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": 76079.09180030337,
-      "required_pages_per_second_upper_estimate": 0.006594342856554981
+      "required_token_positions_per_second": 76366.88537479124,
+      "required_pages_per_second_upper_estimate": 0.006619288074185443
     },
     {
       "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": 9509.886475037922,
-      "required_pages_per_second_upper_estimate": 0.0037936359003661727
+      "required_token_positions_per_second": 9545.860671848905,
+      "required_pages_per_second_upper_estimate": 0.0038079865453362473
     },
     {
       "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": 19019.772950075843,
-      "required_pages_per_second_upper_estimate": 0.004399889735428798
+      "required_token_positions_per_second": 19091.72134369781,
+      "required_pages_per_second_upper_estimate": 0.00441653373004476
     },
     {
       "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": 28529.659425113765,
-      "required_pages_per_second_upper_estimate": 0.0017373065754345794
+      "required_token_positions_per_second": 28637.582015546715,
+      "required_pages_per_second_upper_estimate": 0.0017438784949658742
     },
     {
       "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": 28529.659425113765,
-      "required_pages_per_second_upper_estimate": 0.002196292157697181
+      "required_token_positions_per_second": 28637.582015546715,
+      "required_pages_per_second_upper_estimate": 0.0022046003374575607
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28529.659425113765,
-      "required_pages_per_second_upper_estimate": 0.3542428874320345
+      "required_token_positions_per_second": 28637.582015546715,
+      "required_pages_per_second_upper_estimate": 0.3555829248115365
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10278326035158614,
-    "producer_p95_seconds": 0.19477145499877224,
-    "gpu_update_seconds": 0.9690967420263816,
-    "producer_batches_per_second": 9.729210735088033,
-    "gpu_batches_per_second": 1.0318887234199132,
-    "producer_headroom": 9.42854642586191,
+    "producer_batch_seconds": 0.07175434805472491,
+    "producer_p95_seconds": 0.19397214099990379,
+    "gpu_update_seconds": 0.9654446379233591,
+    "producer_batches_per_second": 13.936437680923945,
+    "gpu_batches_per_second": 1.0357921735947162,
+    "producer_headroom": 13.454859030801076,
     "sustainable": true,
-    "buffer_seconds": 60.083998005635664,
+    "buffer_seconds": 60.82301218917162,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315719.7560790267,
-    "upload_bytes_per_second": 5561178.492265481,
+    "generation_bytes_per_second": 303395.51142809395,
+    "upload_bytes_per_second": 5569777.238370519,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.74611007478809,
+    "estimated_full_shards_per_hour": 65.10161406643023,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 11921886,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.34920726099881,
+      "bytes": 7816018,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 283800, "loss": 0.5926853604614735, "lr": 0.0003, "grad_norm": 0.15031356883188096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41540394660, "tokens_per_second": 134554.33722788843, "data_wait_s": 1.3687998944078572e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 283900, "loss": 0.534436808899045, "lr": 0.0003, "grad_norm": 0.16632695571089925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41555084167, "tokens_per_second": 150336.53961063953, "data_wait_s": 1.3389999367063865e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 284000, "loss": 0.5654576811939478, "lr": 0.0003, "grad_norm": 0.14999594140498718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41569809586, "tokens_per_second": 150860.31564225064, "data_wait_s": 2.1712998204748146e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 284100, "loss": 0.6019531805068254, "lr": 0.0003, "grad_norm": 0.15106356147533648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41584654348, "tokens_per_second": 151793.60427882947, "data_wait_s": 1.229800000146497e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3cbaa3cbe6f6ad8de0c4ba36c1a8eb4212d80a93c264a6f136885404a3edfbb
-size 80649
+oid sha256:749571f3856304d7aa67d2882e7e9ca750e264ca6bc5793e7de9df3774e900c5
+size 82767