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": 75742.33666980336,
-      "required_pages_per_second_upper_estimate": 0.005240030597251204
+      "required_token_positions_per_second": 77143.75945655801,
+      "required_pages_per_second_upper_estimate": 0.005336984277387759
     },
     {
       "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": 9467.79208372542,
-      "required_pages_per_second_upper_estimate": 0.003963966835474774
+      "required_token_positions_per_second": 9642.969932069751,
+      "required_pages_per_second_upper_estimate": 0.004037310142447092
     },
     {
       "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": 18935.58416745084,
-      "required_pages_per_second_upper_estimate": 0.003943782817144677
+      "required_token_positions_per_second": 19285.939864139502,
+      "required_pages_per_second_upper_estimate": 0.004016752669263874
     },
     {
       "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": 28403.376251176265,
-      "required_pages_per_second_upper_estimate": 0.001702739036813654
+      "required_token_positions_per_second": 28928.909796209253,
+      "required_pages_per_second_upper_estimate": 0.0017342439703951211
     },
     {
       "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": 28403.376251176265,
-      "required_pages_per_second_upper_estimate": 0.0022082375702801786
+      "required_token_positions_per_second": 28928.909796209253,
+      "required_pages_per_second_upper_estimate": 0.0022490954918287236
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28403.376251176265,
-      "required_pages_per_second_upper_estimate": 0.36432801338074505
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28928.909796209253,
+      "required_pages_per_second_upper_estimate": 0.3710689934224709
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09974229884375063,
-    "producer_p95_seconds": 0.24404906299969298,
-    "gpu_update_seconds": 0.9734054062976059,
-    "producer_batches_per_second": 10.02583669709208,
-    "gpu_batches_per_second": 1.0273211896403451,
-    "producer_headroom": 9.759203643606362,
+    "producer_batch_seconds": 0.0877402217888914,
+    "producer_p95_seconds": 0.2536444430006668,
+    "gpu_update_seconds": 0.955722154577111,
+    "producer_batches_per_second": 11.397281424773054,
+    "gpu_batches_per_second": 1.0463292026985407,
+    "producer_headroom": 10.892634359605788,
     "sustainable": true,
-    "buffer_seconds": 60.351135190451565,
+    "buffer_seconds": 60.21049573835799,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340960.9975066729,
+    "generation_bytes_per_second": 341665.40568189946,
     "upload_bytes_per_second": 5738258.984628578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.16229289913312,
+    "estimated_full_shards_per_hour": 73.31344249575365,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 706800, "loss": 0.5665615797042847, "lr": 8e-05, "grad_norm": 0.2775661532529385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103455644429, "tokens_per_second": 153525.07218293217, "data_wait_s": 9.347997547592968e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 706900, "loss": 0.5981852412223816, "lr": 8e-05, "grad_norm": 0.22925572127747484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103470381310, "tokens_per_second": 153123.57898828003, "data_wait_s": 1.2748001608997583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707000, "loss": 0.5401789546012878, "lr": 8e-05, "grad_norm": 0.22340848497574609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103485085267, "tokens_per_second": 153346.46700654848, "data_wait_s": 1.2869000784121454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 707100, "loss": 0.5317685008049011, "lr": 8e-05, "grad_norm": 0.21699219140543216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103499593639, "tokens_per_second": 149133.01091071288, "data_wait_s": 1.6828002117108554e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e45b7c130965dc485cb330b48cafb1ad6c226cc3ca1973a94da9ce2d11300d24
-size 469257
+oid sha256:760bce4f4047f960fcb9398025ec22a46cb1b0037e3beefdec0c7a02d58cc177
+size 471375