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": 75867.71998504823,
-      "required_pages_per_second_upper_estimate": 0.006576021683325633
+      "required_token_positions_per_second": 74829.81294767106,
+      "required_pages_per_second_upper_estimate": 0.006486058531877116
     },
     {
       "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": 9483.464998131029,
-      "required_pages_per_second_upper_estimate": 0.00378309597819173
+      "required_token_positions_per_second": 9353.726618458883,
+      "required_pages_per_second_upper_estimate": 0.0037313413987788745
     },
     {
       "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": 18966.929996262057,
-      "required_pages_per_second_upper_estimate": 0.004387665448068203
+      "required_token_positions_per_second": 18707.453236917765,
+      "required_pages_per_second_upper_estimate": 0.004327640066428895
     },
     {
       "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": 28450.394994393086,
-      "required_pages_per_second_upper_estimate": 0.001732479787472018
+      "required_token_positions_per_second": 28061.179855376646,
+      "required_pages_per_second_upper_estimate": 0.001708778627559933
     },
     {
       "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": 28450.394994393086,
-      "required_pages_per_second_upper_estimate": 0.0021901901624023154
+      "required_token_positions_per_second": 28061.179855376646,
+      "required_pages_per_second_upper_estimate": 0.0021602273035843675
     },
     {
       "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": 28450.394994393086,
-      "required_pages_per_second_upper_estimate": 0.3532586884834683
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28061.179855376646,
+      "required_pages_per_second_upper_estimate": 0.3484259390761594
     }
   ],
   "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": 0.10839925453126398,
-    "producer_p95_seconds": 0.1904043829999864,
-    "gpu_update_seconds": 0.9717967010809088,
-    "producer_batches_per_second": 9.225155692482968,
-    "gpu_batches_per_second": 1.029021809693037,
-    "producer_headroom": 8.964975868912715,
+    "producer_batch_seconds": 0.6841837303828555,
+    "producer_p95_seconds": 0.188120565000645,
+    "gpu_update_seconds": 0.9852757490060604,
+    "producer_batches_per_second": 1.461595702429843,
+    "gpu_batches_per_second": 1.0149442945376392,
+    "producer_headroom": 1.4400748004556025,
     "sustainable": true,
-    "buffer_seconds": 60.251395467016344,
+    "buffer_seconds": 60.10182068936969,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373588.8639750047,
-    "upload_bytes_per_second": 5844458.057408248,
+    "generation_bytes_per_second": 313421.77619097265,
+    "upload_bytes_per_second": 5833925.686442722,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.16347350537878,
+    "estimated_full_shards_per_hour": 67.25301708504567,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 7750938,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1050905230003991,
+      "bytes": 6187776,
+      "pages": 15,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -26,3 +26,4 @@
 {"step": 282800, "loss": 0.653716541826725, "lr": 0.0003, "grad_norm": 0.1598144832022321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41394340143, "tokens_per_second": 150733.54300563177, "data_wait_s": 1.3720000424655154e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282900, "loss": 0.5886063631623983, "lr": 0.0003, "grad_norm": 0.1679207484828719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41408818163, "tokens_per_second": 147980.88940932546, "data_wait_s": 8.70500116434414e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 283000, "loss": 0.5849144067615271, "lr": 0.0003, "grad_norm": 0.1710524655987325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41423324239, "tokens_per_second": 127116.93380241522, "data_wait_s": 9.403000149177387e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 283100, "loss": 0.5726057942956686, "lr": 0.0003, "grad_norm": 0.14841560961660563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41437897356, "tokens_per_second": 128897.20193413284, "data_wait_s": 1.4055998690309934e-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:ffd4bce07bd233dcbe3537efe49799d3451c6ff3a5579d2d6b37f7cac46f0568
-size 59469
+oid sha256:950fa7f2e08a161e8a5c8d6dfa4fc52d3522dc37eba999ef4b522564d7f50261
+size 61587