,
208
+ "bytes": 6241738,
209
  "pages": 36,
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.329601728997659,
214
+ "bytes": 12006565,
215
+ "pages": 36,
216
+ "shards": 4
217
  },
218
  {
219
+ "seconds": 1.637713142998109,
220
+ "bytes": 6812710,
221
+ "pages": 30,
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 1.216705439001089,
226
+ "bytes": 6221386,
227
+ "pages": 36,
 
 
 
 
 
 
228
  "shards": 2
229
  },
230
  {
231
+ "seconds": 1.2352482690002944,
232
+ "bytes": 2570616,
233
+ "pages": 35,
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.360712293000688,
238
+ "bytes": 10168781,
239
+ "pages": 33,
240
  "shards": 3
241
  },
242
  {
243
+ "seconds": 1.3043567909990088,
244
+ "bytes": 7359704,
245
  "pages": 30,
246
  "shards": 3
247
  },
248
  {
249
+ "seconds": 1.5744073349997052,
250
+ "bytes": 1306429,
251
+ "pages": 35,
252
+ "shards": 1
253
  },
254
  {
255
+ "seconds": 1.5331175120008993,
256
+ "bytes": 11738494,
257
+ "pages": 32,
258
+ "shards": 3
259
  },
260
  {
261
+ "seconds": 1.0237298360007117,
262
+ "bytes": 7075351,
263
+ "pages": 32,
264
+ "shards": 3
265
  },
266
  {
267
+ "seconds": 1.4313278709996666,
268
+ "bytes": 6586683,
269
+ "pages": 37,
270
  "shards": 2
271
  },
272
  {
273
+ "seconds": 1.1789775520010153,
274
+ "bytes": 937197,
275
+ "pages": 30,
276
  "shards": 1
277
  },
278
  {
279
+ "seconds": 1.0937288770001032,
280
+ "bytes": 7693834,
281
+ "pages": 37,
282
+ "shards": 3
283
  },
284
  {
285
+ "seconds": 1.3860260590008693,
286
+ "bytes": 10598757,
287
+ "pages": 35,
288
  "shards": 3
289
  },
290
  {
291
+ "seconds": 1.0763569690025179,
292
+ "bytes": 2341572,
293
+ "pages": 29,
294
+ "shards": 2
295
  },
296
  {
297
+ "seconds": 1.3760699319973355,
298
+ "bytes": 11906705,
299
+ "pages": 41,
300
  "shards": 3
301
  },
302
  {
303
+ "seconds": 1.2023094430005585,
304
+ "bytes": 5878390,
305
+ "pages": 28,
306
+ "shards": 2
307
  },
308
  {
309
+ "seconds": 0.9879657829987991,
310
+ "bytes": 6822345,
311
+ "pages": 34,
312
+ "shards": 3
313
  },
314
  {
315
+ "seconds": 1.0887722430015856,
316
+ "bytes": 6046508,
317
+ "pages": 36,
318
+ "shards": 2
319
  },
320
  {
321
+ "seconds": 0.9664684920026048,
322
+ "bytes": 1011694,
323
+ "pages": 29,
324
  "shards": 1
325
  },
326
  {
327
+ "seconds": 1.350358371000766,
328
+ "bytes": 7369637,
329
+ "pages": 35,
330
  "shards": 3
331
  },
332
  {
333
+ "seconds": 1.1391658569991705,
334
+ "bytes": 11247930,
335
+ "pages": 27,
336
+ "shards": 3
 
 
 
 
 
 
337
  },
338
  {
339
+ "seconds": 1.2541976089996751,
340
+ "bytes": 7910419,
341
  "pages": 38,
342
  "shards": 3
343
  },
344
  {
345
+ "seconds": 1.501875165002275,
346
+ "bytes": 6144552,
347
+ "pages": 35,
348
  "shards": 2
349
  },
350
  {
351
+ "seconds": 1.1508353210010682,
352
+ "bytes": 1193747,
353
+ "pages": 34,
354
+ "shards": 1
355
  },
356
  {
357
+ "seconds": 1.6794259740017878,
358
+ "bytes": 11268683,
359
+ "pages": 32,
360
+ "shards": 4
361
  },
362
  {
363
+ "seconds": 1.0690768630011007,
364
+ "bytes": 5982757,
365
  "pages": 36,
366
  "shards": 2
367
  },
368
  {
369
+ "seconds": 1.3195378630007326,
370
+ "bytes": 8379086,
371
+ "pages": 35,
372
  "shards": 3
373
  },
374
  {
375
+ "seconds": 1.5084312399994815,
376
+ "bytes": 5915466,
377
+ "pages": 28,
378
+ "shards": 3
379
  },
380
  {
381
+ "seconds": 1.1589784089992463,
382
+ "bytes": 5695458,
383
+ "pages": 37,
384
+ "shards": 2
385
  },
386
  {
387
+ "seconds": 1.9208255039993674,
388
+ "bytes": 7437563,
389
+ "pages": 30,
390
+ "shards": 3
391
  },
392
  {
393
+ "seconds": 1.6179820170000312,
394
+ "bytes": 6082981,
395
  "pages": 37,
396
  "shards": 2
397
  },
398
  {
399
+ "seconds": 0.991337299998122,
400
+ "bytes": 1324376,
401
+ "pages": 36,
402
+ "shards": 1
403
  },
404
  {
405
+ "seconds": 1.2612079769969569,
406
+ "bytes": 16881065,
407
+ "pages": 36,
408
+ "shards": 5
409
  },
410
  {
411
+ "seconds": 1.1940166240019607,
412
+ "bytes": 6068573,
413
+ "pages": 30,
414
  "shards": 2
415
  },
416
  {
417
+ "seconds": 1.0544981790008023,
418
+ "bytes": 2562621,
419
+ "pages": 36,
420
+ "shards": 2
421
  },
422
  {
423
+ "seconds": 1.3444492599992373,
424
+ "bytes": 6214787,
425
  "pages": 36,
426
+ "shards": 2
427
  },
428
  {
429
+ "seconds": 1.2648084209977242,
430
+ "bytes": 5986982,
431
+ "pages": 31,
432
+ "shards": 2
433
+ },
434
+ {
435
+ "seconds": 1.4115151900005003,
436
+ "bytes": 7731251,
437
+ "pages": 39,
438
  "shards": 3
439
  },
440
  {
441
+ "seconds": 1.1964002810018428,
442
+ "bytes": 11692696,
443
+ "pages": 29,
444
+ "shards": 3
445
  },
446
  {
447
+ "seconds": 1.0258839710004395,
448
+ "bytes": 2700953,
449
+ "pages": 38,
450
  "shards": 2
451
  },
452
  {
453
+ "seconds": 1.2136837799989735,
454
+ "bytes": 9835275,
455
+ "pages": 34,
456
+ "shards": 3
457
  },
458
  {
459
+ "seconds": 1.2959440570011793,
460
+ "bytes": 5942943,
461
+ "pages": 35,
462
  "shards": 2
463
  },
464
  {
465
+ "seconds": 0.9916904240017175,
466
+ "bytes": 2484673,
467
+ "pages": 30,
468
  "shards": 2
469
  },
470
  {
471
+ "seconds": 1.0907142409996595,
472
+ "bytes": 6115368,
473
+ "pages": 37,
474
  "shards": 2
475
  },
476
  {
477
+ "seconds": 1.3705160650024482,
478
+ "bytes": 13033658,
479
+ "pages": 37,
480
+ "shards": 4
481
  },
482
  {
483
+ "seconds": 1.2911444290002692,
484
+ "bytes": 5949225,
485
+ "pages": 28,
486
+ "shards": 2
487
  },
488
  {
489
+ "seconds": 1.0684557280001172,
490
+ "bytes": 6155327,
491
+ "pages": 37,
492
+ "shards": 2
493
  },
494
  {
495
+ "seconds": 1.2350120470000547,
496
+ "bytes": 2680827,
497
+ "pages": 36,
498
+ "shards": 2
499
  },
500
  {
501
+ "seconds": 1.5306445749993145,
502
+ "bytes": 9835087,
503
+ "pages": 33,
504
  "shards": 3
505
  },
506
  {
507
+ "seconds": 0.9580864189993008,
508
+ "bytes": 989214,
509
+ "pages": 29,
510
+ "shards": 1
511
  },
512
  {
513
+ "seconds": 1.3733238169988908,
514
+ "bytes": 13560061,
515
+ "pages": 41,
516
+ "shards": 4
517
  },
518
  {
519
+ "seconds": 1.2763601959995867,
520
+ "bytes": 5814992,
521
+ "pages": 27,
522
+ "shards": 2
523
  },
524
  {
525
+ "seconds": 1.203305928996997,
526
+ "bytes": 7121753,
527
+ "pages": 36,
528
  "shards": 3
529
  },
530
  {
531
+ "seconds": 1.1241114900003595,
532
+ "bytes": 6180535,
533
+ "pages": 36,
534
  "shards": 2
535
  },
536
  {
537
+ "seconds": 1.1907369090004067,
538
+ "bytes": 6141214,
539
+ "pages": 35,
540
+ "shards": 2
541
  },
542
  {
543
+ "seconds": 0.9464336839992029,
544
+ "bytes": 2380354,
545
  "pages": 28,
546
  "shards": 2
547
  },
548
  {
549
+ "seconds": 1.6319502810001723,
550
+ "bytes": 16693932,
551
+ "pages": 40,
552
+ "shards": 4
553
  },
554
  {
555
+ "seconds": 0.91435248900234,
556
+ "bytes": 2656824,
557
  "pages": 36,
558
  "shards": 2
559
  },
560
  {
561
+ "seconds": 1.1157520460001251,
562
+ "bytes": 5879867,
563
+ "pages": 28,
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 sha256:7a080253df8bf544a1d7795cc94dbd9bed18537ff0c92c322418f47ac23af393
3
- size 721766975
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d44898a301bc80cd4b55a3e048afe4856a384a2f161cf4889c3ae4e68e12da5a
3
+ size 721775039
training_config.json CHANGED
@@ -53,7 +53,7 @@
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
- "prefetch_batches": 62,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,
 
53
  "cpu_threads": 4,
54
  "page_rows": 4096,
55
  "cortex_page_rows": 128,
56
+ "prefetch_batches": 63,
57
  "prefetch_low_water": 31,
58
  "memory_pages": 12,
59
  "cache_gb": 20,