Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77006.31555546631,
-      "required_pages_per_second_upper_estimate": 0.005327475589396447
+      "required_token_positions_per_second": 76623.24908125607,
+      "required_pages_per_second_upper_estimate": 0.005300974161873905
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9625.789444433289,
-      "required_pages_per_second_upper_estimate": 0.0040301170310430845
+      "required_token_positions_per_second": 9577.906135157009,
+      "required_pages_per_second_upper_estimate": 0.004010069289366308
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19251.578888866577,
-      "required_pages_per_second_upper_estimate": 0.0040095961842743786
+      "required_token_positions_per_second": 19155.812270314018,
+      "required_pages_per_second_upper_estimate": 0.003989650523165447
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28877.368333299866,
-      "required_pages_per_second_upper_estimate": 0.0017311541383929604
+      "required_token_positions_per_second": 28733.718405471023,
+      "required_pages_per_second_upper_estimate": 0.0017225425445603609
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28877.368333299866,
-      "required_pages_per_second_upper_estimate": 0.002245088369794455
+      "required_token_positions_per_second": 28733.718405471023,
+      "required_pages_per_second_upper_estimate": 0.0022339202197550192
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28877.368333299866,
-      "required_pages_per_second_upper_estimate": 0.37040787487718047
+      "required_token_positions_per_second": 28733.718405471023,
+      "required_pages_per_second_upper_estimate": 0.3685652878422676
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16859342357028595,
-    "producer_p95_seconds": 0.3012484790015151,
-    "gpu_update_seconds": 0.9574279650724882,
-    "producer_batches_per_second": 5.931429463991541,
-    "gpu_batches_per_second": 1.0444650004810425,
-    "producer_headroom": 5.67891644168042,
+    "producer_batch_seconds": 0.11021912598431527,
+    "producer_p95_seconds": 0.3112322960005258,
+    "gpu_update_seconds": 0.9622144829934611,
+    "producer_batches_per_second": 9.072835508986934,
+    "gpu_batches_per_second": 1.039269328901585,
+    "producer_headroom": 8.730013728564579,
     "sustainable": true,
-    "buffer_seconds": 60.31796179956675,
+    "buffer_seconds": 60.61951242858805,
     "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": 437078.4339456666,
-    "upload_bytes_per_second": 4868525.538710905,
+    "generation_bytes_per_second": 436256.6226324876,
+    "upload_bytes_per_second": 4813585.146918953,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.78685725953578,
+    "estimated_full_shards_per_hour": 93.61051568251582,
     "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": 2.2123866580004687,
-      "bytes": 13632293,
-      "pages": 49,
-      "shards": 4
-    },
     {
       "seconds": 1.2271638120000716,
       "bytes": 936260,
@@ -562,6 +556,12 @@
       "bytes": 7496140,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.6659519919994636,
+      "bytes": 6121395,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 792200, "loss": 0.5342683792114258, "lr": 3e-05, "grad_norm": 0.2635454912110246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115955309528, "tokens_per_second": 151981.81773728828, "data_wait_s": 9.202998626278713e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 792300, "loss": 0.5348979234695435, "lr": 3e-05, "grad_norm": 0.2578812605807442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115969991276, "tokens_per_second": 152120.4504045221, "data_wait_s": 1.0077998013002798e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792400, "loss": 0.5232592225074768, "lr": 3e-05, "grad_norm": 0.25947454606977255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115984565712, "tokens_per_second": 151426.0776815659, "data_wait_s": 1.357600194751285e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 792500, "loss": 0.5563556551933289, "lr": 3e-05, "grad_norm": 0.26630299494470827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115999186454, "tokens_per_second": 152317.218110856, "data_wait_s": 1.4304998330771923e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25311eefa60c79fcb41456d63199c38d609dd1172bfc590d073750b1a477d261
-size 691095
+oid sha256:10f781c3e68064ba30b0b1bd07898e0bfb9e3cdc11faa644e1e41e3b9706426a
+size 693213