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": 77618.69002823648,
-      "required_pages_per_second_upper_estimate": 0.005369841076327227
+      "required_token_positions_per_second": 77436.16535866597,
+      "required_pages_per_second_upper_estimate": 0.005357213596170756
     },
     {
       "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": 9702.33625352956,
-      "required_pages_per_second_upper_estimate": 0.00406216558153255
+      "required_token_positions_per_second": 9679.520669833246,
+      "required_pages_per_second_upper_estimate": 0.0040526131730824685
     },
     {
       "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": 19404.67250705912,
-      "required_pages_per_second_upper_estimate": 0.004041481547593673
+      "required_token_positions_per_second": 19359.04133966649,
+      "required_pages_per_second_upper_estimate": 0.004031977778800596
     },
     {
       "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": 29107.00876058868,
-      "required_pages_per_second_upper_estimate": 0.0017449207313682952
+      "required_token_positions_per_second": 29038.56200949974,
+      "required_pages_per_second_upper_estimate": 0.0017408174531526506
     },
     {
       "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": 29107.00876058868,
-      "required_pages_per_second_upper_estimate": 0.0022629419029346694
+      "required_token_positions_per_second": 29038.56200949974,
+      "required_pages_per_second_upper_estimate": 0.002257620469102945
     },
     {
       "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": 29107.00876058868,
-      "required_pages_per_second_upper_estimate": 0.37335345570976103
+      "required_token_positions_per_second": 29038.56200949974,
+      "required_pages_per_second_upper_estimate": 0.3724754942791875
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08526713728119262,
-    "producer_p95_seconds": 0.2112126050014922,
-    "gpu_update_seconds": 0.9498743147195462,
-    "producer_batches_per_second": 11.727847701773026,
-    "gpu_batches_per_second": 1.0527708608430513,
-    "producer_headroom": 11.139981298856858,
+    "producer_batch_seconds": 0.41328901357044856,
+    "producer_p95_seconds": 0.2212015179975424,
+    "gpu_update_seconds": 0.9521132620463497,
+    "producer_batches_per_second": 2.419614282414361,
+    "gpu_batches_per_second": 1.0502952115704478,
+    "producer_headroom": 2.303746847323475,
     "sustainable": true,
-    "buffer_seconds": 60.79195614205096,
+    "buffer_seconds": 60.93524877096638,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373934.0418998127,
-    "upload_bytes_per_second": 3535269.575914512,
+    "generation_bytes_per_second": 344246.53764559247,
+    "upload_bytes_per_second": 3570710.3344659293,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.23754065271173,
+    "estimated_full_shards_per_hour": 73.86729332948524,
     "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": 2.040032099999735,
-      "bytes": 7666715,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 34.010384213999714,
       "bytes": 15924428,
@@ -562,6 +556,12 @@
       "bytes": 7619870,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.496031877999485,
+      "bytes": 10106377,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 943700, "loss": 0.5552679896354675, "lr": 3e-05, "grad_norm": 0.35232737785301604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138131372594, "tokens_per_second": 151315.82610603806, "data_wait_s": 1.2191998393973336e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 943800, "loss": 0.5147156715393066, "lr": 3e-05, "grad_norm": 0.3471735735507548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138145894708, "tokens_per_second": 152289.5026058417, "data_wait_s": 9.80800177785568e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 943900, "loss": 0.5266441702842712, "lr": 3e-05, "grad_norm": 0.36553139360250847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138160589670, "tokens_per_second": 137394.35932784676, "data_wait_s": 1.3025997759541497e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 944000, "loss": 0.5293976068496704, "lr": 3e-05, "grad_norm": 0.3551377906111634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138175317139, "tokens_per_second": 153223.9130166885, "data_wait_s": 1.5906000044196844e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6efbad68a8617a51125d068e184d6dea4e746ccb30732a6eb655fe7cba3241c6
-size 402816
+oid sha256:80a128d988fb606cfabcb4d8947c7d3f3bb4f01cce8349506b36f8fc28f457f2
+size 404934