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": 77268.49083886783,
-      "required_pages_per_second_upper_estimate": 0.005345613483832634
+      "required_token_positions_per_second": 77012.45227315802,
+      "required_pages_per_second_upper_estimate": 0.0053279001417655064
     },
     {
       "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": 9658.561354858479,
-      "required_pages_per_second_upper_estimate": 0.004043837945582801
+      "required_token_positions_per_second": 9626.556534144753,
+      "required_pages_per_second_upper_estimate": 0.004030438195486619
     },
     {
       "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": 19317.122709716958,
-      "required_pages_per_second_upper_estimate": 0.00402324723365072
+      "required_token_positions_per_second": 19253.113068289505,
+      "required_pages_per_second_upper_estimate": 0.004009915713389142
     },
     {
       "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": 28975.684064575435,
-      "required_pages_per_second_upper_estimate": 0.0017370480163635014
+      "required_token_positions_per_second": 28879.66960243426,
+      "required_pages_per_second_upper_estimate": 0.0017312920959637328
     },
     {
       "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": 28975.684064575435,
-      "required_pages_per_second_upper_estimate": 0.002252731985455931
+      "required_token_positions_per_second": 28879.66960243426,
+      "required_pages_per_second_upper_estimate": 0.002245267283347441
     },
     {
       "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": 28975.684064575435,
-      "required_pages_per_second_upper_estimate": 0.3716689635147758
+      "required_token_positions_per_second": 28879.66960243426,
+      "required_pages_per_second_upper_estimate": 0.3704373930867262
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06001966298436656,
-    "producer_p95_seconds": 0.13137776099983967,
-    "gpu_update_seconds": 0.9541793711714779,
-    "producer_batches_per_second": 16.661206515945814,
-    "gpu_batches_per_second": 1.048020980344887,
-    "producer_headroom": 15.897779556343307,
+    "producer_batch_seconds": 0.09006620921093145,
+    "producer_p95_seconds": 0.12962477400014905,
+    "gpu_update_seconds": 0.957351672668359,
+    "producer_batches_per_second": 11.10294314328296,
+    "gpu_batches_per_second": 1.0445482350417483,
+    "producer_headroom": 10.62942118976363,
     "sustainable": true,
-    "buffer_seconds": 60.11330038380311,
+    "buffer_seconds": 60.313155378106615,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 63,
+    "upload_pages_per_commit": 67,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338197.0762321755,
-    "upload_bytes_per_second": 4361711.095634004,
+    "generation_bytes_per_second": 340529.8246150719,
+    "upload_bytes_per_second": 4330775.872946521,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.56921973442029,
+    "estimated_full_shards_per_hour": 73.06977323378675,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 1025985,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.7697618709999006,
+      "bytes": 7336838,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -9,3 +9,4 @@
 {"step": 398000, "loss": 0.5890787243843079, "lr": 0.0003, "grad_norm": 0.1762822575074146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58256002437, "tokens_per_second": 152467.2838565922, "data_wait_s": 1.2849000086134765e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 398100, "loss": 0.5921404361724854, "lr": 0.0003, "grad_norm": 0.1505247779724851, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58270731354, "tokens_per_second": 153217.11141987887, "data_wait_s": 8.729999990464421e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 398200, "loss": 0.611016571521759, "lr": 0.0003, "grad_norm": 0.14642271618150887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58285400637, "tokens_per_second": 152303.38380564412, "data_wait_s": 1.460099997530051e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 398300, "loss": 0.5686750411987305, "lr": 0.0003, "grad_norm": 0.16674207471425945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58300080941, "tokens_per_second": 152062.93361204525, "data_wait_s": 1.1905999826922198e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:883406a0176826faa785fb6674985519869ffad0b1610583a861c382fe8facb8
-size 23463
+oid sha256:79dcaed01e872b3a1db72e316261628610c02812526be375537be607400afa1f
+size 25581