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": 74551.81914600158,
-      "required_pages_per_second_upper_estimate": 0.006461962733717255
+      "required_token_positions_per_second": 74593.6392598537,
+      "required_pages_per_second_upper_estimate": 0.006465587595193847
     },
     {
       "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": 9318.977393250198,
-      "required_pages_per_second_upper_estimate": 0.003717479413295914
+      "required_token_positions_per_second": 9324.204907481713,
+      "required_pages_per_second_upper_estimate": 0.0037195647468811683
     },
     {
       "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": 18637.954786500395,
-      "required_pages_per_second_upper_estimate": 0.004311562823055789
+      "required_token_positions_per_second": 18648.409814963426,
+      "required_pages_per_second_upper_estimate": 0.004313981409888494
     },
     {
       "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": 27956.932179750595,
-      "required_pages_per_second_upper_estimate": 0.0017024304910595903
+      "required_token_positions_per_second": 27972.61472244514,
+      "required_pages_per_second_upper_estimate": 0.0017033854756297456
     },
     {
       "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": 27956.932179750595,
-      "required_pages_per_second_upper_estimate": 0.002152202028938638
+      "required_token_positions_per_second": 27972.61472244514,
+      "required_pages_per_second_upper_estimate": 0.002153409314487321
     },
     {
       "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": 27956.932179750595,
-      "required_pages_per_second_upper_estimate": 0.34713153183941037
+      "required_token_positions_per_second": 27972.61472244514,
+      "required_pages_per_second_upper_estimate": 0.3473262565335826
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12547755939047534,
-    "producer_p95_seconds": 0.3064413140018587,
-    "gpu_update_seconds": 0.988949711013916,
-    "producer_batches_per_second": 7.969552522838656,
-    "gpu_batches_per_second": 1.011173762288433,
-    "producer_headroom": 7.881486664371514,
+    "producer_batch_seconds": 0.12515130437481048,
+    "producer_p95_seconds": 0.31810767099887016,
+    "gpu_update_seconds": 0.9883952670972631,
+    "producer_batches_per_second": 7.9903282270645875,
+    "gpu_batches_per_second": 1.0117409838847347,
+    "producer_headroom": 7.897602602184304,
     "sustainable": true,
-    "buffer_seconds": 60.32593237184888,
+    "buffer_seconds": 60.29211129293305,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455905.41175620694,
+    "generation_bytes_per_second": 456233.19046689157,
     "upload_bytes_per_second": 2799686.7873858884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.82668842806488,
+    "estimated_full_shards_per_hour": 97.89702210907994,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 218900, "loss": 0.5877895951271057, "lr": 0.0003, "grad_norm": 0.16471820727193348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32039711182, "tokens_per_second": 149834.3093207871, "data_wait_s": 1.4344997907755896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219000, "loss": 0.635651670396328, "lr": 0.0003, "grad_norm": 0.1789175456113893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32054448077, "tokens_per_second": 149585.77841550292, "data_wait_s": 1.5058998542372137e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219100, "loss": 0.6620151866227388, "lr": 0.0003, "grad_norm": 0.16267989680756995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32069099202, "tokens_per_second": 148261.5626955664, "data_wait_s": 1.4432996977120638e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 219200, "loss": 0.5834460780024529, "lr": 0.0003, "grad_norm": 0.15104916600150237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32083818204, "tokens_per_second": 149474.60381156084, "data_wait_s": 1.1145999451400712e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da524265527595c563d74b583a5f3d1c7f76b2fb3830c46a87e957eccec8d4d4
-size 513735
+oid sha256:3a7bb707e68288e2363fb96e1403735fb0c94eb4f0eecdc487b551fb61c11765
+size 515853