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": 75089.55403435834,
-      "required_pages_per_second_upper_estimate": 0.0065085722309634044
+      "required_token_positions_per_second": 75666.67043597525,
+      "required_pages_per_second_upper_estimate": 0.006558595217967411
     },
     {
       "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": 9386.194254294793,
-      "required_pages_per_second_upper_estimate": 0.003744293224148234
+      "required_token_positions_per_second": 9458.333804496906,
+      "required_pages_per_second_upper_estimate": 0.0037730707693062492
     },
     {
       "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": 18772.388508589585,
-      "required_pages_per_second_upper_estimate": 0.004342661698708421
+      "required_token_positions_per_second": 18916.667608993812,
+      "required_pages_per_second_upper_estimate": 0.004376038129361498
     },
     {
       "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": 28158.58276288438,
-      "required_pages_per_second_upper_estimate": 0.001714709953593593
+      "required_token_positions_per_second": 28375.001413490718,
+      "required_pages_per_second_upper_estimate": 0.0017278887139546124
     },
     {
       "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": 28158.58276288438,
-      "required_pages_per_second_upper_estimate": 0.0021677256490328176
+      "required_token_positions_per_second": 28375.001413490718,
+      "required_pages_per_second_upper_estimate": 0.0021843861558416616
     },
     {
       "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": 28158.58276288438,
-      "required_pages_per_second_upper_estimate": 0.3496353572008441
+      "required_token_positions_per_second": 28375.001413490718,
+      "required_pages_per_second_upper_estimate": 0.3523225525347445
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.12152511371090213,
-    "producer_p95_seconds": 0.422912574998918,
-    "gpu_update_seconds": 0.9818675972727798,
-    "producer_batches_per_second": 8.228751814862848,
-    "gpu_batches_per_second": 1.0184672584955288,
-    "producer_headroom": 8.079544773013412,
+    "producer_batch_seconds": 0.12722127366407676,
+    "producer_p95_seconds": 0.4442140789979021,
+    "gpu_update_seconds": 0.9743788060872106,
+    "producer_batches_per_second": 7.8603206146203535,
+    "gpu_batches_per_second": 1.0262949006615567,
+    "producer_headroom": 7.658929815936469,
     "sustainable": true,
-    "buffer_seconds": 60.87579103091235,
+    "buffer_seconds": 60.41148597740706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299585.8615554249,
-    "upload_bytes_per_second": 5600676.810723267,
+    "generation_bytes_per_second": 332267.7093581426,
+    "upload_bytes_per_second": 5677308.77381865,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.28415188786565,
+    "estimated_full_shards_per_hour": 71.29691563184936,
     "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": 1.0676300510021974,
-      "bytes": 1203686,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.0235948650006321,
       "bytes": 9738901,
@@ -562,6 +556,12 @@
       "bytes": 2377167,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8975952609980595,
+      "bytes": 5947609,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -412,3 +412,4 @@
 {"step": 236200, "loss": 0.5758872497826815, "lr": 0.0003, "grad_norm": 0.16433870787146654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34573002617, "tokens_per_second": 148960.42532426154, "data_wait_s": 1.4053002814762294e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236300, "loss": 0.5963732190430164, "lr": 0.0003, "grad_norm": 0.1587367101003753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34587718002, "tokens_per_second": 148591.4770712124, "data_wait_s": 1.204499858431518e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236400, "loss": 0.6338792107999325, "lr": 0.0003, "grad_norm": 0.14740172653696962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34602370900, "tokens_per_second": 148298.6080736068, "data_wait_s": 1.0199000826105475e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 236500, "loss": 0.5867517907172441, "lr": 0.0003, "grad_norm": 0.14828568283742657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34616852937, "tokens_per_second": 146583.28181564133, "data_wait_s": 1.745999907143414e-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 sha256:8bd2c34923f3decc099cd35416cbada30c557ccc16971c687068d2c7deb74233
-size 883281
+oid sha256:9cec22fd1fa8cdb040ec684fa4f87a0b94ce0f07f8c26cc9801b94bbe69cf58b
+size 885399