Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77379.11304063832,
-      "required_pages_per_second_upper_estimate": 0.005353266584429988
+      "required_token_positions_per_second": 77219.54048268082,
+      "required_pages_per_second_upper_estimate": 0.005342226984611141
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9672.38913007979,
-      "required_pages_per_second_upper_estimate": 0.004049627346311181
+      "required_token_positions_per_second": 9652.442560335103,
+      "required_pages_per_second_upper_estimate": 0.0040412761340908225
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19344.77826015958,
-      "required_pages_per_second_upper_estimate": 0.004029007155482008
+      "required_token_positions_per_second": 19304.885120670206,
+      "required_pages_per_second_upper_estimate": 0.004020698466579207
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29017.16739023937,
-      "required_pages_per_second_upper_estimate": 0.0017395348783956835
+      "required_token_positions_per_second": 28957.32768100531,
+      "required_pages_per_second_upper_estimate": 0.0017359475791970471
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29017.16739023937,
-      "required_pages_per_second_upper_estimate": 0.002255957131560436
+      "required_token_positions_per_second": 28957.32768100531,
+      "required_pages_per_second_upper_estimate": 0.002251304857374552
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29017.16739023937,
-      "required_pages_per_second_upper_estimate": 0.37220106707506795
+      "required_token_positions_per_second": 28957.32768100531,
+      "required_pages_per_second_upper_estimate": 0.37143350753588733
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05875075965617782,
-    "producer_p95_seconds": 0.21500816299885628,
-    "gpu_update_seconds": 0.9528152637428035,
-    "producer_batches_per_second": 17.021056508072693,
-    "gpu_batches_per_second": 1.0495213899826161,
-    "producer_headroom": 16.217922445920443,
+    "producer_batch_seconds": 0.24815502786728416,
+    "producer_p95_seconds": 0.23936947499896633,
+    "gpu_update_seconds": 0.9547842364658473,
+    "producer_batches_per_second": 4.02973902481158,
+    "gpu_batches_per_second": 1.0473570486474721,
+    "producer_headroom": 3.847531297961352,
     "sustainable": true,
-    "buffer_seconds": 60.027361615796615,
+    "buffer_seconds": 60.15140689734838,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445203.7150304803,
-    "upload_bytes_per_second": 5525029.803259725,
+    "generation_bytes_per_second": 430044.96531152044,
+    "upload_bytes_per_second": 5540210.133045894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.53035343347365,
+    "estimated_full_shards_per_hour": 92.27763862141809,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 0.7847848540000086,
-      "bytes": 603387,
-      "pages": 17,
-      "shards": 1
-    },
     {
       "seconds": 1.5595179319998351,
       "bytes": 10763416,
@@ -562,6 +556,12 @@
       "bytes": 5946250,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 0.9391229799985013,
+      "bytes": 2632765,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -107,3 +107,4 @@
 {"step": 567800, "loss": 0.5703026652336121, "lr": 0.0003, "grad_norm": 0.16510065624460302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83109136146, "tokens_per_second": 153065.39365362955, "data_wait_s": 1.9455999790807255e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 567900, "loss": 0.5990357398986816, "lr": 0.0003, "grad_norm": 0.15096866719414792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83123804273, "tokens_per_second": 153176.82087323375, "data_wait_s": 8.633000106783584e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 568000, "loss": 0.5760231018066406, "lr": 0.0003, "grad_norm": 0.14936431707598977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83138360447, "tokens_per_second": 151341.9966538106, "data_wait_s": 1.1370999345672317e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 568100, "loss": 0.5821261405944824, "lr": 0.0003, "grad_norm": 0.19712639032358825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83152926119, "tokens_per_second": 151624.50268147435, "data_wait_s": 1.1381998774595559e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0c79205b453ad00bc598ebd5585800eecd168259acf5535cd4bb85d931455dd
-size 232593
+oid sha256:9f1c10e87e9cdcf0bf605f5e4c97233686005d3ec1ae9bb25d89e08617bd3eec
+size 234711