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": 75564.50328668191,
-      "required_pages_per_second_upper_estimate": 0.0065497396284069365
+      "required_token_positions_per_second": 75460.59299482622,
+      "required_pages_per_second_upper_estimate": 0.006540732947667113
     },
     {
       "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": 9445.562910835239,
-      "required_pages_per_second_upper_estimate": 0.0037679762688827344
+      "required_token_positions_per_second": 9432.574124353278,
+      "required_pages_per_second_upper_estimate": 0.003762794847755416
     },
     {
       "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": 18891.125821670477,
-      "required_pages_per_second_upper_estimate": 0.004370129486384348
+      "required_token_positions_per_second": 18865.148248706555,
+      "required_pages_per_second_upper_estimate": 0.004364120032068814
     },
     {
       "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": 28336.688732505714,
-      "required_pages_per_second_upper_estimate": 0.0017255556726937277
+      "required_token_positions_per_second": 28297.722373059834,
+      "required_pages_per_second_upper_estimate": 0.0017231828258441617
     },
     {
       "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": 28336.688732505714,
-      "required_pages_per_second_upper_estimate": 0.002181436739603147
+      "required_token_positions_per_second": 28297.722373059834,
+      "required_pages_per_second_upper_estimate": 0.002178437001387217
     },
     {
       "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": 28336.688732505714,
-      "required_pages_per_second_upper_estimate": 0.3518468372612056
+      "required_token_positions_per_second": 28297.722373059834,
+      "required_pages_per_second_upper_estimate": 0.35136300548890365
     }
   ],
   "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": 0.061236668226570146,
-    "producer_p95_seconds": 0.21764885099946696,
-    "gpu_update_seconds": 0.97569621704897,
-    "producer_batches_per_second": 16.33008504479849,
-    "gpu_batches_per_second": 1.0249091700125041,
-    "producer_headroom": 15.933202202297847,
+    "producer_batch_seconds": 0.0767402158359829,
+    "producer_p95_seconds": 0.19524387500132434,
+    "gpu_update_seconds": 0.9770397643847695,
+    "producer_batches_per_second": 13.030977162447694,
+    "gpu_batches_per_second": 1.0234997964793053,
+    "producer_headroom": 12.731782856501207,
     "sustainable": true,
-    "buffer_seconds": 60.49316545703614,
+    "buffer_seconds": 60.576465391855706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318958.8426562966,
-    "upload_bytes_per_second": 5239137.393507432,
+    "generation_bytes_per_second": 353905.0656742512,
+    "upload_bytes_per_second": 5240219.602624116,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.44114265219378,
+    "estimated_full_shards_per_hour": 75.93978860541011,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 7511604,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.111224463000326,
+      "bytes": 5878363,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -78,3 +78,4 @@
 {"step": 202800, "loss": 0.6024265885353088, "lr": 0.0003, "grad_norm": 0.14127869008033742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29682049824, "tokens_per_second": 147475.89511952884, "data_wait_s": 1.524700019217562e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202900, "loss": 0.5920193083584309, "lr": 0.0003, "grad_norm": 0.14651542795545644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29696746421, "tokens_per_second": 149019.61192145862, "data_wait_s": 1.833300120779313e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203000, "loss": 0.6431281287223101, "lr": 0.0003, "grad_norm": 0.1728700480867535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29711434566, "tokens_per_second": 148852.6840905166, "data_wait_s": 2.3517999579780735e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 203100, "loss": 0.6051948219537735, "lr": 0.0003, "grad_norm": 0.16271054594507742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29726077619, "tokens_per_second": 146998.69476981432, "data_wait_s": 1.6678000974934548e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81d83e8f358c697048568f7249d35fbb09ea974a2ae66867edfa40786fdd2fc2
-size 170388
+oid sha256:9fb54fa0c48ce608e757eda46011a5cfab982a3934093487600009b4fbfb7249
+size 172506