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": 74391.18559159151,
-      "required_pages_per_second_upper_estimate": 0.006448039424343011
+      "required_token_positions_per_second": 75857.48426310635,
+      "required_pages_per_second_upper_estimate": 0.006575134476889911
     },
     {
       "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": 9298.898198948938,
-      "required_pages_per_second_upper_estimate": 0.003709469522478434
+      "required_token_positions_per_second": 9482.185532888294,
+      "required_pages_per_second_upper_estimate": 0.003782585580376693
     },
     {
       "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": 18597.796397897877,
-      "required_pages_per_second_upper_estimate": 0.004302272886616097
+      "required_token_positions_per_second": 18964.371065776588,
+      "required_pages_per_second_upper_estimate": 0.004387073484536028
     },
     {
       "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": 27896.694596846814,
-      "required_pages_per_second_upper_estimate": 0.0016987623383029225
+      "required_token_positions_per_second": 28446.55659866488,
+      "required_pages_per_second_upper_estimate": 0.0017322460493106762
     },
     {
       "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": 27896.694596846814,
-      "required_pages_per_second_upper_estimate": 0.002147564773058403
+      "required_token_positions_per_second": 28446.55659866488,
+      "required_pages_per_second_upper_estimate": 0.002189894672073799
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27896.694596846814,
-      "required_pages_per_second_upper_estimate": 0.3463835826619667
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28446.55659866488,
+      "required_pages_per_second_upper_estimate": 0.3532110284548081
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15614454342983208,
-    "producer_p95_seconds": 0.38916420999885304,
-    "gpu_update_seconds": 0.9910851589967608,
-    "producer_batches_per_second": 6.404322418409568,
-    "gpu_batches_per_second": 1.008995030267897,
-    "producer_headroom": 6.347228902315966,
+    "producer_batch_seconds": 0.09390940593777941,
+    "producer_p95_seconds": 0.3676033980009379,
+    "gpu_update_seconds": 0.9719278290890799,
+    "producer_batches_per_second": 10.648560599590628,
+    "gpu_batches_per_second": 1.028882978829025,
+    "producer_headroom": 10.34963238648363,
     "sustainable": true,
-    "buffer_seconds": 60.456194698802406,
+    "buffer_seconds": 60.25952540352296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264737.2839899967,
+    "generation_bytes_per_second": 283230.86427743064,
     "upload_bytes_per_second": 5471648.822872058,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.80645837569166,
+    "estimated_full_shards_per_hour": 60.77475019685926,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -361,3 +361,4 @@
 {"step": 151100, "loss": 0.5722631234675646, "lr": 0.0003, "grad_norm": 0.14709183133475207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22113193280, "tokens_per_second": 149919.23307072968, "data_wait_s": 1.3406999642029405e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 151200, "loss": 0.6018664203584194, "lr": 0.0003, "grad_norm": 0.1628181430201471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22127652525, "tokens_per_second": 147863.55401999847, "data_wait_s": 1.0734001989476383e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 151300, "loss": 0.5892051421105862, "lr": 0.0003, "grad_norm": 0.13786515756407766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22142229831, "tokens_per_second": 150042.62136134115, "data_wait_s": 9.256000339519233e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 151400, "loss": 0.5989663805812597, "lr": 0.0003, "grad_norm": 0.1467814174992331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22156842202, "tokens_per_second": 137872.9737155881, "data_wait_s": 1.0670999472495168e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:faae5747b18aea81f58775901f17a7c354a6cf66025786b510e340afc9c69445
-size 774480
+oid sha256:0bcda3b06d4925dfbc41936961627fbba475d80095805c2912627c3aa0192e3d
+size 776598