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": 75501.62196555822,
-      "required_pages_per_second_upper_estimate": 0.006544289234863187
+      "required_token_positions_per_second": 75706.5335479691,
+      "required_pages_per_second_upper_estimate": 0.006562050451482904
     },
     {
       "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": 9437.702745694778,
-      "required_pages_per_second_upper_estimate": 0.003764840731488263
+      "required_token_positions_per_second": 9463.316693496137,
+      "required_pages_per_second_upper_estimate": 0.0037750585182288725
     },
     {
       "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": 18875.405491389556,
-      "required_pages_per_second_upper_estimate": 0.00436649286464222
+      "required_token_positions_per_second": 18926.633386992275,
+      "required_pages_per_second_upper_estimate": 0.00437834353670974
     },
     {
       "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": 28313.108237084336,
-      "required_pages_per_second_upper_estimate": 0.001724119744239864
+      "required_token_positions_per_second": 28389.950080488412,
+      "required_pages_per_second_upper_estimate": 0.0017287990093451775
     },
     {
       "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": 28313.108237084336,
-      "required_pages_per_second_upper_estimate": 0.0021796214477906206
+      "required_token_positions_per_second": 28389.950080488412,
+      "required_pages_per_second_upper_estimate": 0.002185536945607703
     },
     {
       "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": 28313.108237084336,
-      "required_pages_per_second_upper_estimate": 0.35155404642691357
+      "required_token_positions_per_second": 28389.950080488412,
+      "required_pages_per_second_upper_estimate": 0.35250816494888576
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10809764025776758,
-    "producer_p95_seconds": 0.3252332199990633,
-    "gpu_update_seconds": 0.9765088229976396,
-    "producer_batches_per_second": 9.25089574217734,
-    "gpu_batches_per_second": 1.024056287510284,
-    "producer_headroom": 9.03358131286747,
+    "producer_batch_seconds": 0.09329093939049926,
+    "producer_p95_seconds": 0.2874859760013351,
+    "gpu_update_seconds": 0.9738657490279163,
+    "producer_batches_per_second": 10.719154577425552,
+    "gpu_batches_per_second": 1.026835578721369,
+    "producer_headroom": 10.439017501490554,
     "sustainable": true,
-    "buffer_seconds": 60.543547025853655,
+    "buffer_seconds": 60.379676439730815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402222.2851045854,
-    "upload_bytes_per_second": 5564590.260246406,
+    "generation_bytes_per_second": 401955.0661163763,
+    "upload_bytes_per_second": 5562817.457726893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.30753912785694,
+    "estimated_full_shards_per_hour": 86.25020015352695,
     "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": 0.9169029310014594,
-      "bytes": 7362656,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9597948790014925,
       "bytes": 2508350,
@@ -562,6 +556,12 @@
       "bytes": 6113242,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 0.979562092001288,
+      "bytes": 7577775,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 21200, "loss": 0.6821452751755714, "lr": 0.0003, "grad_norm": 0.14234147294255003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3102061033, "tokens_per_second": 149341.20047718784, "data_wait_s": 1.0720003047026694e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 21300, "loss": 0.7059006355702877, "lr": 0.0003, "grad_norm": 0.13807118533059076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3116664584, "tokens_per_second": 148218.4060777925, "data_wait_s": 1.1599000572459772e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 21400, "loss": 0.6534912511706352, "lr": 0.0003, "grad_norm": 0.13234207093614064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3131302397, "tokens_per_second": 148239.98097128168, "data_wait_s": 1.206399974762462e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 21500, "loss": 0.6984033659100533, "lr": 0.0003, "grad_norm": 0.13349629381036066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3146023552, "tokens_per_second": 149947.73587327, "data_wait_s": 1.4861998351989314e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e27cf811d08ffbaaf04555428da1e88c4eb00162ba2d75e23001618b87fa33d
-size 450775
+oid sha256:28810d58795c1bad2fc3e6f3f565ad581803f4826677de7cbf66a2e38f1fc41d
+size 452893