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": 75204.01420684911,
-      "required_pages_per_second_upper_estimate": 0.00651849334861825
+      "required_token_positions_per_second": 76020.7798683107,
+      "required_pages_per_second_upper_estimate": 0.006589288526080097
     },
     {
       "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": 9400.501775856139,
-      "required_pages_per_second_upper_estimate": 0.0037500007084155652
+      "required_token_positions_per_second": 9502.597483538837,
+      "required_pages_per_second_upper_estimate": 0.003790728212677053
     },
     {
       "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": 18801.003551712278,
-      "required_pages_per_second_upper_estimate": 0.004349281285327298
+      "required_token_positions_per_second": 19005.194967077674,
+      "required_pages_per_second_upper_estimate": 0.00439651737562591
     },
     {
       "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": 28201.50532756842,
-      "required_pages_per_second_upper_estimate": 0.0017173237125855583
+      "required_token_positions_per_second": 28507.79245061651,
+      "required_pages_per_second_upper_estimate": 0.0017359749914148451
     },
     {
       "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": 28201.50532756842,
-      "required_pages_per_second_upper_estimate": 0.0021710299468794586
+      "required_token_positions_per_second": 28507.79245061651,
+      "required_pages_per_second_upper_estimate": 0.002194608777468723
     },
     {
       "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": 28201.50532756842,
-      "required_pages_per_second_upper_estimate": 0.35016831180163677
+      "required_token_positions_per_second": 28507.79245061651,
+      "required_pages_per_second_upper_estimate": 0.35397137279283447
     }
   ],
   "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": 0.0970556911094036,
-    "producer_p95_seconds": 0.39262932300334796,
-    "gpu_update_seconds": 0.9803731991913447,
-    "producer_batches_per_second": 10.303362827768389,
-    "gpu_batches_per_second": 1.0200197239427233,
-    "producer_headroom": 10.101140777888476,
+    "producer_batch_seconds": 0.11253044140596558,
+    "producer_p95_seconds": 0.3794181720004417,
+    "gpu_update_seconds": 0.9698400901400587,
+    "producer_batches_per_second": 8.886484292658137,
+    "gpu_batches_per_second": 1.0310978172242662,
+    "producer_headroom": 8.618468727419783,
     "sustainable": true,
-    "buffer_seconds": 60.78313834986337,
+    "buffer_seconds": 60.130085588683635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429183.57444168127,
-    "upload_bytes_per_second": 5656445.629007714,
+    "generation_bytes_per_second": 447992.8977198156,
+    "upload_bytes_per_second": 5608547.544504554,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.09280419290378,
+    "estimated_full_shards_per_hour": 96.12884710975506,
     "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": 1.1233831270001247,
-      "bytes": 6290212,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8791882089972205,
       "bytes": 6587099,
@@ -562,6 +556,12 @@
       "bytes": 11759986,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 0.806070794002153,
+      "bytes": 884841,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -350,3 +350,4 @@
 {"step": 75000, "loss": 0.727703420445323, "lr": 0.0003, "grad_norm": 0.17311887083600086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10975731388, "tokens_per_second": 147207.9602730767, "data_wait_s": 1.3761993614025414e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 75100, "loss": 0.6338124442845583, "lr": 0.0003, "grad_norm": 0.139546121747869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10990242509, "tokens_per_second": 138263.8792476853, "data_wait_s": 7.650000043213367e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 75200, "loss": 0.6243664398789406, "lr": 0.0003, "grad_norm": 0.1403619385012886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11004993501, "tokens_per_second": 150818.62802642008, "data_wait_s": 8.654002158436924e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 75300, "loss": 0.6142660826444626, "lr": 0.0003, "grad_norm": 0.13490544914266286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11019686667, "tokens_per_second": 149725.26537174347, "data_wait_s": 1.0880001354962587e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68bff7de791c1d69df9adb7c30da780f4a7da18a37a95d5382895e08936c66a6
-size 751182
+oid sha256:9f31fb94420061ed9db8d04dbebfb2bf99127b5f3207485cf8a02c3fe1a6df2f
+size 753300