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": 75814.08589160604,
-      "required_pages_per_second_upper_estimate": 0.006571372816040431
+      "required_token_positions_per_second": 75434.74103291096,
+      "required_pages_per_second_upper_estimate": 0.006538492165129496
     },
     {
       "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": 9476.760736450755,
-      "required_pages_per_second_upper_estimate": 0.003780421547969824
+      "required_token_positions_per_second": 9429.34262911387,
+      "required_pages_per_second_upper_estimate": 0.003761505755989257
     },
     {
       "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": 18953.52147290151,
-      "required_pages_per_second_upper_estimate": 0.004384563622170691
+      "required_token_positions_per_second": 18858.68525822774,
+      "required_pages_per_second_upper_estimate": 0.004362624932966287
     },
     {
       "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": 28430.282209352266,
-      "required_pages_per_second_upper_estimate": 0.00173125502438666
+      "required_token_positions_per_second": 28288.02788734161,
+      "required_pages_per_second_upper_estimate": 0.0017225924825269574
     },
     {
       "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": 28430.282209352266,
-      "required_pages_per_second_upper_estimate": 0.002188641824534121
+      "required_token_positions_per_second": 28288.02788734161,
+      "required_pages_per_second_upper_estimate": 0.0021776906930405736
     },
     {
       "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": 28430.282209352266,
-      "required_pages_per_second_upper_estimate": 0.3530089550064227
+      "required_token_positions_per_second": 28288.02788734161,
+      "required_pages_per_second_upper_estimate": 0.3512426324216399
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08339257680466972,
-    "producer_p95_seconds": 0.2141006700003345,
-    "gpu_update_seconds": 0.9724841912017697,
-    "producer_batches_per_second": 11.991475000734157,
-    "gpu_batches_per_second": 1.0282943507433546,
-    "producer_headroom": 11.661519867405197,
+    "producer_batch_seconds": 0.16526897266396645,
+    "producer_p95_seconds": 0.27403312000024016,
+    "gpu_update_seconds": 0.9773746020793478,
+    "producer_batches_per_second": 6.050742519185694,
+    "gpu_batches_per_second": 1.0231491568048905,
+    "producer_headroom": 5.913842061973709,
     "sustainable": true,
-    "buffer_seconds": 60.29401985450972,
+    "buffer_seconds": 60.59722532891956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484501.17288276687,
+    "generation_bytes_per_second": 495377.05611306115,
     "upload_bytes_per_second": 5882090.189655369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.96267309057478,
+    "estimated_full_shards_per_hour": 106.29638445419194,
     "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": [

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 95900, "loss": 0.6055783797055483, "lr": 0.0003, "grad_norm": 0.15178553308913817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14033585492, "tokens_per_second": 147557.5283446402, "data_wait_s": 9.029998182086274e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 96000, "loss": 0.6212404612451792, "lr": 0.0003, "grad_norm": 0.1366567958379472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14048197658, "tokens_per_second": 149438.04337328838, "data_wait_s": 1.1753996659535915e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 96100, "loss": 0.6387616209685802, "lr": 0.0003, "grad_norm": 0.14333084500028356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14062985718, "tokens_per_second": 150250.28948042548, "data_wait_s": 1.4741999621037394e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 96200, "loss": 0.6147613078355789, "lr": 0.0003, "grad_norm": 0.1378868176610411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14077682953, "tokens_per_second": 150106.42102338027, "data_wait_s": 8.982999133877456e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2713418d9b01150cf05bfb2592c7b89ee4aa16668c3288d290a2038420cd07c0
-size 236829
+oid sha256:31ef6cc9d322d9f5561cb7521ad863ae8ac648123792e8e3a99d797c7de4503e
+size 238947