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": 75510.18194507554,
-      "required_pages_per_second_upper_estimate": 0.006545031192192678
+      "required_token_positions_per_second": 75442.67548914204,
+      "required_pages_per_second_upper_estimate": 0.006539179903685911
     },
     {
       "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": 9438.772743134443,
-      "required_pages_per_second_upper_estimate": 0.0037652675694648326
+      "required_token_positions_per_second": 9430.334436142755,
+      "required_pages_per_second_upper_estimate": 0.003761901402641916
     },
     {
       "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": 18877.545486268886,
-      "required_pages_per_second_upper_estimate": 0.00436698791479493
+      "required_token_positions_per_second": 18860.66887228551,
+      "required_pages_per_second_upper_estimate": 0.004363083807168137
     },
     {
       "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": 28316.318229403332,
-      "required_pages_per_second_upper_estimate": 0.0017243152159305626
+      "required_token_positions_per_second": 28291.003308428262,
+      "required_pages_per_second_upper_estimate": 0.0017227736700603066
     },
     {
       "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": 28316.318229403332,
-      "required_pages_per_second_upper_estimate": 0.002179868561885164
+      "required_token_positions_per_second": 28291.003308428262,
+      "required_pages_per_second_upper_estimate": 0.002177919749192317
     },
     {
       "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": 28316.318229403332,
-      "required_pages_per_second_upper_estimate": 0.35159390378836225
+      "required_token_positions_per_second": 28291.003308428262,
+      "required_pages_per_second_upper_estimate": 0.35127957719344227
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09369171724989656,
-    "producer_p95_seconds": 0.30448047499885433,
-    "gpu_update_seconds": 0.9763981240785267,
-    "producer_batches_per_second": 10.673302073573682,
-    "gpu_batches_per_second": 1.0241723896630255,
-    "producer_headroom": 10.42139212236079,
+    "producer_batch_seconds": 0.08593740876540323,
+    "producer_p95_seconds": 0.301379200998781,
+    "gpu_update_seconds": 0.9772718096485216,
+    "producer_batches_per_second": 11.636375989993557,
+    "gpu_batches_per_second": 1.0232567747550732,
+    "producer_headroom": 11.37190222149161,
     "sustainable": true,
-    "buffer_seconds": 60.53668369286865,
+    "buffer_seconds": 60.59085219820834,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359502.79462678975,
+    "generation_bytes_per_second": 387406.9563303268,
     "upload_bytes_per_second": 5967371.084257196,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.14093093016405,
+    "estimated_full_shards_per_hour": 83.12851445610383,
     "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

@@ -149,3 +149,4 @@
 {"step": 99900, "loss": 0.6030363850295544, "lr": 0.0003, "grad_norm": 0.12983061304929588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14618405878, "tokens_per_second": 148787.18523982857, "data_wait_s": 1.7555998056195676e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 100000, "loss": 0.5686009973287582, "lr": 0.0003, "grad_norm": 0.18355357225654073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14633093467, "tokens_per_second": 149100.08511330638, "data_wait_s": 1.8392998754279688e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 100100, "loss": 0.6103905867785215, "lr": 0.0003, "grad_norm": 0.1358339739445815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14647761693, "tokens_per_second": 124416.37253416263, "data_wait_s": 1.0945001122308895e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 100200, "loss": 0.6084248721599579, "lr": 0.0003, "grad_norm": 0.13180838639884784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14662512988, "tokens_per_second": 150314.62526358146, "data_wait_s": 9.77099989540875e-06, "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 sha256:4811b3c48a4d475a9d490a2a383f44b7768bb8210e73753330f779463b5decc7
-size 322332
+oid sha256:4d1203ffe67ae63401d00878c6c11c3702040315858c7e51ac3620e800ac937b
+size 324450