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": 75337.0549066585,
-      "required_pages_per_second_upper_estimate": 0.006530024979289156
+      "required_token_positions_per_second": 75075.29130432717,
+      "required_pages_per_second_upper_estimate": 0.0065073359736728615
     },
     {
       "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": 9417.131863332312,
-      "required_pages_per_second_upper_estimate": 0.003756634698951776
+      "required_token_positions_per_second": 9384.411413040896,
+      "required_pages_per_second_upper_estimate": 0.0037435820221162026
     },
     {
       "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": 18834.263726664623,
-      "required_pages_per_second_upper_estimate": 0.004356975441443436
+      "required_token_positions_per_second": 18768.822826081792,
+      "required_pages_per_second_upper_estimate": 0.004341836840814113
     },
     {
       "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": 28251.395589996937,
-      "required_pages_per_second_upper_estimate": 0.0017203617678134774
+      "required_token_positions_per_second": 28153.23423912269,
+      "required_pages_per_second_upper_estimate": 0.0017143842565580415
     },
     {
       "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": 28251.395589996937,
-      "required_pages_per_second_upper_estimate": 0.0021748706373862915
+      "required_token_positions_per_second": 28153.23423912269,
+      "required_pages_per_second_upper_estimate": 0.0021673139048679815
     },
     {
       "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": 28251.395589996937,
-      "required_pages_per_second_upper_estimate": 0.3507877818890316
+      "required_token_positions_per_second": 28153.23423912269,
+      "required_pages_per_second_upper_estimate": 0.3495689464360814
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08134996351566315,
-    "producer_p95_seconds": 0.19123474700063525,
-    "gpu_update_seconds": 0.9786419191903362,
-    "producer_batches_per_second": 12.292568512430368,
-    "gpu_batches_per_second": 1.0218242039206067,
-    "producer_headroom": 12.03002284078355,
+    "producer_batch_seconds": 0.09975972471103489,
+    "producer_p95_seconds": 0.1937923690002208,
+    "gpu_update_seconds": 0.9820541315135795,
+    "producer_batches_per_second": 10.024085400161347,
+    "gpu_batches_per_second": 1.0182738078386389,
+    "producer_headroom": 9.844194481873403,
     "sustainable": true,
-    "buffer_seconds": 60.67579898980084,
+    "buffer_seconds": 60.88735615384193,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 67,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417856.7945421776,
-    "upload_bytes_per_second": 5310584.374241293,
+    "generation_bytes_per_second": 446660.70615734736,
+    "upload_bytes_per_second": 5095049.216793044,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.66234090041156,
+    "estimated_full_shards_per_hour": 95.84298981228176,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 10173347,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9085161899984087,
+      "bytes": 2572379,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -9,3 +9,4 @@
 {"step": 281100, "loss": 0.610587639734149, "lr": 0.0003, "grad_norm": 0.14617332899983632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41145514716, "tokens_per_second": 150904.03889516508, "data_wait_s": 9.845000022323802e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 281200, "loss": 0.6069388054311275, "lr": 0.0003, "grad_norm": 0.15914330458428488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41160209644, "tokens_per_second": 150465.07544183324, "data_wait_s": 1.5604999134666286e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 281300, "loss": 0.5778026226907969, "lr": 0.0003, "grad_norm": 0.1437492068679232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41174926929, "tokens_per_second": 150813.47408636252, "data_wait_s": 1.4073000784264877e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 281400, "loss": 0.5655051227658987, "lr": 0.0003, "grad_norm": 0.13728999638525194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41189387898, "tokens_per_second": 147921.70543950234, "data_wait_s": 1.2397998943924904e-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 sha256:2fe2a19e01ea74274035d292040e94f14ba8b3a133e1aeb00e18a6d73a2d269b
-size 23463
+oid sha256:aac60ca9e4953383db630c65d0e8c7f4875c4a01c36c9f201a23ae2b96c5bafe
+size 25581