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": 75891.39846743303,
-      "required_pages_per_second_upper_estimate": 0.0065780740741661825
+      "required_token_positions_per_second": 75835.51900253801,
+      "required_pages_per_second_upper_estimate": 0.006573230583774292
     },
     {
       "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": 9486.424808429128,
-      "required_pages_per_second_upper_estimate": 0.0037842766907727494
+      "required_token_positions_per_second": 9479.439875317252,
+      "required_pages_per_second_upper_estimate": 0.0037814902965203653
     },
     {
       "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": 18972.849616858257,
-      "required_pages_per_second_upper_estimate": 0.004389034848111369
+      "required_token_positions_per_second": 18958.879750634504,
+      "required_pages_per_second_upper_estimate": 0.0043858031654217524
     },
     {
       "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": 28459.274425287385,
-      "required_pages_per_second_upper_estimate": 0.0017330204982267073
+      "required_token_positions_per_second": 28438.319625951757,
+      "required_pages_per_second_upper_estimate": 0.001731744460888503
     },
     {
       "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": 28459.274425287385,
-      "required_pages_per_second_upper_estimate": 0.0021908737255723967
+      "required_token_positions_per_second": 28438.319625951757,
+      "required_pages_per_second_upper_estimate": 0.0021892605671129427
     },
     {
       "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": 28459.274425287385,
-      "required_pages_per_second_upper_estimate": 0.35336894129763197
+      "required_token_positions_per_second": 28438.319625951757,
+      "required_pages_per_second_upper_estimate": 0.35310875282108545
     }
   ],
   "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": 0.43261033196868937,
-    "producer_p95_seconds": 0.2808711109992146,
-    "gpu_update_seconds": 0.9714934958226998,
-    "producer_batches_per_second": 2.31154904564872,
-    "gpu_batches_per_second": 1.029342969664619,
-    "producer_headroom": 2.2456548631229007,
+    "producer_batch_seconds": 0.4411348145470413,
+    "producer_p95_seconds": 0.2608998679970682,
+    "gpu_update_seconds": 0.9722093416085478,
+    "producer_batches_per_second": 2.266880706359128,
+    "gpu_batches_per_second": 1.0285850559155005,
+    "producer_headroom": 2.203882599034528,
     "sustainable": true,
-    "buffer_seconds": 60.23259674100739,
+    "buffer_seconds": 60.27697917972996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319229.5111620504,
-    "upload_bytes_per_second": 5179174.581816344,
+    "generation_bytes_per_second": 318394.2210353851,
+    "upload_bytes_per_second": 5284157.991582592,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.4992218126882,
+    "estimated_full_shards_per_hour": 68.3199879960648,
     "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": 1.1802483909996226,
-      "bytes": 1055971,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.59867495799881,
       "bytes": 13042114,
@@ -562,6 +556,12 @@
       "bytes": 6123105,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9817681739987165,
+      "bytes": 8356010,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -158,3 +158,4 @@
 {"step": 210800, "loss": 0.5892199259251356, "lr": 0.0003, "grad_norm": 0.13936088710343394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30853566654, "tokens_per_second": 146726.40569528582, "data_wait_s": 1.5040999642224051e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210900, "loss": 0.617307286709547, "lr": 0.0003, "grad_norm": 0.1489735590135397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30868213471, "tokens_per_second": 150020.40876580813, "data_wait_s": 1.0015999578172341e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 211000, "loss": 0.6516569443047047, "lr": 0.0003, "grad_norm": 0.16912665173726843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30882882175, "tokens_per_second": 149258.46164579518, "data_wait_s": 1.1176001862622797e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 211100, "loss": 0.5443230774253607, "lr": 0.0003, "grad_norm": 0.2085922932129044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30897658932, "tokens_per_second": 150112.22063284108, "data_wait_s": 1.3487999240169302e-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:a33b88effd27f2a2730cb766fc6ef367b97c6a34d90d85ff6430b0135d33ddb3
-size 341394
+oid sha256:6b1955c0efbe2022b9ad8b8ffa736424d700e392ff0989e80ef5946a3b33bd09
+size 343512