appvoid commited on
Commit
cf8541e
·
verified ·
1 Parent(s): f319dec

Qwen90M step 805000; 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. performance.json +3 -3
  6. shard_scheduler.json +44 -380
  7. trainer.pt +2 -2
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 800000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "8dbf79dba5ff97079bc140e423cf40b0ef487b5e04e4c30f7760173437d08164",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "84eb35d156f31624644b274e41f57e3691a9338e88ee83073b01c24b4744a8cf",
12
  "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
- "performance.json": "d91455eeac02a2960a60e3a2edc7d99796576b62edb422642a8beb3b824ff02d",
14
- "shard_scheduler.json": "cc275791320ae7c7d3b065796c547bbaa39e7bc94e15c07e5ba7950b8dffb861",
15
- "coverage.json": "666363f885b45a74f2b6a41ece402cf5806e0f534f2f95e854c20588602a307a",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
 
1
  {
2
+ "step": 805000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "f81353344f2caa3305d0563c6e2992bc9eca917fdb0f44f55a65bbf690de87f9",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "1232789bccab1559e521ddbb828d7a33d03698d8a2d36aeb0341e6a2e27d4c92",
12
  "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd",
13
+ "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7",
14
+ "shard_scheduler.json": "671cb6a206e18282c9946eb8a1b247f024ce1af7927d526a978429fd995d40c4",
15
+ "coverage.json": "413a5d11bd5c4ca98af7d6cdbc7f9d3951a2feb0417fce131129dc6f92afcc10",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 800000,
3
  "stats": {
4
- "climbmix/blocks": 28800000,
5
- "climbmix/targets": 48143279504,
6
- "climbmix/input_tokens": 48146330672,
7
- "ultra_qa/blocks": 10800000,
8
- "ultra_qa/targets": 17619033390,
9
- "ultra_qa/input_tokens": 17619067325,
10
- "cortex/blocks": 10800000,
11
- "cortex/targets": 410679183,
12
- "cortex/input_tokens": 17167407037,
13
- "ultra_style/blocks": 10800000,
14
- "ultra_style/targets": 17082628887,
15
- "ultra_style/input_tokens": 17083524729,
16
- "openmath/blocks": 7200000,
17
- "openmath/targets": 11125217942,
18
- "openmath/input_tokens": 11128924680,
19
- "rewrite6/blocks": 3600000,
20
- "rewrite6/targets": 5901745728,
21
- "rewrite6/input_tokens": 5908216171
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 3944,
26
- "row": 2788,
27
  "start": {
28
- "file": 190,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 75,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
- "documents": 16157412
38
  },
39
  "rewrite6": {
40
- "page": 44,
41
- "row": 113,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 180224
47
  }
48
  },
49
- "epoch": 44,
50
- "offset": 2048,
51
- "documents": 9218113
52
  },
53
  "openmath": {
54
- "page": 2465,
55
- "row": 197,
56
  "start": {
57
  "file": 23,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 53,
61
- "row": 690
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
- "documents": 10096837
67
  },
68
  "ultra_qa": {
69
- "page": 1065,
70
- "row": 3552,
71
  "start": {
72
  "file": 8,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 3,
76
- "row": 11688
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 4365792
82
  },
83
  "ultra_style": {
84
- "page": 1626,
85
- "row": 2808,
86
  "start": {
87
  "file": 9,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 5,
91
- "row": 70413
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 6662904
97
  },
98
  "cortex": {
99
- "page": 209452,
100
- "row": 75,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 26809931
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 805000,
3
  "stats": {
4
+ "climbmix/blocks": 28980000,
5
+ "climbmix/targets": 48444394688,
6
+ "climbmix/input_tokens": 48447464623,
7
+ "ultra_qa/blocks": 10867500,
8
+ "ultra_qa/targets": 17729141745,
9
+ "ultra_qa/input_tokens": 17729175912,
10
+ "cortex/blocks": 10867500,
11
+ "cortex/targets": 413246363,
12
+ "cortex/input_tokens": 17274632186,
13
+ "ultra_style/blocks": 10867500,
14
+ "ultra_style/targets": 17189552467,
15
+ "ultra_style/input_tokens": 17190453888,
16
+ "openmath/blocks": 7245000,
17
+ "openmath/targets": 11194750151,
18
+ "openmath/input_tokens": 11198479888,
19
+ "rewrite6/blocks": 3622500,
20
+ "rewrite6/targets": 5938628484,
21
+ "rewrite6/input_tokens": 5945139165
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 3969,
26
+ "row": 558,
27
  "start": {
28
+ "file": 192,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 9,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
  "offset": 2048,
37
+ "documents": 16257582
38
  },
39
  "rewrite6": {
40
+ "page": 7,
41
+ "row": 3564,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 28672
47
  }
48
  },
49
+ "epoch": 45,
50
+ "offset": 0,
51
+ "documents": 9275416
52
  },
53
  "openmath": {
54
+ "page": 2480,
55
+ "row": 1695,
56
  "start": {
57
  "file": 23,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 115,
61
+ "row": 130
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 0,
66
+ "documents": 10159775
67
  },
68
  "ultra_qa": {
69
+ "page": 1072,
70
+ "row": 2161,
71
  "start": {
72
  "file": 8,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 3,
76
+ "row": 40360
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 4393073
82
  },
83
  "ultra_style": {
84
+ "page": 1636,
85
+ "row": 3207,
86
  "start": {
87
  "file": 9,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 6,
91
+ "row": 27157
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 6704263
97
  },
98
  "cortex": {
99
+ "page": 210760,
100
+ "row": 21,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 26977301
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 800000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "8dbf79dba5ff97079bc140e423cf40b0ef487b5e04e4c30f7760173437d08164", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "84eb35d156f31624644b274e41f57e3691a9338e88ee83073b01c24b4744a8cf", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "d91455eeac02a2960a60e3a2edc7d99796576b62edb422642a8beb3b824ff02d", "shard_scheduler.json": "cc275791320ae7c7d3b065796c547bbaa39e7bc94e15c07e5ba7950b8dffb861", "coverage.json": "666363f885b45a74f2b6a41ece402cf5806e0f534f2f95e854c20588602a307a", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000800000"}
 
1
+ {"step": 805000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "f81353344f2caa3305d0563c6e2992bc9eca917fdb0f44f55a65bbf690de87f9", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "1232789bccab1559e521ddbb828d7a33d03698d8a2d36aeb0341e6a2e27d4c92", "training_config.json": "4e8f8b2fc7193aabdbca12082e7c7cc3a0ed080bd5ff16e060cc02e10ffc53dd", "performance.json": "eb45b2d9ecb051b6b7e592c18f2a0bfebf4278147b783ba131f7a9a4453a9ec7", "shard_scheduler.json": "671cb6a206e18282c9946eb8a1b247f024ce1af7927d526a978429fd995d40c4", "coverage.json": "413a5d11bd5c4ca98af7d6cdbc7f9d3951a2feb0417fce131129dc6f92afcc10", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "052aedaef69b550457da29474e17df28fa635c99594a7b9c0881f3ac3827000e", "folder": "step-000000805000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8dbf79dba5ff97079bc140e423cf40b0ef487b5e04e4c30f7760173437d08164
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f81353344f2caa3305d0563c6e2992bc9eca917fdb0f44f55a65bbf690de87f9
3
  size 360622744
performance.json CHANGED
@@ -49,11 +49,11 @@
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.9481044159999783,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9481044159999783,
56
- "input_tokens_per_second": 194408.96687059014
57
  },
58
  "measurements": [
59
  {
 
49
  "selected": {
50
  "microbatch": 90,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.9411944559999483,
53
  "peak_gb": 89.37790203094482,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9411944559999483,
56
+ "input_tokens_per_second": 195836.25766704488
57
  },
58
  "measurements": [
59
  {
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": 77170.39332101293,
26
- "required_pages_per_second_upper_estimate": 0.005338826869929831
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": 9646.299165126617,
40
- "required_pages_per_second_upper_estimate": 0.004038704022805709
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": 19292.598330253233,
54
- "required_pages_per_second_upper_estimate": 0.004018139452159805
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": 28938.89749537985,
68
- "required_pages_per_second_upper_estimate": 0.0017348427177100649
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": 28938.89749537985,
82
- "required_pages_per_second_upper_estimate": 0.0022498719915079917
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": 28938.89749537985,
96
- "required_pages_per_second_upper_estimate": 0.37119710490347546
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,409 +158,73 @@
158
  "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
- "producer_batch_seconds": 0.1137480395858006,
162
- "producer_p95_seconds": 0.34392818799824454,
163
- "gpu_update_seconds": 0.9553923056126812,
164
- "producer_batches_per_second": 8.79136030512153,
165
- "gpu_batches_per_second": 1.0466904476048846,
166
- "producer_headroom": 8.399197991381865,
167
  "sustainable": true,
168
- "buffer_seconds": 60.18971525359891,
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": 331531.55085523945,
174
- "upload_bytes_per_second": 4593959.83944083,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 71.13895315401923,
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": 1.108235388997855,
184
- "bytes": 5901441,
185
- "pages": 28,
186
- "shards": 2
187
- },
188
- {
189
- "seconds": 1.0388182249989768,
190
- "bytes": 6175327,
191
- "pages": 37,
192
- "shards": 2
193
- },
194
- {
195
- "seconds": 1.7779917830011982,
196
- "bytes": 7496140,
197
- "pages": 33,
198
- "shards": 3
199
- },
200
- {
201
- "seconds": 1.6659519919994636,
202
- "bytes": 6121395,
203
- "pages": 38,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 1.0959868669997377,
208
- "bytes": 2315562,
209
- "pages": 28,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.3342183659988223,
214
- "bytes": 11912121,
215
- "pages": 41,
216
- "shards": 3
217
- },
218
- {
219
- "seconds": 1.4696884980003233,
220
- "bytes": 5989675,
221
- "pages": 29,
222
- "shards": 2
223
- },
224
- {
225
- "seconds": 1.5069389330019476,
226
- "bytes": 6220019,
227
- "pages": 34,
228
- "shards": 3
229
- },
230
- {
231
- "seconds": 0.9491098379985488,
232
- "bytes": 6427108,
233
- "pages": 37,
234
- "shards": 2
235
  },
236
  {
237
- "seconds": 1.2226353290025145,
238
- "bytes": 7620412,
239
  "pages": 40,
240
  "shards": 3
241
  },
242
  {
243
- "seconds": 1.2109157329978188,
244
- "bytes": 5386870,
245
- "pages": 32,
246
- "shards": 2
247
- },
248
- {
249
- "seconds": 1.3874870750005357,
250
- "bytes": 11872388,
251
- "pages": 33,
252
- "shards": 3
253
- },
254
- {
255
- "seconds": 1.3592714799960959,
256
- "bytes": 2722837,
257
- "pages": 36,
258
- "shards": 2
259
- },
260
- {
261
- "seconds": 1.3202570079956786,
262
- "bytes": 6097373,
263
  "pages": 30,
264
  "shards": 2
265
  },
266
  {
267
- "seconds": 1.580753474998346,
268
- "bytes": 10855257,
269
- "pages": 30,
270
- "shards": 3
271
- },
272
- {
273
- "seconds": 1.1245584639982553,
274
- "bytes": 2759850,
275
- "pages": 40,
276
- "shards": 2
277
- },
278
- {
279
- "seconds": 1.2774146450028638,
280
- "bytes": 6152659,
281
- "pages": 37,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 1.6375758160065743,
286
- "bytes": 13187614,
287
- "pages": 36,
288
- "shards": 4
289
- },
290
- {
291
- "seconds": 1.1867789399984758,
292
- "bytes": 1047762,
293
- "pages": 32,
294
- "shards": 1
295
- },
296
- {
297
- "seconds": 2.737723589001689,
298
- "bytes": 9866131,
299
  "pages": 36,
300
- "shards": 4
301
- },
302
- {
303
- "seconds": 1.0393312469968805,
304
- "bytes": 2566382,
305
- "pages": 31,
306
  "shards": 2
307
  },
308
  {
309
- "seconds": 1.564457227003004,
310
- "bytes": 6307452,
311
  "pages": 38,
312
- "shards": 2
313
- },
314
- {
315
- "seconds": 1.4970643039996503,
316
- "bytes": 11855130,
317
- "pages": 30,
318
- "shards": 4
319
- },
320
- {
321
- "seconds": 1.1368510719985352,
322
- "bytes": 7197379,
323
- "pages": 42,
324
- "shards": 2
325
- },
326
- {
327
- "seconds": 1.1000944999977946,
328
- "bytes": 5635539,
329
- "pages": 19,
330
- "shards": 2
331
- },
332
- {
333
- "seconds": 1.2645733180033858,
334
- "bytes": 2430189,
335
- "pages": 29,
336
- "shards": 2
337
- },
338
- {
339
- "seconds": 1.1291347480000695,
340
- "bytes": 6037077,
341
- "pages": 34,
342
- "shards": 2
343
- },
344
- {
345
- "seconds": 1.49954066099599,
346
- "bytes": 11232631,
347
- "pages": 40,
348
  "shards": 3
349
  },
350
  {
351
- "seconds": 1.3497194699957618,
352
- "bytes": 2429729,
353
- "pages": 29,
354
- "shards": 2
355
- },
356
- {
357
- "seconds": 1.1435100689996034,
358
- "bytes": 625562,
359
- "pages": 19,
360
- "shards": 1
361
- },
362
- {
363
- "seconds": 1.4279748699991615,
364
- "bytes": 13559996,
365
- "pages": 51,
366
- "shards": 4
367
- },
368
- {
369
- "seconds": 1.198663317001774,
370
- "bytes": 9767212,
371
- "pages": 29,
372
- "shards": 3
373
- },
374
- {
375
- "seconds": 1.4251030520026688,
376
- "bytes": 1167175,
377
- "pages": 35,
378
- "shards": 1
379
- },
380
- {
381
- "seconds": 1.788670624999213,
382
- "bytes": 7398476,
383
- "pages": 31,
384
- "shards": 3
385
- },
386
- {
387
- "seconds": 1.539240851001523,
388
- "bytes": 6176798,
389
- "pages": 38,
390
- "shards": 2
391
- },
392
- {
393
- "seconds": 1.0939249449947965,
394
- "bytes": 5048949,
395
- "pages": 21,
396
- "shards": 2
397
- },
398
- {
399
- "seconds": 1.358410709995951,
400
- "bytes": 13879443,
401
- "pages": 50,
402
- "shards": 4
403
- },
404
- {
405
- "seconds": 1.1252228140001534,
406
- "bytes": 887251,
407
- "pages": 27,
408
- "shards": 1
409
- },
410
- {
411
- "seconds": 1.1445283560024109,
412
- "bytes": 7481150,
413
- "pages": 37,
414
- "shards": 3
415
- },
416
- {
417
- "seconds": 2.095004743998288,
418
- "bytes": 10399963,
419
- "pages": 34,
420
- "shards": 3
421
- },
422
- {
423
- "seconds": 0.994942393997917,
424
- "bytes": 1292130,
425
- "pages": 36,
426
- "shards": 1
427
- },
428
- {
429
- "seconds": 1.574394162002136,
430
- "bytes": 12830257,
431
- "pages": 24,
432
- "shards": 4
433
- },
434
- {
435
- "seconds": 1.6963619910020498,
436
- "bytes": 6365269,
437
- "pages": 45,
438
- "shards": 2
439
- },
440
- {
441
- "seconds": 1.0425699910047115,
442
- "bytes": 4497395,
443
- "pages": 30,
444
- "shards": 2
445
- },
446
- {
447
- "seconds": 1.688526213998557,
448
- "bytes": 7586011,
449
- "pages": 40,
450
- "shards": 3
451
- },
452
- {
453
- "seconds": 1.5618516569957137,
454
- "bytes": 5919633,
455
- "pages": 30,
456
- "shards": 2
457
- },
458
- {
459
- "seconds": 2.862179110001307,
460
- "bytes": 6924568,
461
- "pages": 39,
462
- "shards": 3
463
- },
464
- {
465
- "seconds": 1.8365363409975544,
466
- "bytes": 11520166,
467
- "pages": 32,
468
- "shards": 3
469
- },
470
- {
471
- "seconds": 1.1648409480039845,
472
- "bytes": 7473766,
473
- "pages": 40,
474
  "shards": 3
475
  },
476
  {
477
- "seconds": 1.0564176369953202,
478
- "bytes": 870650,
479
  "pages": 27,
480
- "shards": 1
481
- },
482
- {
483
- "seconds": 1.4489900529952138,
484
- "bytes": 10666654,
485
- "pages": 41,
486
- "shards": 4
487
- },
488
- {
489
- "seconds": 1.0202623110017157,
490
- "bytes": 2376232,
491
- "pages": 30,
492
- "shards": 2
493
- },
494
- {
495
- "seconds": 1.3351656059967354,
496
- "bytes": 6483786,
497
- "pages": 37,
498
- "shards": 2
499
- },
500
- {
501
- "seconds": 1.8462947400039411,
502
- "bytes": 11598158,
503
- "pages": 34,
504
- "shards": 3
505
- },
506
- {
507
- "seconds": 3.023687531000178,
508
- "bytes": 6495633,
509
- "pages": 34,
510
- "shards": 3
511
- },
512
- {
513
- "seconds": 1.523345616995357,
514
- "bytes": 6337115,
515
- "pages": 38,
516
  "shards": 2
517
  },
518
  {
519
- "seconds": 1.3345436189993052,
520
- "bytes": 2354584,
521
- "pages": 29,
522
- "shards": 2
523
- },
524
- {
525
- "seconds": 1.836964503992931,
526
- "bytes": 6252743,
527
- "pages": 37,
528
- "shards": 2
529
- },
530
- {
531
- "seconds": 2.1564883190003457,
532
- "bytes": 10319080,
533
- "pages": 32,
534
- "shards": 3
535
- },
536
- {
537
- "seconds": 1.3421651660028147,
538
- "bytes": 8322765,
539
  "pages": 36,
540
- "shards": 3
541
- },
542
- {
543
- "seconds": 1.1529908149968833,
544
- "bytes": 5931904,
545
- "pages": 28,
546
- "shards": 2
547
- },
548
- {
549
- "seconds": 1.4366694560012547,
550
- "bytes": 7813323,
551
- "pages": 41,
552
- "shards": 3
553
- },
554
- {
555
- "seconds": 1.6668840290003573,
556
- "bytes": 4874073,
557
- "pages": 32,
558
- "shards": 2
559
- },
560
- {
561
- "seconds": 1.9653244989967789,
562
- "bytes": 6133990,
563
- "pages": 38,
564
  "shards": 2
565
  }
566
  ],
@@ -606,7 +270,7 @@
606
  },
607
  "cache_reused": true,
608
  "resume_probe_seconds": [
609
- 3.0803219919999947,
610
- 0.10207374499998423
611
  ]
612
  }
 
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": 76909.91644078335,
26
+ "required_pages_per_second_upper_estimate": 0.005320806474965917
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": 9613.739555097918,
40
+ "required_pages_per_second_upper_estimate": 0.004025071994008668
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": 19227.479110195836,
54
+ "required_pages_per_second_upper_estimate": 0.004004576835931166
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": 28841.218665293753,
68
+ "required_pages_per_second_upper_estimate": 0.001728987020993341
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": 28841.218665293753,
82
+ "required_pages_per_second_upper_estimate": 0.0022422778921125652
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": 28841.218665293753,
96
+ "required_pages_per_second_upper_estimate": 0.36994418575048743
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
158
  "prefetch_batches": 63,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 720750,
161
+ "producer_batch_seconds": 0.0739909046796825,
162
+ "producer_p95_seconds": 0.15531819999978325,
163
+ "gpu_update_seconds": 0.9586280080900458,
164
+ "producer_batches_per_second": 13.515174660036216,
165
+ "gpu_batches_per_second": 1.0431575038083678,
166
+ "producer_headroom": 12.95602496333958,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.393564509672885,
169
  "buffer_drain_seconds": null,
170
  "shard_target_mib": 16,
171
+ "upload_pages_per_commit": 78,
172
  "max_pending_pages": 4096,
173
+ "generation_bytes_per_second": 383905.7653317095,
174
+ "upload_bytes_per_second": 3532908.8334232424,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 82.37724037135567,
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": 1.6723634379995929,
184
+ "bytes": 50251,
185
+ "pages": 1,
186
+ "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
187
  },
188
  {
189
+ "seconds": 2.1410375759996896,
190
+ "bytes": 10006036,
191
  "pages": 40,
192
  "shards": 3
193
  },
194
  {
195
+ "seconds": 1.5198384699997405,
196
+ "bytes": 2430008,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
197
  "pages": 30,
198
  "shards": 2
199
  },
200
  {
201
+ "seconds": 1.2644712469991646,
202
+ "bytes": 6316580,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
203
  "pages": 36,
 
 
 
 
 
 
204
  "shards": 2
205
  },
206
  {
207
+ "seconds": 1.6103071739998995,
208
+ "bytes": 7727369,
209
  "pages": 38,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
210
  "shards": 3
211
  },
212
  {
213
+ "seconds": 1.5575974499997756,
214
+ "bytes": 11152630,
215
+ "pages": 33,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
216
  "shards": 3
217
  },
218
  {
219
+ "seconds": 1.9014979050007241,
220
+ "bytes": 5893171,
221
  "pages": 27,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 1.375044964999688,
226
+ "bytes": 2500711,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
227
  "pages": 36,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
228
  "shards": 2
229
  }
230
  ],
 
270
  },
271
  "cache_reused": true,
272
  "resume_probe_seconds": [
273
+ 3.328129581999974,
274
+ 0.10924991900014902
275
  ]
276
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:84eb35d156f31624644b274e41f57e3691a9338e88ee83073b01c24b4744a8cf
3
- size 721788351
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1232789bccab1559e521ddbb828d7a33d03698d8a2d36aeb0341e6a2e27d4c92
3
+ size 721769471