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": 75950.02866387811,
-      "required_pages_per_second_upper_estimate": 0.006583155990997168
+      "required_token_positions_per_second": 76108.15534855808,
+      "required_pages_per_second_upper_estimate": 0.006596862011256809
     },
     {
       "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": 9493.753582984764,
-      "required_pages_per_second_upper_estimate": 0.003787200248517937
+      "required_token_positions_per_second": 9513.51941856976,
+      "required_pages_per_second_upper_estimate": 0.0037950851358583694
     },
     {
       "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": 18987.507165969528,
-      "required_pages_per_second_upper_estimate": 0.004392425614134208
+      "required_token_positions_per_second": 19027.03883713952,
+      "required_pages_per_second_upper_estimate": 0.004401570570525736
     },
     {
       "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": 28481.26074895429,
-      "required_pages_per_second_upper_estimate": 0.0017343593499846966
+      "required_token_positions_per_second": 28540.55825570928,
+      "required_pages_per_second_upper_estimate": 0.0017379702570361013
     },
     {
       "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": 28481.26074895429,
-      "required_pages_per_second_upper_estimate": 0.002192566293630314
+      "required_token_positions_per_second": 28540.55825570928,
+      "required_pages_per_second_upper_estimate": 0.0021971311798463177
     },
     {
       "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": 28481.26074895429,
-      "required_pages_per_second_upper_estimate": 0.3536419378540831
+      "required_token_positions_per_second": 28540.55825570928,
+      "required_pages_per_second_upper_estimate": 0.35437821443199125
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09505588549245658,
+    "producer_batch_seconds": 0.2528820457423251,
     "producer_p95_seconds": 0.26696207799977856,
-    "gpu_update_seconds": 0.9707435441043499,
-    "producer_batches_per_second": 10.520127131731972,
-    "gpu_batches_per_second": 1.0301381925981732,
-    "producer_headroom": 10.212345496285824,
+    "gpu_update_seconds": 0.9687266714367533,
+    "producer_batches_per_second": 3.9544128056404326,
+    "gpu_batches_per_second": 1.0322829230218924,
+    "producer_headroom": 3.830745154694929,
     "sustainable": true,
-    "buffer_seconds": 60.18609973446969,
+    "buffer_seconds": 60.06105362907871,
     "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": 438338.2496630286,
-    "upload_bytes_per_second": 3956068.731694604,
+    "generation_bytes_per_second": 416055.0775628595,
+    "upload_bytes_per_second": 3965797.3101255987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.05718438547271,
+    "estimated_full_shards_per_hour": 89.27573437847461,
     "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": 1.7388112589997036,
-      "bytes": 12464285,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 0.8487006530012877,
       "bytes": 592565,
@@ -562,6 +556,12 @@
       "bytes": 11940488,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.3719915590008895,
+      "bytes": 12062246,
+      "pages": 44,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 291200, "loss": 0.5675464998930693, "lr": 0.0003, "grad_norm": 0.16711411768102727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42623302588, "tokens_per_second": 148623.04284251956, "data_wait_s": 9.75500006461516e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 291300, "loss": 0.5636729337275028, "lr": 0.0003, "grad_norm": 0.1601071855519649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42637864279, "tokens_per_second": 148545.567102421, "data_wait_s": 1.0681000276235864e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 291400, "loss": 0.5936985500156879, "lr": 0.0003, "grad_norm": 0.16376809380638718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42652512807, "tokens_per_second": 150215.8051876788, "data_wait_s": 1.1799002095358446e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 291500, "loss": 0.565649164840579, "lr": 0.0003, "grad_norm": 0.1470336876474914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42667207381, "tokens_per_second": 150758.1432170335, "data_wait_s": 8.61400258145295e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:589e8654b55b419795c6dd0ddf7adab120d416493ee0bcb23fdf00d6e4c472eb
-size 238947
+oid sha256:54f3f1f668364d37c0b794fef2dc56b52691dab26f016c2d51a8b310b391759a
+size 241065