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": 75071.13350699151,
-      "required_pages_per_second_upper_estimate": 0.006506975586337647
+      "required_token_positions_per_second": 75598.2496106376,
+      "required_pages_per_second_upper_estimate": 0.00655266467423814
     },
     {
       "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": 9383.891688373938,
-      "required_pages_per_second_upper_estimate": 0.0037433746961759765
+      "required_token_positions_per_second": 9449.7812013297,
+      "required_pages_per_second_upper_estimate": 0.0037696590080300387
     },
     {
       "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": 18767.783376747877,
-      "required_pages_per_second_upper_estimate": 0.004341596382504395
+      "required_token_positions_per_second": 18899.5624026594,
+      "required_pages_per_second_upper_estimate": 0.004372081140917381
     },
     {
       "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": 28151.675065121817,
-      "required_pages_per_second_upper_estimate": 0.0017142893110417422
+      "required_token_positions_per_second": 28349.343603989102,
+      "required_pages_per_second_upper_estimate": 0.0017263262879720867
     },
     {
       "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": 28151.675065121817,
-      "required_pages_per_second_upper_estimate": 0.0021671938753373243
+      "required_token_positions_per_second": 28349.343603989102,
+      "required_pages_per_second_upper_estimate": 0.0021824109466408647
     },
     {
       "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": 28151.675065121817,
-      "required_pages_per_second_upper_estimate": 0.3495495867132103
+      "required_token_positions_per_second": 28349.343603989102,
+      "required_pages_per_second_upper_estimate": 0.3520039684118989
     }
   ],
   "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": 0.13525272735159888,
-    "producer_p95_seconds": 0.3286041680003109,
-    "gpu_update_seconds": 0.982108522354116,
-    "producer_batches_per_second": 7.393566248763549,
-    "gpu_batches_per_second": 1.0182174141030749,
-    "producer_headroom": 7.261284423500434,
+    "producer_batch_seconds": 0.16837734289049422,
+    "producer_p95_seconds": 0.3110309090006922,
+    "gpu_update_seconds": 0.9752606757395816,
+    "producer_batches_per_second": 5.939041339132898,
+    "gpu_batches_per_second": 1.0253668838248373,
+    "producer_headroom": 5.7921134696480605,
     "sustainable": true,
-    "buffer_seconds": 60.89072838595519,
+    "buffer_seconds": 60.466161895854064,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 255991.2591051141,
-    "upload_bytes_per_second": 2255153.384844526,
+    "generation_bytes_per_second": 265505.33179643203,
+    "upload_bytes_per_second": 2278907.642506723,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.9297650324351,
+    "estimated_full_shards_per_hour": 56.971263555714806,
     "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": 1.586299536000297,
-      "bytes": 6350565,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 0.9621634939976502,
       "bytes": 2554887,
@@ -562,6 +556,12 @@
       "bytes": 7460093,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.418776655002148,
+      "bytes": 10393041,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -176,3 +176,4 @@
 {"step": 297800, "loss": 0.5744358487427235, "lr": 0.0003, "grad_norm": 0.15290834698088446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43590185194, "tokens_per_second": 150211.0022263145, "data_wait_s": 9.883999155135825e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 297900, "loss": 0.6208792291581631, "lr": 0.0003, "grad_norm": 0.14864424516271077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43604950684, "tokens_per_second": 150923.5094337729, "data_wait_s": 9.46100044529885e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 298000, "loss": 0.572965245693922, "lr": 0.0003, "grad_norm": 0.14221758515788127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43619684108, "tokens_per_second": 149950.25379936505, "data_wait_s": 1.3213000784162432e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 298100, "loss": 0.5779675506055355, "lr": 0.0003, "grad_norm": 0.15523465119748014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43634281008, "tokens_per_second": 148972.73237010153, "data_wait_s": 1.3013999705435708e-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 sha256:9e8e41e22382aad6755f989845606e669e1cb71e90f1d8de55c42162ee328dec
-size 379518
+oid sha256:f370c001867f5b35710ea71307143d72a485b27a13706611ff8496809f3fcdc4
+size 381636