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": 77453.72648399828,
-      "required_pages_per_second_upper_estimate": 0.00535842851556867
+      "required_token_positions_per_second": 77249.26554278136,
+      "required_pages_per_second_upper_estimate": 0.005344283433240539
     },
     {
       "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": 9681.715810499785,
-      "required_pages_per_second_upper_estimate": 0.004053532232639799
+      "required_token_positions_per_second": 9656.15819284767,
+      "required_pages_per_second_upper_estimate": 0.004042831791832605
     },
     {
       "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": 19363.43162099957,
-      "required_pages_per_second_upper_estimate": 0.004032892158622772
+      "required_token_positions_per_second": 19312.31638569534,
+      "required_pages_per_second_upper_estimate": 0.004022246203108305
     },
     {
       "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": 29045.147431499354,
-      "required_pages_per_second_upper_estimate": 0.0017412122391461703
+      "required_token_positions_per_second": 28968.47457854301,
+      "required_pages_per_second_upper_estimate": 0.0017366158186841053
     },
     {
       "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": 29045.147431499354,
-      "required_pages_per_second_upper_estimate": 0.0022581324566972047
+      "required_token_positions_per_second": 28968.47457854301,
+      "required_pages_per_second_upper_estimate": 0.0022521714796280875
     },
     {
       "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": 29045.147431499354,
-      "required_pages_per_second_upper_estimate": 0.37255996500172334
+      "required_token_positions_per_second": 28968.47457854301,
+      "required_pages_per_second_upper_estimate": 0.3715764879688948
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11005713519494975,
-    "producer_p95_seconds": 0.3142559350017109,
-    "gpu_update_seconds": 0.9518973888910561,
-    "producer_batches_per_second": 9.086189625312795,
-    "gpu_batches_per_second": 1.0505333995767996,
-    "producer_headroom": 8.649120179304253,
+    "producer_batch_seconds": 0.12104418198424582,
+    "producer_p95_seconds": 0.3251895110006444,
+    "gpu_update_seconds": 0.9544168411435414,
+    "producer_batches_per_second": 8.261446222422753,
+    "gpu_batches_per_second": 1.0477602205780892,
+    "producer_headroom": 7.884863406881967,
     "sustainable": true,
-    "buffer_seconds": 60.92143288902759,
+    "buffer_seconds": 60.12826099204311,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314172.3132429037,
+    "generation_bytes_per_second": 314840.43812545785,
     "upload_bytes_per_second": 5685149.198085611,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.4140648647817,
+    "estimated_full_shards_per_hour": 67.55742891142656,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 521300, "loss": 0.6548756957054138, "lr": 0.0003, "grad_norm": 0.16964390841660704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76300837391, "tokens_per_second": 149669.21883760186, "data_wait_s": 1.5433004591614008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521400, "loss": 0.5528781414031982, "lr": 0.0003, "grad_norm": 0.14258201655071912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76315482106, "tokens_per_second": 152911.77912135163, "data_wait_s": 1.5258003259077668e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521500, "loss": 0.57232266664505, "lr": 0.0003, "grad_norm": 0.16306147595004902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76330023305, "tokens_per_second": 151847.95394713708, "data_wait_s": 1.5318000805564225e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 521600, "loss": 0.6300690174102783, "lr": 0.0003, "grad_norm": 0.20283871393591768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76344462949, "tokens_per_second": 150974.9642086803, "data_wait_s": 1.1862000974360853e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d129a7c28ef083c7e859f4b3c389d6610f48a7aabe8a0a0c839a5dc73748cf0e
-size 778716
+oid sha256:53a2e9c76f7d19efcc5692139dc488d6bc333fda5821ca262a7abe38f146033b
+size 780834