Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76876.01424022709,
-      "required_pages_per_second_upper_estimate": 0.005318461042067492
+      "required_token_positions_per_second": 76872.39957130198,
+      "required_pages_per_second_upper_estimate": 0.005318210970884066
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9609.501780028386,
-      "required_pages_per_second_upper_estimate": 0.004023297726081861
+      "required_token_positions_per_second": 9609.049946412748,
+      "required_pages_per_second_upper_estimate": 0.004023108552782352
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19219.00356005677,
-      "required_pages_per_second_upper_estimate": 0.004002811602352487
+      "required_token_positions_per_second": 19218.099892825496,
+      "required_pages_per_second_upper_estimate": 0.004002623392299519
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28828.50534008516,
-      "required_pages_per_second_upper_estimate": 0.0017282248765592095
+      "required_token_positions_per_second": 28827.149839238245,
+      "required_pages_per_second_upper_estimate": 0.0017281436163531668
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28828.50534008516,
-      "required_pages_per_second_upper_estimate": 0.002241289486997605
+      "required_token_positions_per_second": 28827.149839238245,
+      "required_pages_per_second_upper_estimate": 0.0022411841027689754
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28828.50534008516,
-      "required_pages_per_second_upper_estimate": 0.3697811128652167
+      "required_token_positions_per_second": 28827.149839238245,
+      "required_pages_per_second_upper_estimate": 0.3697637259557759
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08136494914069203,
-    "producer_p95_seconds": 0.20143434099918522,
-    "gpu_update_seconds": 0.9590507615237442,
-    "producer_batches_per_second": 12.290304493042232,
-    "gpu_batches_per_second": 1.04269767578433,
-    "producer_headroom": 11.787025883410848,
+    "producer_batch_seconds": 0.06951333892970979,
+    "producer_p95_seconds": 0.18227206300070975,
+    "gpu_update_seconds": 0.9590958576961626,
+    "producer_batches_per_second": 14.385728198312785,
+    "gpu_batches_per_second": 1.0426486486993,
+    "producer_headroom": 13.797292324944673,
     "sustainable": true,
-    "buffer_seconds": 60.42019797599589,
+    "buffer_seconds": 60.42303903485824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300920.58324105124,
-    "upload_bytes_per_second": 4317504.904458179,
+    "generation_bytes_per_second": 301520.7081352791,
+    "upload_bytes_per_second": 4317809.417392553,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.57055209087041,
+    "estimated_full_shards_per_hour": 64.69932492297917,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 13214903,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.4333472569996957,
+      "bytes": 6210705,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 532200, "loss": 0.6312143802642822, "lr": 0.0003, "grad_norm": 0.17787364407634543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77896848290, "tokens_per_second": 143940.44404040152, "data_wait_s": 1.2266998965060338e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532300, "loss": 0.5710451602935791, "lr": 0.0003, "grad_norm": 0.16897770926561098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77911518820, "tokens_per_second": 152876.33175186967, "data_wait_s": 1.3333999959286302e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532400, "loss": 0.568967878818512, "lr": 0.0003, "grad_norm": 0.14396788905138194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77926188670, "tokens_per_second": 153249.58756346416, "data_wait_s": 9.742998372530565e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 532500, "loss": 0.6018000245094299, "lr": 0.0003, "grad_norm": 0.16666585614226395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77940993934, "tokens_per_second": 154926.7151380044, "data_wait_s": 1.0138999641640112e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59f68a1e5f13a9ee9966770c303029b60eb6fe26ad777bf5ff29b5919aaf96da
-size 157680
+oid sha256:f169624bf198e371e4d92b84c960118783f78bb0ba700fff3eeb56c32f30de63
+size 159798