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": 75401.73314238236,
-      "required_pages_per_second_upper_estimate": 0.006535631124836211
+      "required_token_positions_per_second": 73715.4478653424,
+      "required_pages_per_second_upper_estimate": 0.006389468190873348
     },
     {
       "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": 9425.216642797795,
-      "required_pages_per_second_upper_estimate": 0.003759859838358782
+      "required_token_positions_per_second": 9214.4309831678,
+      "required_pages_per_second_upper_estimate": 0.0036757742872059198
     },
     {
       "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": 18850.43328559559,
-      "required_pages_per_second_upper_estimate": 0.004360715984327598
+      "required_token_positions_per_second": 18428.8619663356,
+      "required_pages_per_second_upper_estimate": 0.004263192879018614
     },
     {
       "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": 28275.649928393388,
-      "required_pages_per_second_upper_estimate": 0.0017218387297691423
+      "required_token_positions_per_second": 27643.2929495034,
+      "required_pages_per_second_upper_estimate": 0.0016833315074754034
     },
     {
       "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": 28275.649928393388,
-      "required_pages_per_second_upper_estimate": 0.0021767378034963565
+      "required_token_positions_per_second": 27643.2929495034,
+      "required_pages_per_second_upper_estimate": 0.002128057212785252
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28275.649928393388,
-      "required_pages_per_second_upper_estimate": 0.3510889395978667
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27643.2929495034,
+      "required_pages_per_second_upper_estimate": 0.34323718228271977
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07115390350013229,
-    "producer_p95_seconds": 0.19268069399913657,
-    "gpu_update_seconds": 0.9778024579458694,
-    "producer_batches_per_second": 14.054042727229165,
-    "gpu_batches_per_second": 1.0227014586369136,
-    "producer_headroom": 13.742077522760948,
+    "producer_batch_seconds": 0.3760872807812916,
+    "producer_p95_seconds": 0.20377543799986597,
+    "gpu_update_seconds": 1.000170278211977,
+    "producer_batches_per_second": 2.6589572450378514,
+    "gpu_batches_per_second": 0.9998297507777562,
+    "producer_headroom": 2.6594100075232596,
     "sustainable": true,
-    "buffer_seconds": 60.6237523926439,
+    "buffer_seconds": 60.01021669271862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268058.95697252767,
+    "generation_bytes_per_second": 253500.79179574145,
     "upload_bytes_per_second": 3666752.2604912724,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.519212073153234,
+    "estimated_full_shards_per_hour": 54.39536872295554,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 245900, "loss": 0.61958840303123, "lr": 0.0003, "grad_norm": 0.16966736014105382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35992654748, "tokens_per_second": 121703.00145173802, "data_wait_s": 1.2460999641916715e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246000, "loss": 0.6332357935607433, "lr": 0.0003, "grad_norm": 0.15884492809949485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36007415822, "tokens_per_second": 151095.6949204287, "data_wait_s": 1.1965999874519184e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246100, "loss": 0.5597578007727861, "lr": 0.0003, "grad_norm": 0.13582123623624906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36022129649, "tokens_per_second": 149542.31262298633, "data_wait_s": 1.5158000678638928e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 246200, "loss": 0.6082983128726482, "lr": 0.0003, "grad_norm": 0.14485082765122037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36036857721, "tokens_per_second": 150080.03273420228, "data_wait_s": 9.61499972618185e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1074608dbdc7deed1541c1cae404b544522570ba41eaad42c6e06a00cf444d56
-size 192351
+oid sha256:8cacd0bc670c2be40644231aff0bbed9123498ac0cf2a3fcdaa4260bf083407c
+size 194469