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": 75270.03680702043,
-      "required_pages_per_second_upper_estimate": 0.006524216020268349
+      "required_token_positions_per_second": 74551.87473408281,
+      "required_pages_per_second_upper_estimate": 0.006461967551951249
     },
     {
       "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": 9408.754600877553,
-      "required_pages_per_second_upper_estimate": 0.0037532928837073374
+      "required_token_positions_per_second": 9318.984341760352,
+      "required_pages_per_second_upper_estimate": 0.003717482185160504
     },
     {
       "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": 18817.509201755107,
-      "required_pages_per_second_upper_estimate": 0.004353099576975187
+      "required_token_positions_per_second": 18637.968683520703,
+      "required_pages_per_second_upper_estimate": 0.004311566037886858
     },
     {
       "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": 28226.26380263266,
-      "required_pages_per_second_upper_estimate": 0.001718831373819291
+      "required_token_positions_per_second": 27956.95302528106,
+      "required_pages_per_second_upper_estimate": 0.0017024317604430275
     },
     {
       "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": 28226.26380263266,
-      "required_pages_per_second_upper_estimate": 0.0021729359228257494
+      "required_token_positions_per_second": 27956.95302528106,
+      "required_pages_per_second_upper_estimate": 0.0021522036336852775
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28226.26380263266,
-      "required_pages_per_second_upper_estimate": 0.3504757292006489
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27956.95302528106,
+      "required_pages_per_second_upper_estimate": 0.34713179067113326
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1365862909766804,
-    "producer_p95_seconds": 0.29147882200049935,
-    "gpu_update_seconds": 0.9795132715163413,
-    "producer_batches_per_second": 7.321378982102469,
-    "gpu_batches_per_second": 1.0209152127688317,
-    "producer_headroom": 7.17138787877017,
+    "producer_batch_seconds": 0.7983091551562893,
+    "producer_p95_seconds": 0.3582648830015387,
+    "gpu_update_seconds": 0.9889489736237824,
+    "producer_batches_per_second": 1.2526475407941733,
+    "gpu_batches_per_second": 1.0111745162500383,
+    "producer_headroom": 1.2388044997807528,
     "sustainable": true,
-    "buffer_seconds": 60.729822834013156,
+    "buffer_seconds": 60.32588739105073,
     "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": 285295.552513365,
-    "upload_bytes_per_second": 4398674.9641365735,
+    "generation_bytes_per_second": 263295.85887776804,
+    "upload_bytes_per_second": 4407349.330201067,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.21778422880852,
+    "estimated_full_shards_per_hour": 56.49716210126668,
     "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": 1.0509260079998057,
-      "bytes": 6234779,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.154916157000116,
       "bytes": 2367312,
@@ -562,6 +556,12 @@
       "bytes": 7512493,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9002696629977436,
+      "bytes": 6383764,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -291,3 +291,4 @@
 {"step": 266200, "loss": 0.6267967186868191, "lr": 0.0003, "grad_norm": 0.14775866779638372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38964363490, "tokens_per_second": 148285.72594981416, "data_wait_s": 9.677001798991114e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 266300, "loss": 0.6215458959341049, "lr": 0.0003, "grad_norm": 0.16557022819382566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38979085931, "tokens_per_second": 150119.21623880844, "data_wait_s": 1.4631001249654219e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 266400, "loss": 0.6018778197467327, "lr": 0.0003, "grad_norm": 0.15254020031180637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38993517013, "tokens_per_second": 147141.5573219985, "data_wait_s": 9.244002285413444e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 266500, "loss": 0.5274950340390205, "lr": 0.0003, "grad_norm": 0.1670249046064422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39008052295, "tokens_per_second": 116731.62916602202, "data_wait_s": 1.4427998394239694e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8674f4067a020ee42c03a3098bde953b7257e1a5db9be3d42d8f979aa37461d
-size 625437
+oid sha256:cbd8becc5cf4dfc7dd8421f4f45365e0bfc47ca043051b6bae012b339f70945d
+size 627555