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": 75616.3776869974,
-      "required_pages_per_second_upper_estimate": 0.006554235969951817
+      "required_token_positions_per_second": 76323.61804239494,
+      "required_pages_per_second_upper_estimate": 0.006615537771473384
     },
     {
       "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": 9452.047210874674,
-      "required_pages_per_second_upper_estimate": 0.0037705629531173904
+      "required_token_positions_per_second": 9540.452255299368,
+      "required_pages_per_second_upper_estimate": 0.0038058290471116034
     },
     {
       "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": 18904.09442174935,
-      "required_pages_per_second_upper_estimate": 0.004373129543773029
+      "required_token_positions_per_second": 19080.904510598735,
+      "required_pages_per_second_upper_estimate": 0.00441403144607705
     },
     {
       "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": 28356.141632624025,
-      "required_pages_per_second_upper_estimate": 0.0017267402522494538
+      "required_token_positions_per_second": 28621.3567658981,
+      "required_pages_per_second_upper_estimate": 0.0017428904623895795
     },
     {
       "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": 28356.141632624025,
-      "required_pages_per_second_upper_estimate": 0.0021829342777033287
+      "required_token_positions_per_second": 28621.3567658981,
+      "required_pages_per_second_upper_estimate": 0.0022033512728252447
     },
     {
       "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": 28356.141632624025,
-      "required_pages_per_second_upper_estimate": 0.3520883771760064
+      "required_token_positions_per_second": 28621.3567658981,
+      "required_pages_per_second_upper_estimate": 0.3553814615133181
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11199002101564304,
-    "producer_p95_seconds": 0.20070495999971172,
-    "gpu_update_seconds": 0.9750268692476377,
-    "producer_batches_per_second": 8.929367017980267,
-    "gpu_batches_per_second": 1.0256127615966444,
-    "producer_headroom": 8.706372767904416,
+    "producer_batch_seconds": 0.07985346209379429,
+    "producer_p95_seconds": 0.19511583799976506,
+    "gpu_update_seconds": 0.965991941826537,
+    "producer_batches_per_second": 12.52293856496065,
+    "gpu_batches_per_second": 1.035205322840643,
+    "producer_headroom": 12.097057741740766,
     "sustainable": true,
-    "buffer_seconds": 60.451665893353535,
+    "buffer_seconds": 60.85749233507183,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297138.42189612245,
-    "upload_bytes_per_second": 5486306.114488939,
+    "generation_bytes_per_second": 334590.61790641217,
+    "upload_bytes_per_second": 5496123.6982138995,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.75898831045871,
+    "estimated_full_shards_per_hour": 71.79535773176454,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 1049873,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.218386226999428,
+      "bytes": 7220161,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -55,3 +55,4 @@
 {"step": 285700, "loss": 0.608401644974947, "lr": 0.0003, "grad_norm": 0.1501933163471568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41818387961, "tokens_per_second": 148086.78253792008, "data_wait_s": 1.215599877468776e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285800, "loss": 0.553261948749423, "lr": 0.0003, "grad_norm": 0.14661093419425655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41832941344, "tokens_per_second": 148412.1016033091, "data_wait_s": 9.921999662765302e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 285900, "loss": 0.5836225003004074, "lr": 0.0003, "grad_norm": 0.1586045129972069, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41847504610, "tokens_per_second": 149271.3635139179, "data_wait_s": 1.3619999663205817e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 286000, "loss": 0.6062596887350082, "lr": 0.0003, "grad_norm": 0.1471150243766587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41862184336, "tokens_per_second": 139687.25513113348, "data_wait_s": 1.1812999218818732e-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:cd7b5ab6bd1a360b0c54d633730a61031b5957f8675cb085a77240a68ad45180
-size 121674
+oid sha256:662573738c3896bd44f6b43c988dbf38b92e740cd7385ff05e9119a4c95aab74
+size 123792