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": 74747.38446791789,
-      "required_pages_per_second_upper_estimate": 0.006478913840165183
+      "required_token_positions_per_second": 74794.12533314222,
+      "required_pages_per_second_upper_estimate": 0.0064829652198991885
     },
     {
       "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": 9343.423058489736,
-      "required_pages_per_second_upper_estimate": 0.0037272311546552323
+      "required_token_positions_per_second": 9349.265666642777,
+      "required_pages_per_second_upper_estimate": 0.003729561858402257
     },
     {
       "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": 18686.846116979472,
-      "required_pages_per_second_upper_estimate": 0.004322872971904095
+      "required_token_positions_per_second": 18698.531333285555,
+      "required_pages_per_second_upper_estimate": 0.004325576141043725
     },
     {
       "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": 28030.269175469206,
-      "required_pages_per_second_upper_estimate": 0.0017068963293293737
+      "required_token_positions_per_second": 28047.79699992833,
+      "required_pages_per_second_upper_estimate": 0.0017079636818775467
     },
     {
       "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": 28030.269175469206,
-      "required_pages_per_second_upper_estimate": 0.002157847713878854
+      "required_token_positions_per_second": 28047.79699992833,
+      "required_pages_per_second_upper_estimate": 0.002159197054325841
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28030.269175469206,
-      "required_pages_per_second_upper_estimate": 0.3480421318830998
+      "required_token_positions_per_second": 28047.79699992833,
+      "required_pages_per_second_upper_estimate": 0.3482597688010272
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09926299432813579,
-    "producer_p95_seconds": 0.1950109989993507,
-    "gpu_update_seconds": 0.9863622724035861,
-    "producer_batches_per_second": 10.07424777751796,
-    "gpu_batches_per_second": 1.0138262867284868,
-    "producer_headroom": 9.936857930589394,
+    "producer_batch_seconds": 0.053408991187517074,
+    "producer_p95_seconds": 0.18466416200044478,
+    "gpu_update_seconds": 0.9857458680291592,
+    "producer_batches_per_second": 18.723439214364404,
+    "gpu_batches_per_second": 1.0144602502867597,
+    "producer_headroom": 18.456552840854837,
     "sustainable": true,
-    "buffer_seconds": 60.16809861661875,
+    "buffer_seconds": 60.130497949778714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444182.0761238619,
-    "upload_bytes_per_second": 5967616.168923454,
+    "generation_bytes_per_second": 443974.4644154879,
+    "upload_bytes_per_second": 5883279.233415565,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.3111335066499,
+    "estimated_full_shards_per_hour": 95.26658486698607,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 12430977,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.9270829400002185,
+      "bytes": 1118634,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 87500, "loss": 0.6118768639862537, "lr": 0.0003, "grad_norm": 0.13208805657286993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12804833533, "tokens_per_second": 147901.13554139592, "data_wait_s": 9.451000551052857e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 87600, "loss": 0.6148227006196976, "lr": 0.0003, "grad_norm": 0.12887817111968602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12819477179, "tokens_per_second": 149061.14271705193, "data_wait_s": 1.0055000529973768e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 87700, "loss": 0.5609135627746582, "lr": 0.0003, "grad_norm": 0.13802842621014572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12834214746, "tokens_per_second": 149762.05390319825, "data_wait_s": 1.2090000382158905e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 87800, "loss": 0.6796466615051031, "lr": 0.0003, "grad_norm": 0.13775670145550264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12848973134, "tokens_per_second": 149532.93794615607, "data_wait_s": 9.22599974728655e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1a4a79bfd6622f0417dc9d7bfd08855968b182eab4c195c8d87f65fadef80bb
-size 123792
+oid sha256:75bd771748361e812d4e31431a75361a96621ccc99a21a863fcc601b450b6740
+size 125910