Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75014.49488267668,
-      "required_pages_per_second_upper_estimate": 0.0065020662939313184
+      "required_token_positions_per_second": 76047.5338972513,
+      "required_pages_per_second_upper_estimate": 0.006591607497501201
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9376.811860334585,
-      "required_pages_per_second_upper_estimate": 0.0037405504469182164
+      "required_token_positions_per_second": 9505.941737156412,
+      "required_pages_per_second_upper_estimate": 0.003792062285446151
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18753.62372066917,
-      "required_pages_per_second_upper_estimate": 0.0043383207952720215
+      "required_token_positions_per_second": 19011.883474312825,
+      "required_pages_per_second_upper_estimate": 0.0043980646440610555
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28130.435581003752,
-      "required_pages_per_second_upper_estimate": 0.001712995937893913
+      "required_token_positions_per_second": 28517.825211469233,
+      "required_pages_per_second_upper_estimate": 0.0017365859339129478
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28130.435581003752,
-      "required_pages_per_second_upper_estimate": 0.002165558801055261
+      "required_token_positions_per_second": 28517.825211469233,
+      "required_pages_per_second_upper_estimate": 0.002195381127171625
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28130.435581003752,
-      "required_pages_per_second_upper_estimate": 0.34928586340444456
+      "required_token_positions_per_second": 28517.825211469233,
+      "required_pages_per_second_upper_estimate": 0.3540959461051347
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.23557111860938562,
-    "producer_p95_seconds": 0.3164607930011698,
-    "gpu_update_seconds": 0.9828500493846054,
-    "producer_batches_per_second": 4.245002553382442,
-    "gpu_batches_per_second": 1.0174492035953324,
-    "producer_headroom": 4.172200969229709,
+    "producer_batch_seconds": 0.12770199576561936,
+    "producer_p95_seconds": 0.3432198820009944,
+    "gpu_update_seconds": 0.9694988939367154,
+    "producer_batches_per_second": 7.8307311800778105,
+    "gpu_batches_per_second": 1.0314606919657565,
+    "producer_headroom": 7.591885217801187,
     "sustainable": true,
-    "buffer_seconds": 60.936703061845535,
+    "buffer_seconds": 60.10893142407635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 507009.5376736355,
-    "upload_bytes_per_second": 2651718.7494431166,
+    "generation_bytes_per_second": 523116.1927496768,
+    "upload_bytes_per_second": 2696666.7026607064,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.79244420677948,
+    "estimated_full_shards_per_hour": 112.24855744235734,
     "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": 0.9669066969981941,
-      "bytes": 1217353,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.4986801279992505,
       "bytes": 15911648,
@@ -562,6 +556,12 @@
       "bytes": 10571255,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.0644437480004854,
+      "bytes": 8538245,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 300900, "loss": 0.6143244709819555, "lr": 0.0003, "grad_norm": 0.15830787618421865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44043880447, "tokens_per_second": 149355.67107736587, "data_wait_s": 1.402800262439996e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 301000, "loss": 0.5659272521734238, "lr": 0.0003, "grad_norm": 0.1554755789510048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44058550815, "tokens_per_second": 151017.71921907787, "data_wait_s": 2.089900226565078e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 301100, "loss": 0.5770027451217175, "lr": 0.0003, "grad_norm": 0.15096028628473723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44073284767, "tokens_per_second": 150545.78282708215, "data_wait_s": 1.6408001101808622e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 301200, "loss": 0.6090783625841141, "lr": 0.0003, "grad_norm": 0.1565635430873337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44087920495, "tokens_per_second": 150003.4241925961, "data_wait_s": 1.224000152433291e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9de39448ba6b61aba1a82cef929f7841017c61a9ede782b03f2b02f28cdd888
-size 445959
+oid sha256:c06e8307bd2aa09878811c04e62644b26872d563896c858a1b20eaeaf8aad014
+size 448077