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": 76857.83044944204,
-      "required_pages_per_second_upper_estimate": 0.005317203045228775
+      "required_token_positions_per_second": 77089.37780631066,
+      "required_pages_per_second_upper_estimate": 0.005333222028640834
     },
     {
       "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": 9607.228806180256,
-      "required_pages_per_second_upper_estimate": 0.0040223460793967405
+      "required_token_positions_per_second": 9636.172225788832,
+      "required_pages_per_second_upper_estimate": 0.00403446408477952
     },
     {
       "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": 19214.45761236051,
-      "required_pages_per_second_upper_estimate": 0.004001864801332058
+      "required_token_positions_per_second": 19272.344451577665,
+      "required_pages_per_second_upper_estimate": 0.004013921103362383
     },
     {
       "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": 28821.686418540765,
-      "required_pages_per_second_upper_estimate": 0.0017278160926245132
+      "required_token_positions_per_second": 28908.516677366497,
+      "required_pages_per_second_upper_estimate": 0.0017330214340589875
     },
     {
       "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": 28821.686418540765,
-      "required_pages_per_second_upper_estimate": 0.002240759345840794
+      "required_token_positions_per_second": 28908.516677366497,
+      "required_pages_per_second_upper_estimate": 0.0022475100165385495
     },
     {
       "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": 28821.686418540765,
-      "required_pages_per_second_upper_estimate": 0.36969364706123276
+      "required_token_positions_per_second": 28908.516677366497,
+      "required_pages_per_second_upper_estimate": 0.37080741239038106
     }
   ],
   "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": 0.11394102725770949,
-    "producer_p95_seconds": 0.1822635349999473,
-    "gpu_update_seconds": 0.9592776633019732,
-    "producer_batches_per_second": 8.77646993420746,
-    "gpu_batches_per_second": 1.0424510423372673,
-    "producer_headroom": 8.419071570526555,
+    "producer_batch_seconds": 0.05339074917958442,
+    "producer_p95_seconds": 0.18270901699906972,
+    "gpu_update_seconds": 0.956396355737152,
+    "producer_batches_per_second": 18.72983644856552,
+    "gpu_batches_per_second": 1.0455916043607674,
+    "producer_headroom": 17.913147322960945,
     "sustainable": true,
-    "buffer_seconds": 60.43449278802431,
+    "buffer_seconds": 60.25297041144057,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347384.5460282837,
-    "upload_bytes_per_second": 3141706.6507950164,
+    "generation_bytes_per_second": 347480.7131843553,
+    "upload_bytes_per_second": 3173805.4771586163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.54063687931426,
+    "estimated_full_shards_per_hour": 74.56127211235041,
     "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": [
@@ -514,6 +514,12 @@
       "bytes": 1093074,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.176304970000274,
+      "bytes": 7496411,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 696000, "loss": 0.5266731977462769, "lr": 8e-05, "grad_norm": 0.23273954291675364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101875729910, "tokens_per_second": 151681.54074112067, "data_wait_s": 1.4149998605716974e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696100, "loss": 0.4949779808521271, "lr": 8e-05, "grad_norm": 0.2101428819671386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101890331384, "tokens_per_second": 151445.37746768864, "data_wait_s": 1.6972000594250858e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696200, "loss": 0.6068347096443176, "lr": 8e-05, "grad_norm": 0.22772669219068942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101904911710, "tokens_per_second": 151933.08169458772, "data_wait_s": 1.4774999726796523e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 696300, "loss": 0.5331386923789978, "lr": 8e-05, "grad_norm": 0.20653918218985734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101919565504, "tokens_per_second": 134281.24536770766, "data_wait_s": 1.8845999875338748e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91442296ffe86b5e5e0fca6af09c0d291f0c156796ddf02aab61e5e835cb98ae
-size 238947
+oid sha256:3ec95d126ba48bb830babfd596785d8252615c9c571f4ca4776200c8a5cd2507
+size 241065