appvoid commited on
Commit
c532528
·
verified ·
1 Parent(s): 1abb80c

Qwen90M step 45000; complete resumable checkpoint

Browse files
complete.json CHANGED
@@ -1,18 +1,18 @@
1
  {
2
- "step": 40000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
- "model.safetensors": "1f54edcf173e14eaa220773d6904ca3660dca9333bec4cca80621f41c11ae4ba",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
- "trainer.pt": "ddb86ed5b8deb7d4e788e14749c412fa48be7f3cefd33dadef82d54b10488b02",
12
- "training_config.json": "07f331dbc4eba0ea92b45763c59c20af4874302e74a327f3562642d69645fed2",
13
- "performance.json": "b95a1befd575f8d9a02b43fd0cc41c54243f5297127b93d2656e0292ba15e77e",
14
- "shard_scheduler.json": "dd35db67300912a92de8d12b39d40858b1385a9f601def538c82db10aecb48b5",
15
- "coverage.json": "496553c7362fb891e4a087907a1ee7bc57bd29676eca9ccbf872f730f0c086c9",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
 
1
  {
2
+ "step": 45000,
3
  "files": {
4
  "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
5
  "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
6
+ "model.safetensors": "8538b491fbf01a73501462b1b8b4d11ed8192e49c9b77a3ed172e5e05d009d88",
7
  "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
8
  "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
9
  "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
10
  "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
11
+ "trainer.pt": "d5d9f25a02093a0b4efb7c3de3f9c0397ade770ae17612d84206b659abe4a816",
12
+ "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
13
+ "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f",
14
+ "shard_scheduler.json": "1fa5611cb9e27a2e65056abc0c9c52bc5694633d8adb64378769c8274e989bc6",
15
+ "coverage.json": "e0f2cf7ac209336f996ae01d22aab46c0dbcfe6c255e055d3cc863bf5d710ea4",
16
  "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
17
  },
18
  "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"
coverage.json CHANGED
@@ -1,107 +1,107 @@
1
  {
2
- "step": 40000,
3
  "stats": {
4
- "climbmix/blocks": 1440000,
5
- "climbmix/targets": 2405645996,
6
- "climbmix/input_tokens": 2405798466,
7
- "ultra_qa/blocks": 540000,
8
- "ultra_qa/targets": 881103422,
9
- "ultra_qa/input_tokens": 881105161,
10
- "cortex/blocks": 540000,
11
- "cortex/targets": 20545868,
12
- "cortex/input_tokens": 858494555,
13
- "ultra_style/blocks": 540000,
14
- "ultra_style/targets": 853948670,
15
- "ultra_style/input_tokens": 853993129,
16
- "openmath/blocks": 360000,
17
- "openmath/targets": 556374754,
18
- "openmath/input_tokens": 556560449,
19
- "rewrite6/blocks": 180000,
20
- "rewrite6/targets": 295083056,
21
- "rewrite6/input_tokens": 295406501
22
  },
23
  "sources": {
24
  "climbmix": {
25
- "page": 197,
26
- "row": 1098,
27
  "start": {
28
- "file": 9,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
- "row_group": 43,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
- "offset": 4096,
37
- "documents": 808010
38
  },
39
  "rewrite6": {
40
- "page": 12,
41
- "row": 807,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
- "row": 49152
47
  }
48
  },
49
  "epoch": 2,
50
  "offset": 0,
51
- "documents": 460767
52
  },
53
  "openmath": {
54
- "page": 123,
55
- "row": 1391,
56
  "start": {
57
  "file": 1,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
- "row_group": 67,
61
- "row": 158
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 2048,
66
- "documents": 505199
67
  },
68
  "ultra_qa": {
69
- "page": 53,
70
- "row": 1357,
71
  "start": {
72
  "file": 0,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 3,
76
- "row": 23181
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
- "documents": 218445
82
  },
83
  "ultra_style": {
84
- "page": 81,
85
- "row": 1172,
86
  "start": {
87
  "file": 0,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
- "row_group": 3,
91
- "row": 76752
92
  }
93
  },
94
  "epoch": 0,
95
- "offset": 0,
96
- "documents": 332948
97
  },
98
  "cortex": {
99
- "page": 10471,
100
- "row": 124,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
- "documents": 1340412
105
  }
106
  }
107
  }
 
1
  {
2
+ "step": 45000,
3
  "stats": {
4
+ "climbmix/blocks": 1620000,
5
+ "climbmix/targets": 2706557052,
6
+ "climbmix/input_tokens": 2706728542,
7
+ "ultra_qa/blocks": 607500,
8
+ "ultra_qa/targets": 991206075,
9
+ "ultra_qa/input_tokens": 991208036,
10
+ "cortex/blocks": 607500,
11
+ "cortex/targets": 23108530,
12
+ "cortex/input_tokens": 965718686,
13
+ "ultra_style/blocks": 607500,
14
+ "ultra_style/targets": 960329583,
15
+ "ultra_style/input_tokens": 960379775,
16
+ "openmath/blocks": 405000,
17
+ "openmath/targets": 625892819,
18
+ "openmath/input_tokens": 626101608,
19
+ "rewrite6/blocks": 202500,
20
+ "rewrite6/targets": 331937575,
21
+ "rewrite6/input_tokens": 332301831
22
  },
23
  "sources": {
24
  "climbmix": {
25
+ "page": 221,
26
+ "row": 3842,
27
  "start": {
28
+ "file": 10,
29
  "reader": {
30
  "format": "parquet-row-v1",
31
+ "row_group": 56,
32
  "row": 1024
33
  }
34
  },
35
  "epoch": 0,
36
+ "offset": 2048,
37
+ "documents": 909058
38
  },
39
  "rewrite6": {
40
+ "page": 26,
41
+ "row": 1558,
42
  "start": {
43
  "file": 0,
44
  "reader": {
45
  "format": "jsonl-row-v1",
46
+ "row": 106496
47
  }
48
  },
49
  "epoch": 2,
50
  "offset": 0,
51
+ "documents": 518862
52
  },
53
  "openmath": {
54
+ "page": 138,
55
+ "row": 2902,
56
  "start": {
57
  "file": 1,
58
  "reader": {
59
  "format": "parquet-row-v1",
60
+ "row_group": 128,
61
+ "row": 598
62
  }
63
  },
64
  "epoch": 0,
65
  "offset": 2048,
66
+ "documents": 568150
67
  },
68
  "ultra_qa": {
69
+ "page": 60,
70
+ "row": 12,
71
  "start": {
72
  "file": 0,
73
  "reader": {
74
  "format": "parquet-row-v1",
75
  "row_group": 3,
76
+ "row": 51853
77
  }
78
  },
79
  "epoch": 0,
80
  "offset": 2048,
81
+ "documents": 245772
82
  },
83
  "ultra_style": {
84
+ "page": 91,
85
+ "row": 2493,
86
  "start": {
87
  "file": 0,
88
  "reader": {
89
  "format": "parquet-row-v1",
90
+ "row_group": 4,
91
+ "row": 32715
92
  }
93
  },
94
  "epoch": 0,
95
+ "offset": 2048,
96
+ "documents": 375229
97
  },
98
  "cortex": {
99
+ "page": 11778,
100
+ "row": 67,
101
  "start": null,
102
  "epoch": 0,
103
  "offset": 0,
104
+ "documents": 1507651
105
  }
106
  }
107
  }
latest.json CHANGED
@@ -1 +1 @@
1
- {"step": 40000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "1f54edcf173e14eaa220773d6904ca3660dca9333bec4cca80621f41c11ae4ba", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "ddb86ed5b8deb7d4e788e14749c412fa48be7f3cefd33dadef82d54b10488b02", "training_config.json": "07f331dbc4eba0ea92b45763c59c20af4874302e74a327f3562642d69645fed2", "performance.json": "b95a1befd575f8d9a02b43fd0cc41c54243f5297127b93d2656e0292ba15e77e", "shard_scheduler.json": "dd35db67300912a92de8d12b39d40858b1385a9f601def538c82db10aecb48b5", "coverage.json": "496553c7362fb891e4a087907a1ee7bc57bd29676eca9ccbf872f730f0c086c9", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000040000"}
 
1
+ {"step": 45000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "8538b491fbf01a73501462b1b8b4d11ed8192e49c9b77a3ed172e5e05d009d88", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "d5d9f25a02093a0b4efb7c3de3f9c0397ade770ae17612d84206b659abe4a816", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "1fa5611cb9e27a2e65056abc0c9c52bc5694633d8adb64378769c8274e989bc6", "coverage.json": "e0f2cf7ac209336f996ae01d22aab46c0dbcfe6c255e055d3cc863bf5d710ea4", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000045000"}
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1f54edcf173e14eaa220773d6904ca3660dca9333bec4cca80621f41c11ae4ba
3
  size 360622744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8538b491fbf01a73501462b1b8b4d11ed8192e49c9b77a3ed172e5e05d009d88
3
  size 360622744
performance.json CHANGED
@@ -44,16 +44,16 @@
44
  "source": "e6d135a4487a9fd1907e12903cda79e75fd61305fad149fda7ec5281279dc18b",
45
  "requested_microbatch_ceiling": 90
46
  },
47
- "cache_reused": false,
48
  "attention": "PyTorch FlashAttention (forced; forward/backward verified)",
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
- "microbatch_seconds": 0.8818629400000191,
53
- "peak_gb": 84.78368663787842,
54
  "fits_budget": true,
55
- "estimated_update_seconds": 0.9349941874999956,
56
- "input_tokens_per_second": 199722.64624250587
57
  },
58
  "measurements": [
59
  {
@@ -218,5 +218,6 @@
218
  "selection_policy": "max_batch",
219
  "budget_gb": 89.6964394569397,
220
  "reserve_gb": 2.1791439056396484,
221
- "note": "Forward/backward throughput; excludes optimizer, data, logging and checkpoint time."
 
222
  }
 
44
  "source": "e6d135a4487a9fd1907e12903cda79e75fd61305fad149fda7ec5281279dc18b",
45
  "requested_microbatch_ceiling": 90
46
  },
47
+ "cache_reused": true,
48
  "attention": "PyTorch FlashAttention (forced; forward/backward verified)",
49
  "selected": {
50
  "microbatch": 86,
51
  "mode": "compiled",
52
+ "microbatch_seconds": 0.8974938420000171,
53
+ "peak_gb": 85.39913082122803,
54
  "fits_budget": true,
55
+ "estimated_update_seconds": 0.9533413040000482,
56
+ "input_tokens_per_second": 196244.2434228943
57
  },
58
  "measurements": [
59
  {
 
218
  "selection_policy": "max_batch",
219
  "budget_gb": 89.6964394569397,
220
  "reserve_gb": 2.1791439056396484,
221
+ "note": "Forward/backward throughput; excludes optimizer, data, logging and checkpoint time.",
222
+ "legacy_seed_verified": false
223
  }
shard_scheduler.json CHANGED
@@ -22,8 +22,8 @@
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
- "required_token_positions_per_second": 75414.82490767069,
26
- "required_pages_per_second_upper_estimate": 0.006536765885870627
27
  },
28
  {
29
  "source": "rewrite6",
@@ -36,8 +36,8 @@
36
  "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
- "required_token_positions_per_second": 9426.853113458836,
40
- "required_pages_per_second_upper_estimate": 0.003760512650972888
41
  },
42
  {
43
  "source": "openmath",
@@ -50,8 +50,8 @@
50
  "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
- "required_token_positions_per_second": 18853.70622691767,
54
- "required_pages_per_second_upper_estimate": 0.004361473121700661
55
  },
56
  {
57
  "source": "ultra_qa",
@@ -64,8 +64,8 @@
64
  "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
- "required_token_positions_per_second": 28280.55934037651,
68
- "required_pages_per_second_upper_estimate": 0.0017221376872012205
69
  },
70
  {
71
  "source": "ultra_style",
@@ -78,8 +78,8 @@
78
  "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
- "required_token_positions_per_second": 28280.55934037651,
82
- "required_pages_per_second_upper_estimate": 0.002177115743620939
83
  },
84
  {
85
  "source": "cortex",
@@ -92,8 +92,8 @@
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
  "estimated_pages_for_buffer": 22,
95
- "required_token_positions_per_second": 28280.55934037651,
96
- "required_pages_per_second_upper_estimate": 0.3511498980639521
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,410 +158,140 @@
158
  "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
- "producer_batch_seconds": 0.1152253453594767,
162
- "producer_p95_seconds": 0.4034826810020604,
163
- "gpu_update_seconds": 0.9776327146587446,
164
- "producer_batches_per_second": 8.678646150985522,
165
- "gpu_batches_per_second": 1.0228790270680161,
166
- "producer_headroom": 8.48452839615064,
167
  "sustainable": true,
168
- "buffer_seconds": 60.61322830884216,
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": 346780.9633111328,
174
- "upload_bytes_per_second": 3718135.8829697864,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
- "estimated_full_shards_per_hour": 74.41112207890022,
179
  "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
180
  },
181
  "upload_measurements": [
182
  {
183
- "seconds": 0.9655926410050597,
184
- "bytes": 5648372,
185
- "pages": 25,
186
- "shards": 2
187
- },
188
- {
189
- "seconds": 2.317049978002615,
190
- "bytes": 7094864,
191
- "pages": 41,
192
- "shards": 2
193
- },
194
- {
195
- "seconds": 1.0156349210010376,
196
- "bytes": 12015881,
197
- "pages": 35,
198
- "shards": 4
199
- },
200
- {
201
- "seconds": 1.0830463239981327,
202
- "bytes": 6195752,
203
- "pages": 36,
204
- "shards": 2
205
- },
206
- {
207
- "seconds": 0.7641328610043274,
208
- "bytes": 2288843,
209
- "pages": 28,
210
- "shards": 2
211
- },
212
- {
213
- "seconds": 1.014886224998918,
214
- "bytes": 6118927,
215
- "pages": 37,
216
- "shards": 2
217
  },
218
  {
219
- "seconds": 0.9393350060054217,
220
- "bytes": 807847,
221
- "pages": 26,
222
  "shards": 1
223
  },
224
  {
225
- "seconds": 1.4682032510027057,
226
- "bytes": 16949786,
227
- "pages": 39,
228
  "shards": 5
229
  },
230
  {
231
- "seconds": 0.9623997290036641,
232
- "bytes": 6092433,
233
- "pages": 37,
234
- "shards": 2
235
- },
236
- {
237
- "seconds": 0.7187483379966579,
238
- "bytes": 1132933,
239
- "pages": 34,
240
  "shards": 1
241
  },
242
  {
243
- "seconds": 1.4302320659990073,
244
- "bytes": 11751486,
245
  "pages": 35,
246
- "shards": 4
247
- },
248
- {
249
- "seconds": 0.8936330840006121,
250
- "bytes": 843076,
251
- "pages": 27,
252
- "shards": 1
253
- },
254
- {
255
- "seconds": 1.0585707080026623,
256
- "bytes": 7490877,
257
- "pages": 34,
258
- "shards": 3
259
- },
260
- {
261
- "seconds": 33.798858237001696,
262
- "bytes": 12238534,
263
- "pages": 38,
264
- "shards": 3
265
- },
266
- {
267
- "seconds": 0.911944024002878,
268
- "bytes": 5136667,
269
- "pages": 41,
270
  "shards": 2
271
  },
272
  {
273
- "seconds": 1.2792566019998048,
274
- "bytes": 7551543,
275
- "pages": 38,
276
- "shards": 3
277
- },
278
- {
279
- "seconds": 0.9091981460005627,
280
- "bytes": 5814772,
281
- "pages": 27,
282
- "shards": 2
283
- },
284
- {
285
- "seconds": 1.0604712630010908,
286
- "bytes": 6743320,
287
- "pages": 36,
288
- "shards": 4
289
- },
290
- {
291
- "seconds": 1.0128854309950839,
292
- "bytes": 6059900,
293
- "pages": 26,
294
- "shards": 2
295
- },
296
- {
297
- "seconds": 1.0482589550010744,
298
- "bytes": 7056720,
299
- "pages": 42,
300
- "shards": 2
301
- },
302
- {
303
- "seconds": 1.9086563699966064,
304
- "bytes": 7678172,
305
- "pages": 33,
306
- "shards": 3
307
- },
308
- {
309
- "seconds": 1.1078665649984032,
310
- "bytes": 9761910,
311
- "pages": 33,
312
- "shards": 3
313
- },
314
- {
315
- "seconds": 0.90664235100121,
316
- "bytes": 2695308,
317
- "pages": 37,
318
- "shards": 2
319
- },
320
- {
321
- "seconds": 0.7943546639944543,
322
- "bytes": 258770,
323
- "pages": 8,
324
- "shards": 1
325
- },
326
- {
327
- "seconds": 1.1189483790003578,
328
- "bytes": 5921688,
329
- "pages": 28,
330
- "shards": 2
331
- },
332
- {
333
- "seconds": 1.2605290909996256,
334
- "bytes": 6824656,
335
- "pages": 33,
336
- "shards": 2
337
- },
338
- {
339
- "seconds": 11.391802897000161,
340
- "bytes": 12565360,
341
- "pages": 37,
342
- "shards": 4
343
- },
344
- {
345
- "seconds": 0.941881225000543,
346
- "bytes": 6023516,
347
- "pages": 37,
348
- "shards": 2
349
- },
350
- {
351
- "seconds": 0.9670609070017235,
352
- "bytes": 2689085,
353
  "pages": 35,
354
- "shards": 2
355
- },
356
- {
357
- "seconds": 1.1650014270053362,
358
- "bytes": 5886860,
359
- "pages": 31,
360
- "shards": 2
361
- },
362
- {
363
- "seconds": 1.1598776869941503,
364
- "bytes": 10363400,
365
- "pages": 37,
366
  "shards": 3
367
  },
368
  {
369
- "seconds": 1.1336681269967812,
370
- "bytes": 8324153,
371
  "pages": 31,
372
- "shards": 3
373
- },
374
- {
375
- "seconds": 1.0718341040046653,
376
- "bytes": 6483305,
377
- "pages": 36,
378
  "shards": 2
379
  },
380
  {
381
- "seconds": 0.9573828110005707,
382
- "bytes": 873310,
383
- "pages": 27,
384
- "shards": 1
385
- },
386
- {
387
- "seconds": 1.2830225369980326,
388
- "bytes": 12398676,
389
- "pages": 40,
390
- "shards": 4
391
- },
392
- {
393
- "seconds": 1.2898951360039064,
394
- "bytes": 6161152,
395
- "pages": 36,
396
- "shards": 2
397
- },
398
- {
399
- "seconds": 0.9979428110018489,
400
- "bytes": 2470155,
401
- "pages": 30,
402
- "shards": 2
403
- },
404
- {
405
- "seconds": 1.098158920998685,
406
- "bytes": 11654949,
407
- "pages": 31,
408
- "shards": 3
409
- },
410
- {
411
- "seconds": 1.0713990720032598,
412
- "bytes": 5874732,
413
- "pages": 36,
414
- "shards": 2
415
- },
416
- {
417
- "seconds": 0.9352636539988453,
418
- "bytes": 6076485,
419
- "pages": 31,
420
- "shards": 3
421
- },
422
- {
423
- "seconds": 1.2692730029957602,
424
- "bytes": 6086693,
425
  "pages": 37,
426
- "shards": 2
427
- },
428
- {
429
- "seconds": 1.036868928000331,
430
- "bytes": 2682212,
431
- "pages": 38,
432
- "shards": 2
433
- },
434
- {
435
- "seconds": 1.1587374009977793,
436
- "bytes": 10973458,
437
- "pages": 30,
438
  "shards": 3
439
  },
440
  {
441
- "seconds": 0.8646178900016821,
442
- "bytes": 947338,
443
- "pages": 27,
444
- "shards": 1
445
- },
446
- {
447
- "seconds": 1.0228583920033998,
448
- "bytes": 13452885,
449
- "pages": 43,
450
- "shards": 4
451
- },
452
- {
453
- "seconds": 1.0817572929954622,
454
- "bytes": 6130820,
455
- "pages": 28,
456
- "shards": 2
457
- },
458
- {
459
- "seconds": 0.9214774369975203,
460
- "bytes": 1243052,
461
  "pages": 35,
462
- "shards": 1
463
- },
464
- {
465
- "seconds": 1.3106369909946807,
466
- "bytes": 11452061,
467
- "pages": 33,
468
- "shards": 4
469
- },
470
- {
471
- "seconds": 1.0961634519990184,
472
- "bytes": 6066738,
473
- "pages": 37,
474
  "shards": 2
475
  },
476
  {
477
- "seconds": 0.9469924939985503,
478
- "bytes": 890694,
479
- "pages": 27,
480
  "shards": 2
481
  },
482
  {
483
- "seconds": 1.2476149669964798,
484
- "bytes": 13021486,
485
- "pages": 30,
486
  "shards": 4
487
  },
488
  {
489
- "seconds": 0.8867707609970239,
490
- "bytes": 6143048,
491
- "pages": 43,
492
- "shards": 2
493
- },
494
- {
495
- "seconds": 1.05023945299763,
496
- "bytes": 7400872,
497
- "pages": 31,
498
- "shards": 3
499
- },
500
- {
501
- "seconds": 1.1047442850031075,
502
- "bytes": 6203278,
503
- "pages": 37,
504
- "shards": 2
505
- },
506
- {
507
- "seconds": 1.0125872680000612,
508
- "bytes": 1072664,
509
- "pages": 31,
510
  "shards": 1
511
  },
512
  {
513
- "seconds": 1.132626735001395,
514
- "bytes": 11459123,
515
- "pages": 36,
516
- "shards": 4
517
- },
518
- {
519
- "seconds": 1.060930259001907,
520
- "bytes": 6808205,
521
- "pages": 34,
522
  "shards": 2
523
  },
524
  {
525
- "seconds": 3.3303450339953997,
526
- "bytes": 5952538,
527
  "pages": 29,
528
- "shards": 2
529
  },
530
  {
531
- "seconds": 0.9529956519982079,
532
- "bytes": 7196338,
533
- "pages": 30,
534
  "shards": 3
535
  },
536
  {
537
- "seconds": 1.1464728400023887,
538
- "bytes": 6003121,
539
  "pages": 37,
540
- "shards": 2
541
- },
542
- {
543
- "seconds": 0.9798986519963364,
544
- "bytes": 7781027,
545
- "pages": 39,
546
  "shards": 3
547
  },
548
  {
549
- "seconds": 0.9489700769991032,
550
- "bytes": 911408,
551
- "pages": 26,
552
  "shards": 1
553
  },
554
  {
555
- "seconds": 1.3251626939963899,
556
- "bytes": 11870118,
557
- "pages": 36,
558
  "shards": 3
559
  },
560
  {
561
- "seconds": 1.1823667450007633,
562
- "bytes": 10876587,
563
- "pages": 32,
564
- "shards": 4
565
  }
566
  ],
567
  "cache_key": {
@@ -604,5 +334,9 @@
604
  "cpu_count": 20,
605
  "microbatch": 86
606
  },
607
- "cache_reused": false
 
 
 
 
608
  }
 
22
  "tokens_per_second": 1062967.6168812835,
23
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
24
  "estimated_pages_for_buffer": 1,
25
+ "required_token_positions_per_second": 75734.50043516197,
26
+ "required_pages_per_second_upper_estimate": 0.006564474550383355
27
  },
28
  {
29
  "source": "rewrite6",
 
36
  "tokens_per_second": 1206672.263377472,
37
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
38
  "estimated_pages_for_buffer": 1,
39
+ "required_token_positions_per_second": 9466.812554395246,
40
+ "required_pages_per_second_upper_estimate": 0.0037764530694093055
41
  },
42
  {
43
  "source": "openmath",
 
50
  "tokens_per_second": 410042.7581134458,
51
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
52
  "estimated_pages_for_buffer": 1,
53
+ "required_token_positions_per_second": 18933.625108790493,
54
+ "required_pages_per_second_upper_estimate": 0.004379960948497657
55
  },
56
  {
57
  "source": "ultra_qa",
 
64
  "tokens_per_second": 2163486.5624068812,
65
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
66
  "estimated_pages_for_buffer": 1,
67
+ "required_token_positions_per_second": 28400.437663185738,
68
+ "required_pages_per_second_upper_estimate": 0.0017294376481073507
69
  },
70
  {
71
  "source": "ultra_style",
 
78
  "tokens_per_second": 1546694.2737633225,
79
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
80
  "estimated_pages_for_buffer": 1,
81
+ "required_token_positions_per_second": 28400.437663185738,
82
+ "required_pages_per_second_upper_estimate": 0.00218634430875523
83
  },
84
  {
85
  "source": "cortex",
 
92
  "tokens_per_second": 1015177.8395898147,
93
  "note": "Bounded fresh sample; full-page latency is an extrapolation.",
94
  "estimated_pages_for_buffer": 22,
95
+ "required_token_positions_per_second": 28400.437663185738,
96
+ "required_pages_per_second_upper_estimate": 0.3526383856263052
97
  }
98
  ],
99
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
 
158
  "prefetch_batches": 62,
159
  "prefetch_capacity_limit": 512,
160
  "batch_memory_estimate_bytes": 688946,
161
+ "producer_batch_seconds": 0.053367566671880695,
162
+ "producer_p95_seconds": 0.17484046899971872,
163
+ "gpu_update_seconds": 0.9735061243735307,
164
+ "producer_batches_per_second": 18.737972562029867,
165
+ "gpu_batches_per_second": 1.0272149039057341,
166
+ "producer_headroom": 18.241531047479253,
167
  "sustainable": true,
168
+ "buffer_seconds": 60.3573797111589,
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": 449162.1108758026,
174
+ "upload_bytes_per_second": 6093636.035551512,
175
  "upload_sustainable": true,
176
  "logical_page_rows": 4096,
177
  "logical_cortex_page_rows": 128,
178
+ "estimated_full_shards_per_hour": 96.37973303514059,
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": 1.0632401930001834,
184
+ "bytes": 22792,
185
+ "pages": 1,
186
+ "shards": 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
187
  },
188
  {
189
+ "seconds": 0.9811290180000469,
190
+ "bytes": 1073130,
191
+ "pages": 29,
192
  "shards": 1
193
  },
194
  {
195
+ "seconds": 1.3945427510002446,
196
+ "bytes": 16977562,
197
+ "pages": 43,
198
  "shards": 5
199
  },
200
  {
201
+ "seconds": 0.7539118750000853,
202
+ "bytes": 789098,
203
+ "pages": 25,
 
 
 
 
 
 
204
  "shards": 1
205
  },
206
  {
207
+ "seconds": 0.9344837560001906,
208
+ "bytes": 6249982,
209
  "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
210
  "shards": 2
211
  },
212
  {
213
+ "seconds": 1.0289287979999244,
214
+ "bytes": 7839831,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
215
  "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
216
  "shards": 3
217
  },
218
  {
219
+ "seconds": 1.02437976300007,
220
+ "bytes": 6057442,
221
  "pages": 31,
 
 
 
 
 
 
222
  "shards": 2
223
  },
224
  {
225
+ "seconds": 1.1024301279999236,
226
+ "bytes": 7712415,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
227
  "pages": 37,
 
 
 
 
 
 
 
 
 
 
 
 
228
  "shards": 3
229
  },
230
  {
231
+ "seconds": 0.9960975080002754,
232
+ "bytes": 6877408,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
233
  "pages": 35,
 
 
 
 
 
 
 
 
 
 
 
 
234
  "shards": 2
235
  },
236
  {
237
+ "seconds": 1.0195401410001068,
238
+ "bytes": 6014632,
239
+ "pages": 34,
240
  "shards": 2
241
  },
242
  {
243
+ "seconds": 1.221793560000151,
244
+ "bytes": 10970126,
245
+ "pages": 34,
246
  "shards": 4
247
  },
248
  {
249
+ "seconds": 0.999679897000533,
250
+ "bytes": 1109612,
251
+ "pages": 33,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
252
  "shards": 1
253
  },
254
  {
255
+ "seconds": 0.9920720150003035,
256
+ "bytes": 6122911,
257
+ "pages": 35,
 
 
 
 
 
 
258
  "shards": 2
259
  },
260
  {
261
+ "seconds": 1.0328473609997673,
262
+ "bytes": 7480746,
263
  "pages": 29,
264
+ "shards": 3
265
  },
266
  {
267
+ "seconds": 1.141071416000159,
268
+ "bytes": 10868715,
269
+ "pages": 32,
270
  "shards": 3
271
  },
272
  {
273
+ "seconds": 1.1746170310007074,
274
+ "bytes": 7476573,
275
  "pages": 37,
 
 
 
 
 
 
276
  "shards": 3
277
  },
278
  {
279
+ "seconds": 1.0039610859994355,
280
+ "bytes": 1092998,
281
+ "pages": 34,
282
  "shards": 1
283
  },
284
  {
285
+ "seconds": 1.1609742079999705,
286
+ "bytes": 10556650,
287
+ "pages": 33,
288
  "shards": 3
289
  },
290
  {
291
+ "seconds": 1.1113881370001764,
292
+ "bytes": 7415466,
293
+ "pages": 37,
294
+ "shards": 3
295
  }
296
  ],
297
  "cache_key": {
 
334
  "cpu_count": 20,
335
  "microbatch": 86
336
  },
337
+ "cache_reused": true,
338
+ "resume_probe_seconds": [
339
+ 2.657314213999996,
340
+ 0.016167582000036873
341
+ ]
342
  }
trainer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ddb86ed5b8deb7d4e788e14749c412fa48be7f3cefd33dadef82d54b10488b02
3
- size 721776959
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d5d9f25a02093a0b4efb7c3de3f9c0397ade770ae17612d84206b659abe4a816
3
+ size 721769471
training_config.json CHANGED
@@ -71,5 +71,7 @@
71
  "hub_resume": true,
72
  "work_dir": "/marimo/qwen90m_workspace/runs/void-byte-v1",
73
  "data_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
74
- "trainer_recipe": "bc45094ca6fe6129cd1201f38cc5978b80014fa21be8d01ea684cb56d8d1644b"
 
 
75
  }
 
71
  "hub_resume": true,
72
  "work_dir": "/marimo/qwen90m_workspace/runs/void-byte-v1",
73
  "data_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
74
+ "trainer_recipe": "bc45094ca6fe6129cd1201f38cc5978b80014fa21be8d01ea684cb56d8d1644b",
75
+ "archive_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658",
76
+ "cortex_archive_recipe": "c0b0679d54277afcf4c85726fd0998265697a759188400782be4a859641ba658"
77
  }