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": 75260.5304577392,
-      "required_pages_per_second_upper_estimate": 0.006523392034006274
+      "required_token_positions_per_second": 76066.20672241565,
+      "required_pages_per_second_upper_estimate": 0.006593226010660604
     },
     {
       "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": 9407.5663072174,
-      "required_pages_per_second_upper_estimate": 0.0037528188555997284
+      "required_token_positions_per_second": 9508.275840301956,
+      "required_pages_per_second_upper_estimate": 0.0037929933940888606
     },
     {
       "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": 18815.1326144348,
-      "required_pages_per_second_upper_estimate": 0.004352549795325142
+      "required_token_positions_per_second": 19016.551680603912,
+      "required_pages_per_second_upper_estimate": 0.004399144551428874
     },
     {
       "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": 28222.698921652198,
-      "required_pages_per_second_upper_estimate": 0.001718614291271066
+      "required_token_positions_per_second": 28524.827520905867,
+      "required_pages_per_second_upper_estimate": 0.0017370123378193604
     },
     {
       "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": 28222.698921652198,
-      "required_pages_per_second_upper_estimate": 0.0021726614884196513
+      "required_token_positions_per_second": 28524.827520905867,
+      "required_pages_per_second_upper_estimate": 0.00219592018433569
     },
     {
       "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": 28222.698921652198,
-      "required_pages_per_second_upper_estimate": 0.3504314653097607
+      "required_token_positions_per_second": 28524.827520905867,
+      "required_pages_per_second_upper_estimate": 0.35418289135311554
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.4736153821485374,
-    "producer_p95_seconds": 0.20226775399896724,
-    "gpu_update_seconds": 0.9796369963323637,
-    "producer_batches_per_second": 2.1114179093245236,
-    "gpu_batches_per_second": 1.0207862746546659,
-    "producer_headroom": 2.068423098693035,
+    "producer_batch_seconds": 0.07829836524996381,
+    "producer_p95_seconds": 0.19687288199929753,
+    "gpu_update_seconds": 0.9692609001662416,
+    "producer_batches_per_second": 12.771658728857844,
+    "gpu_batches_per_second": 1.0317139583660975,
+    "producer_headroom": 12.379069436148791,
     "sustainable": true,
-    "buffer_seconds": 60.73749377260655,
+    "buffer_seconds": 60.09417581030698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311516.6844266971,
-    "upload_bytes_per_second": 5511704.383295727,
+    "generation_bytes_per_second": 299521.5781140328,
+    "upload_bytes_per_second": 5569064.774391282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.84422874069867,
+    "estimated_full_shards_per_hour": 64.27035815778483,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 2374553,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2559754029989563,
+      "bytes": 9926404,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -52,3 +52,4 @@
 {"step": 285400, "loss": 0.5829807613044977, "lr": 0.0003, "grad_norm": 0.1464344673972657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41774615383, "tokens_per_second": 150879.59424880205, "data_wait_s": 1.4166000255499966e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285500, "loss": 0.6272663809359074, "lr": 0.0003, "grad_norm": 0.14719175495472536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41789375829, "tokens_per_second": 151690.0631004838, "data_wait_s": 1.4142999134492129e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285600, "loss": 0.6442608311772346, "lr": 0.0003, "grad_norm": 0.16393942543920254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41803927346, "tokens_per_second": 131592.77003216252, "data_wait_s": 6.4779997046571225e-06, "gpu_peak_gb": 85.46991729736328}
+{"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}

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 sha256:2740a077e1243a9d19f69e5734e74baa7a783b94b7f4d90e65d1c2ab5c923083
-size 115320
+oid sha256:e1f4474e7f4123175d4a2a297632200519d41586607137df66028f3b722f4a27
+size 117438