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": 76496.57104859932,
-      "required_pages_per_second_upper_estimate": 0.005292210281641179
+      "required_token_positions_per_second": 77041.42165718443,
+      "required_pages_per_second_upper_estimate": 0.005329904310970686
     },
     {
       "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": 9562.071381074915,
-      "required_pages_per_second_upper_estimate": 0.004003439608499402
+      "required_token_positions_per_second": 9630.177707148054,
+      "required_pages_per_second_upper_estimate": 0.0040319543050044105
     },
     {
       "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": 19124.14276214983,
-      "required_pages_per_second_upper_estimate": 0.00398305459979594
+      "required_token_positions_per_second": 19260.355414296107,
+      "required_pages_per_second_upper_estimate": 0.004011424103068809
     },
     {
       "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": 28686.214143224744,
-      "required_pages_per_second_upper_estimate": 0.001719694736573505
+      "required_token_positions_per_second": 28890.53312144416,
+      "required_pages_per_second_upper_estimate": 0.0017319433473407433
     },
     {
       "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": 28686.214143224744,
-      "required_pages_per_second_upper_estimate": 0.0022302269723145384
+      "required_token_positions_per_second": 28890.53312144416,
+      "required_pages_per_second_upper_estimate": 0.0022461118741668906
     },
     {
       "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": 28686.214143224744,
-      "required_pages_per_second_upper_estimate": 0.36795595417227517
+      "required_token_positions_per_second": 28890.53312144416,
+      "required_pages_per_second_upper_estimate": 0.3705767386442473
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10688356539859001,
-    "producer_p95_seconds": 0.3191386459984642,
-    "gpu_update_seconds": 0.9638079065421062,
-    "producer_batches_per_second": 9.355975320160791,
-    "gpu_batches_per_second": 1.0375511481201078,
-    "producer_headroom": 9.017362986983784,
+    "producer_batch_seconds": 0.12012242067990542,
+    "producer_p95_seconds": 0.3231012260002899,
+    "gpu_update_seconds": 0.9569916859539749,
+    "producer_batches_per_second": 8.324840561319824,
+    "gpu_batches_per_second": 1.0449411574596412,
+    "producer_headroom": 7.966803204075494,
     "sustainable": true,
-    "buffer_seconds": 60.71989811215269,
+    "buffer_seconds": 60.29047621510042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307596.45699518896,
+    "generation_bytes_per_second": 308338.13761648576,
     "upload_bytes_per_second": 4680928.332677589,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.00303919212104,
+    "estimated_full_shards_per_hour": 66.16218658801012,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 873700, "loss": 0.5502350926399231, "lr": 3e-05, "grad_norm": 0.3079950430755203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127885308473, "tokens_per_second": 152757.36369143435, "data_wait_s": 1.617399902897887e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 873800, "loss": 0.5340554118156433, "lr": 3e-05, "grad_norm": 0.30434055765120077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127899912309, "tokens_per_second": 151690.4765906996, "data_wait_s": 9.535000572213903e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 873900, "loss": 0.5336430668830872, "lr": 3e-05, "grad_norm": 0.31682278954037923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127914685429, "tokens_per_second": 152967.68867623864, "data_wait_s": 1.412599885952659e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 874000, "loss": 0.5015284419059753, "lr": 3e-05, "grad_norm": 0.32555418062398894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127929404004, "tokens_per_second": 153155.30740216115, "data_wait_s": 1.6776000848039985e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:335a81625b9041e7cac6f111d8223f21607a839c65cbb5590ba229fcb365f71e
-size 669915
+oid sha256:6266000c904f6653fe965e878ee2abd63dfd2ad3b4832b31133c4d6e02e7fda5
+size 672033