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": 76123.99454394318,
-      "required_pages_per_second_upper_estimate": 0.006598234912568716
+      "required_token_positions_per_second": 74147.94139234241,
+      "required_pages_per_second_upper_estimate": 0.006426955633648887
     },
     {
       "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": 9515.499317992897,
-      "required_pages_per_second_upper_estimate": 0.003795874947340393
+      "required_token_positions_per_second": 9268.492674042802,
+      "required_pages_per_second_upper_estimate": 0.0036973403039902673
     },
     {
       "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": 19030.998635985794,
-      "required_pages_per_second_upper_estimate": 0.004402486600298741
+      "required_token_positions_per_second": 18536.985348085604,
+      "required_pages_per_second_upper_estimate": 0.0042882053204799505
     },
     {
       "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": 28546.49795397869,
-      "required_pages_per_second_upper_estimate": 0.00173833195349752
+      "required_token_positions_per_second": 27805.47802212841,
+      "required_pages_per_second_upper_estimate": 0.001693207727478953
     },
     {
       "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": 28546.49795397869,
-      "required_pages_per_second_upper_estimate": 0.002197588434261234
+      "required_token_positions_per_second": 27805.47802212841,
+      "required_pages_per_second_upper_estimate": 0.0021405426686329407
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28546.49795397869,
-      "required_pages_per_second_upper_estimate": 0.354451965605606
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27805.47802212841,
+      "required_pages_per_second_upper_estimate": 0.3452509780862015
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07604392072652644,
-    "producer_p95_seconds": 0.18557740400046896,
-    "gpu_update_seconds": 0.9685251075131105,
-    "producer_batches_per_second": 13.150295124790555,
-    "gpu_batches_per_second": 1.0324977558586041,
-    "producer_headroom": 12.736390999566904,
+    "producer_batch_seconds": 0.2636729442187118,
+    "producer_p95_seconds": 0.2116567800003395,
+    "gpu_update_seconds": 0.9943364389562703,
+    "producer_batches_per_second": 3.792577213271145,
+    "gpu_batches_per_second": 1.0056958196661028,
+    "producer_headroom": 3.7710977207107255,
     "sustainable": true,
-    "buffer_seconds": 60.04855666581285,
+    "buffer_seconds": 60.65452277633249,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267233.1446026912,
-    "upload_bytes_per_second": 4583888.070190101,
+    "generation_bytes_per_second": 252709.38758476966,
+    "upload_bytes_per_second": 4585319.739734161,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.34201196251441,
+    "estimated_full_shards_per_hour": 54.22555180222814,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 7549840,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1135005340001953,
+      "bytes": 5221142,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 6100, "loss": 0.7579893618822098, "lr": 0.0003, "grad_norm": 0.16477188436430226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 891895665, "tokens_per_second": 139470.8196672479, "data_wait_s": 1.0054999620479066e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 6200, "loss": 0.7633047662675381, "lr": 0.0003, "grad_norm": 0.16709487775004292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 906568607, "tokens_per_second": 149296.6951368186, "data_wait_s": 9.794000106921885e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 6300, "loss": 0.7785959988832474, "lr": 0.0003, "grad_norm": 0.1863095277344159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 921415758, "tokens_per_second": 151907.7625647174, "data_wait_s": 9.184000191453379e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 6400, "loss": 0.7400187775492668, "lr": 0.0003, "grad_norm": 0.1717913486976937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 936159645, "tokens_per_second": 147453.16503139984, "data_wait_s": 1.0286000360792968e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99b5d49393441470d6cde76b9fe7094ff6d61b074f65d5b65711b9f67ae631cd
-size 132128
+oid sha256:e12c21f1199d52027682b81fff17bb79ff77ef34cc883d0d77c6da6da03727e4
+size 134211