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": 77400.91565948205,
-      "required_pages_per_second_upper_estimate": 0.005354774940190642
+      "required_token_positions_per_second": 76509.59968554687,
+      "required_pages_per_second_upper_estimate": 0.005293111632975813
     },
     {
       "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": 9675.114457435257,
-      "required_pages_per_second_upper_estimate": 0.004050768383963609
+      "required_token_positions_per_second": 9563.699960693359,
+      "required_pages_per_second_upper_estimate": 0.004004121460777034
     },
     {
       "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": 19350.228914870513,
-      "required_pages_per_second_upper_estimate": 0.004030142383114864
+      "required_token_positions_per_second": 19127.399921386717,
+      "required_pages_per_second_upper_estimate": 0.0039837329801679176
     },
     {
       "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": 29025.343372305768,
-      "required_pages_per_second_upper_estimate": 0.0017400250160365623
+      "required_token_positions_per_second": 28691.099882080078,
+      "required_pages_per_second_upper_estimate": 0.001719987629157791
     },
     {
       "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": 29025.343372305768,
-      "required_pages_per_second_upper_estimate": 0.0022565927782037536
+      "required_token_positions_per_second": 28691.099882080078,
+      "required_pages_per_second_upper_estimate": 0.002230606817020446
     },
     {
       "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": 29025.343372305768,
-      "required_pages_per_second_upper_estimate": 0.3723059397943301
+      "required_token_positions_per_second": 28691.099882080078,
+      "required_pages_per_second_upper_estimate": 0.3680186231844137
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40104013639847125,
-    "producer_p95_seconds": 0.17060623599991231,
-    "gpu_update_seconds": 0.9525468706902552,
-    "producer_batches_per_second": 2.4935160080995122,
-    "gpu_batches_per_second": 1.0498171069265687,
-    "producer_headroom": 2.375190870531248,
+    "producer_batch_seconds": 0.09504132128127196,
+    "producer_p95_seconds": 0.16387522999957582,
+    "gpu_update_seconds": 0.9636437819962567,
+    "producer_batches_per_second": 10.52173924477049,
+    "gpu_batches_per_second": 1.0377278603182898,
+    "producer_headroom": 10.139208599009073,
     "sustainable": true,
-    "buffer_seconds": 60.01045285348608,
+    "buffer_seconds": 60.70955826576417,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304554.9955122595,
+    "generation_bytes_per_second": 279098.6073348513,
     "upload_bytes_per_second": 5870910.389131389,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.35041235948408,
+    "estimated_full_shards_per_hour": 59.88806405100016,
     "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": [