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": 74810.35857793319,
-      "required_pages_per_second_upper_estimate": 0.006484372276414888
+      "required_token_positions_per_second": 75738.8509841473,
+      "required_pages_per_second_upper_estimate": 0.006564851644943037
     },
     {
       "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": 9351.294822241649,
-      "required_pages_per_second_upper_estimate": 0.0037303713189092264
+      "required_token_positions_per_second": 9467.356373018412,
+      "required_pages_per_second_upper_estimate": 0.003776670006788899
     },
     {
       "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": 18702.589644483298,
-      "required_pages_per_second_upper_estimate": 0.004326514959915485
+      "required_token_positions_per_second": 18934.712746036825,
+      "required_pages_per_second_upper_estimate": 0.004380212554232834
     },
     {
       "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": 28053.884466724947,
-      "required_pages_per_second_upper_estimate": 0.0017083343766669907
+      "required_token_positions_per_second": 28402.069119055235,
+      "required_pages_per_second_upper_estimate": 0.0017295369952102157
     },
     {
       "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": 28053.884466724947,
-      "required_pages_per_second_upper_estimate": 0.0021596656843710315
+      "required_token_positions_per_second": 28402.069119055235,
+      "required_pages_per_second_upper_estimate": 0.002186469902744223
     },
     {
       "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": 28053.884466724947,
-      "required_pages_per_second_upper_estimate": 0.34833535476520044
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28402.069119055235,
+      "required_pages_per_second_upper_estimate": 0.35265864284807275
     }
   ],
   "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": 0.12061177719539273,
-    "producer_p95_seconds": 0.31256341300104395,
-    "gpu_update_seconds": 0.9855319691215535,
-    "producer_batches_per_second": 8.291064299467093,
-    "gpu_batches_per_second": 1.0146804277605956,
-    "producer_headroom": 8.171108925167218,
+    "producer_batch_seconds": 0.1219421662970035,
+    "producer_p95_seconds": 0.31680942499951925,
+    "gpu_update_seconds": 0.9734502047757738,
+    "producer_batches_per_second": 8.20060878338335,
+    "gpu_batches_per_second": 1.0272739120028658,
+    "producer_headroom": 7.982884299470531,
     "sustainable": true,
-    "buffer_seconds": 60.117450116414766,
+    "buffer_seconds": 60.35391269609798,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334725.19702663005,
+    "generation_bytes_per_second": 333321.0479882551,
     "upload_bytes_per_second": 5995529.720062554,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.82423527812172,
+    "estimated_full_shards_per_hour": 71.52293758140316,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -176,3 +176,4 @@
 {"step": 102600, "loss": 0.5845480412244797, "lr": 0.0003, "grad_norm": 0.14226792437529956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15013821521, "tokens_per_second": 146951.14672099773, "data_wait_s": 1.2579002941492945e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 102700, "loss": 0.6356773227453232, "lr": 0.0003, "grad_norm": 0.12828419148673992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15028512680, "tokens_per_second": 149277.5575517598, "data_wait_s": 2.0441999367903918e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 102800, "loss": 0.587541026994586, "lr": 0.0003, "grad_norm": 0.1717889755210538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15043150245, "tokens_per_second": 148986.56333670835, "data_wait_s": 1.480399805586785e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 102900, "loss": 0.5999361556023359, "lr": 0.0003, "grad_norm": 0.13805758448796915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15057865595, "tokens_per_second": 148841.67114829342, "data_wait_s": 1.5591998817399144e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a68b239eeb74d5e7d3aead588f7cd811e5542a52f7a91467be5ef265304c43ee
-size 379518
+oid sha256:f57daeb8060f38319a39aec5f4f9eedf4a55a23d1081634b318484d5f7a06454
+size 381636