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": 75835.51900253801,
-      "required_pages_per_second_upper_estimate": 0.006573230583774292
+      "required_token_positions_per_second": 75620.72331506542,
+      "required_pages_per_second_upper_estimate": 0.00655461263797886
     },
     {
       "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": 9479.439875317252,
-      "required_pages_per_second_upper_estimate": 0.0037814902965203653
+      "required_token_positions_per_second": 9452.590414383178,
+      "required_pages_per_second_upper_estimate": 0.0037707796451185486
     },
     {
       "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": 18958.879750634504,
-      "required_pages_per_second_upper_estimate": 0.0043858031654217524
+      "required_token_positions_per_second": 18905.180828766355,
+      "required_pages_per_second_upper_estimate": 0.004373380864916303
     },
     {
       "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": 28438.319625951757,
-      "required_pages_per_second_upper_estimate": 0.001731744460888503
+      "required_token_positions_per_second": 28357.771243149535,
+      "required_pages_per_second_upper_estimate": 0.0017268394869805516
     },
     {
       "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": 28438.319625951757,
-      "required_pages_per_second_upper_estimate": 0.0021892605671129427
+      "required_token_positions_per_second": 28357.771243149535,
+      "required_pages_per_second_upper_estimate": 0.002183059729632633
     },
     {
       "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": 28438.319625951757,
-      "required_pages_per_second_upper_estimate": 0.35310875282108545
+      "required_token_positions_per_second": 28357.771243149535,
+      "required_pages_per_second_upper_estimate": 0.3521086114847776
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.337962294234444,
+    "producer_p95_seconds": 0.27580087000023923,
+    "gpu_update_seconds": 0.9749708382558099,
+    "producer_batches_per_second": 2.95890996439473,
+    "gpu_batches_per_second": 1.0256717029495634,
+    "producer_headroom": 2.8848509283093984,
     "sustainable": true,
-    "buffer_seconds": 60.27697917972996,
+    "buffer_seconds": 60.44819197186021,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318394.2210353851,
+    "generation_bytes_per_second": 305434.18842347484,
     "upload_bytes_per_second": 5284157.991582592,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3199879960648,
+    "estimated_full_shards_per_hour": 65.5390666916674,
     "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

@@ -159,3 +159,4 @@
 {"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}
+{"step": 211200, "loss": 0.6435531042516232, "lr": 0.0003, "grad_norm": 0.16996013089216458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30912504703, "tokens_per_second": 150602.83732676326, "data_wait_s": 1.7579000996192917e-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 sha256:6b1955c0efbe2022b9ad8b8ffa736424d700e392ff0989e80ef5946a3b33bd09
-size 343512
+oid sha256:a0c31f14d14cac549aa25cae397dadcf2f180f0d112503b050a778c49cbfd397
+size 345630