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": 75757.99663891453,
-      "required_pages_per_second_upper_estimate": 0.006566511141774042
+      "required_token_positions_per_second": 76005.52867466047,
+      "required_pages_per_second_upper_estimate": 0.006587966591268291
     },
     {
       "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": 9469.749579864316,
-      "required_pages_per_second_upper_estimate": 0.003777624692781361
+      "required_token_positions_per_second": 9500.691084332559,
+      "required_pages_per_second_upper_estimate": 0.003789967721530461
     },
     {
       "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": 18939.49915972863,
-      "required_pages_per_second_upper_estimate": 0.004381319806802429
+      "required_token_positions_per_second": 19001.382168665117,
+      "required_pages_per_second_upper_estimate": 0.004395635351816125
     },
     {
       "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": 28409.248739592946,
-      "required_pages_per_second_upper_estimate": 0.0017299741964324044
+      "required_token_positions_per_second": 28502.073252997674,
+      "required_pages_per_second_upper_estimate": 0.001735626722286169
     },
     {
       "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": 28409.248739592946,
-      "required_pages_per_second_upper_estimate": 0.0021870226098076773
+      "required_token_positions_per_second": 28502.073252997674,
+      "required_pages_per_second_upper_estimate": 0.0021941684978042724
     },
     {
       "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": 28409.248739592946,
-      "required_pages_per_second_upper_estimate": 0.3527477897065069
+      "required_token_positions_per_second": 28502.073252997674,
+      "required_pages_per_second_upper_estimate": 0.3539003594993317
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1419517003670876,
-    "producer_p95_seconds": 0.39495470599649707,
-    "gpu_update_seconds": 0.9732041932340146,
-    "producer_batches_per_second": 7.044649676009491,
-    "gpu_batches_per_second": 1.0275335915651385,
-    "producer_headroom": 6.855882604557079,
+    "producer_batch_seconds": 0.11146049896080967,
+    "producer_p95_seconds": 0.3835207550000632,
+    "gpu_update_seconds": 0.9700346972861755,
+    "producer_batches_per_second": 8.971788295615,
+    "gpu_batches_per_second": 1.0308909596715015,
+    "producer_headroom": 8.702945943452548,
     "sustainable": true,
-    "buffer_seconds": 60.33865998050891,
+    "buffer_seconds": 60.142151231742886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439914.42004433676,
-    "upload_bytes_per_second": 5684944.417797205,
+    "generation_bytes_per_second": 451540.410979131,
+    "upload_bytes_per_second": 5741301.671945947,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.39539385793283,
+    "estimated_full_shards_per_hour": 96.89006087332199,
     "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": [
-    {
-      "seconds": 0.9416690770012792,
-      "bytes": 2624850,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0678014170043753,
       "bytes": 5764775,
@@ -562,6 +556,12 @@
       "bytes": 6217543,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0400478770025074,
+      "bytes": 7313545,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -331,3 +331,4 @@
 {"step": 313300, "loss": 0.5946486592292786, "lr": 0.0003, "grad_norm": 0.1711374375078365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45858487289, "tokens_per_second": 148793.99725079077, "data_wait_s": 8.533999789506197e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 313400, "loss": 0.6749385930597782, "lr": 0.0003, "grad_norm": 0.1880838228090751, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45873017512, "tokens_per_second": 148866.20174536938, "data_wait_s": 1.0744995961431414e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 313500, "loss": 0.5657958760857582, "lr": 0.0003, "grad_norm": 0.14890168583214022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45887546884, "tokens_per_second": 148605.22678918854, "data_wait_s": 1.1411997547838837e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 313600, "loss": 0.5694801658391953, "lr": 0.0003, "grad_norm": 0.15211941678975824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45902177731, "tokens_per_second": 149532.32139601294, "data_wait_s": 9.321003744844347e-06, "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:d915261c8c710240d06230619279ae3659080fcb0bf27411d6b28b058bdd3ab4
-size 710157
+oid sha256:081fd117e3ace4a44548c5cbf6a134fd7e9a0f22cfa7b75923eb3050ae469176
+size 712275