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": 75223.55866632314,
-      "required_pages_per_second_upper_estimate": 0.006520187412830478
+      "required_token_positions_per_second": 75071.13350699151,
+      "required_pages_per_second_upper_estimate": 0.006506975586337647
     },
     {
       "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": 9402.944833290392,
-      "required_pages_per_second_upper_estimate": 0.0037509752805530527
+      "required_token_positions_per_second": 9383.891688373938,
+      "required_pages_per_second_upper_estimate": 0.0037433746961759765
     },
     {
       "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": 18805.889666580784,
-      "required_pages_per_second_upper_estimate": 0.0043504116020094425
+      "required_token_positions_per_second": 18767.783376747877,
+      "required_pages_per_second_upper_estimate": 0.004341596382504395
     },
     {
       "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": 28208.834499871176,
-      "required_pages_per_second_upper_estimate": 0.001717770020725601
+      "required_token_positions_per_second": 28151.675065121817,
+      "required_pages_per_second_upper_estimate": 0.0017142893110417422
     },
     {
       "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": 28208.834499871176,
-      "required_pages_per_second_upper_estimate": 0.0021715941668517726
+      "required_token_positions_per_second": 28151.675065121817,
+      "required_pages_per_second_upper_estimate": 0.0021671938753373243
     },
     {
       "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": 28208.834499871176,
-      "required_pages_per_second_upper_estimate": 0.3502593155924752
+      "required_token_positions_per_second": 28151.675065121817,
+      "required_pages_per_second_upper_estimate": 0.3495495867132103
     }
   ],
   "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": 0.11000797910969595,
-    "producer_p95_seconds": 0.3336559440031124,
-    "gpu_update_seconds": 0.9801184802628503,
-    "producer_batches_per_second": 9.090249708185588,
-    "gpu_batches_per_second": 1.0202848126400166,
-    "producer_headroom": 8.909521729196678,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.76734577629672,
+    "buffer_seconds": 60.89072838595519,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 250173.04071104777,
-    "upload_bytes_per_second": 2218716.0986200282,
+    "generation_bytes_per_second": 255991.2591051141,
+    "upload_bytes_per_second": 2255153.384844526,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.681310806260825,
+    "estimated_full_shards_per_hour": 54.9297650324351,
     "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": 1.4697227199976624,
-      "bytes": 921486,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.586299536000297,
       "bytes": 6350565,
@@ -562,6 +556,12 @@
       "bytes": 5875069,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3580094259996258,
+      "bytes": 7460093,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -175,3 +175,4 @@
 {"step": 297700, "loss": 0.5460868366062641, "lr": 0.0003, "grad_norm": 0.1479280371962989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43575536298, "tokens_per_second": 148930.1567997591, "data_wait_s": 8.151000656653196e-06, "gpu_peak_gb": 85.46991729736328}
 {"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}

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 sha256:89fa77474499ab6b9dfb7585db261d8bbd09ff3c946fc0cddabd4a0248f5c24e
-size 377400
+oid sha256:9e8e41e22382aad6755f989845606e669e1cb71e90f1d8de55c42162ee328dec
+size 379518