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": 75780.19990541549,
-      "required_pages_per_second_upper_estimate": 0.006568435664640681
+      "required_token_positions_per_second": 74562.41581806284,
+      "required_pages_per_second_upper_estimate": 0.006462881226394506
     },
     {
       "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": 9472.524988176936,
-      "required_pages_per_second_upper_estimate": 0.0037787318446533174
+      "required_token_positions_per_second": 9320.301977257855,
+      "required_pages_per_second_upper_estimate": 0.0037180078096608646
     },
     {
       "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": 18945.049976353872,
-      "required_pages_per_second_upper_estimate": 0.004382603890537642
+      "required_token_positions_per_second": 18640.60395451571,
+      "required_pages_per_second_upper_estimate": 0.004312175661452367
     },
     {
       "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": 28417.574964530806,
-      "required_pages_per_second_upper_estimate": 0.001730481219846794
+      "required_token_positions_per_second": 27960.905931773566,
+      "required_pages_per_second_upper_estimate": 0.0017026724717091237
     },
     {
       "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": 28417.574964530806,
-      "required_pages_per_second_upper_estimate": 0.0021876635856518598
+      "required_token_positions_per_second": 27960.905931773566,
+      "required_pages_per_second_upper_estimate": 0.002152507939369416
     },
     {
       "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": 28417.574964530806,
-      "required_pages_per_second_upper_estimate": 0.3528511735541528
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27960.905931773566,
+      "required_pages_per_second_upper_estimate": 0.3471808725402432
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.4492049770547055,
-    "producer_p95_seconds": 0.23380935699969996,
-    "gpu_update_seconds": 0.9729190486700098,
-    "producer_batches_per_second": 2.226155209937082,
-    "gpu_batches_per_second": 1.0278347426407266,
-    "producer_headroom": 2.1658688090437717,
+    "producer_batch_seconds": 0.07415354603915603,
+    "producer_p95_seconds": 0.27997002999836695,
+    "gpu_update_seconds": 0.9888091633176308,
+    "producer_batches_per_second": 13.485531756929873,
+    "gpu_batches_per_second": 1.0113174888517638,
+    "producer_headroom": 13.334617373463168,
     "sustainable": true,
-    "buffer_seconds": 60.3209810175406,
+    "buffer_seconds": 60.31735896237548,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386285.4674081516,
+    "generation_bytes_per_second": 387054.7945179258,
     "upload_bytes_per_second": 5799553.831827062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.887869040331,
+    "estimated_full_shards_per_hour": 83.05294872907001,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -160,3 +160,4 @@
 {"step": 331000, "loss": 0.5964784566313028, "lr": 0.0003, "grad_norm": 0.1577556859694088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48449730099, "tokens_per_second": 150677.61027192196, "data_wait_s": 1.105499904952012e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331100, "loss": 0.6439957339316607, "lr": 0.0003, "grad_norm": 0.16588003596327064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48464504517, "tokens_per_second": 149608.35486879697, "data_wait_s": 1.7828999261837453e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331200, "loss": 0.5629096124321222, "lr": 0.0003, "grad_norm": 0.15336929508148103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48479199854, "tokens_per_second": 148608.9672753796, "data_wait_s": 1.3127999409334734e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 331300, "loss": 0.6037327870726585, "lr": 0.0003, "grad_norm": 0.1496243188968997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48493695266, "tokens_per_second": 147551.1896298099, "data_wait_s": 1.3197000953368843e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78bd9f5b05df149f57367e49a8a7f10d23de8506d690fb632b91891a632d9f62
-size 345630
+oid sha256:5718dc2e3d37813af279de91435a3d73b0fc7da1c1ca66136a530ff2742b0052
+size 347748