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": 74391.20552088397,
-      "required_pages_per_second_upper_estimate": 0.006448041151763572
+      "required_token_positions_per_second": 74717.25735989158,
+      "required_pages_per_second_upper_estimate": 0.00647630249879792
     },
     {
       "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": 9298.900690110497,
-      "required_pages_per_second_upper_estimate": 0.003709470516240026
+      "required_token_positions_per_second": 9339.657169986447,
+      "required_pages_per_second_upper_estimate": 0.0037257288854262195
     },
     {
       "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": 18597.801380220993,
-      "required_pages_per_second_upper_estimate": 0.004302274039188864
+      "required_token_positions_per_second": 18679.314339972894,
+      "required_pages_per_second_upper_estimate": 0.004321130627848372
     },
     {
       "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": 27896.70207033149,
-      "required_pages_per_second_upper_estimate": 0.001698762793398929
+      "required_token_positions_per_second": 28018.97150995934,
+      "required_pages_per_second_upper_estimate": 0.001706208360774093
     },
     {
       "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": 27896.70207033149,
-      "required_pages_per_second_upper_estimate": 0.0021475653483879415
+      "required_token_positions_per_second": 28018.97150995934,
+      "required_pages_per_second_upper_estimate": 0.002156977988313965
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27896.70207033149,
-      "required_pages_per_second_upper_estimate": 0.3463836754576342
+      "required_token_positions_per_second": 28018.97150995934,
+      "required_pages_per_second_upper_estimate": 0.3479018526883214
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.39943989945311387,
-    "producer_p95_seconds": 0.37381959799677134,
-    "gpu_update_seconds": 0.991084893486532,
-    "producer_batches_per_second": 2.503505537051087,
-    "gpu_batches_per_second": 1.0089953005762258,
-    "producer_headroom": 2.4811865185312194,
+    "producer_batch_seconds": 0.1377993589376274,
+    "producer_p95_seconds": 0.3479638210046687,
+    "gpu_update_seconds": 0.9867599883233586,
+    "producer_batches_per_second": 7.2569278094583405,
+    "gpu_batches_per_second": 1.0134176616738766,
+    "producer_headroom": 7.160846000524568,
     "sustainable": true,
-    "buffer_seconds": 60.45617850267845,
+    "buffer_seconds": 60.19235928772488,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370867.69314111,
-    "upload_bytes_per_second": 6006710.253326657,
+    "generation_bytes_per_second": 386261.2749985883,
+    "upload_bytes_per_second": 6032141.1302028615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.57957359003997,
+    "estimated_full_shards_per_hour": 82.88267791240918,
     "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": 1.015233548998367,
-      "bytes": 5854223,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8780484679991787,
       "bytes": 2637899,
@@ -562,6 +556,12 @@
       "bytes": 9406921,
       "pages": 24,
       "shards": 3
+    },
+    {
+      "seconds": 0.883425399006228,
+      "bytes": 6847863,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 114000, "loss": 0.584840752184391, "lr": 0.0003, "grad_norm": 0.1348682865008997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16682054620, "tokens_per_second": 149707.6124610946, "data_wait_s": 1.875000452855602e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114100, "loss": 0.6048310548067093, "lr": 0.0003, "grad_norm": 0.1343600350262735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16696750733, "tokens_per_second": 139348.7362881526, "data_wait_s": 8.791001164354384e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 114200, "loss": 0.5817533507943153, "lr": 0.0003, "grad_norm": 0.151245004184825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16711235399, "tokens_per_second": 147684.33645460388, "data_wait_s": 1.0333998943679035e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 114300, "loss": 0.6063061691820621, "lr": 0.0003, "grad_norm": 0.1685130957539429, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16725834847, "tokens_per_second": 148726.3324610475, "data_wait_s": 1.841800258262083e-05, "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:e8b6c3e04e88da7328073ebfdb96f05bb470af7c1169e8a7d36e6488ae242b1f
-size 622536
+oid sha256:8e022d5546d1689fec398fad1cf8553e3fbc1797bfcf9b9a1e42e47b3af9e277
+size 624654