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": 73948.90425320524,
-      "required_pages_per_second_upper_estimate": 0.006409703598883494
+      "required_token_positions_per_second": 75004.89503419884,
+      "required_pages_per_second_upper_estimate": 0.006501234203395853
     },
     {
       "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": 9243.613031650655,
-      "required_pages_per_second_upper_estimate": 0.003687415442656237
+      "required_token_positions_per_second": 9375.611879274855,
+      "required_pages_per_second_upper_estimate": 0.0037400717565321743
     },
     {
       "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": 18487.22606330131,
-      "required_pages_per_second_upper_estimate": 0.004276694385678606
+      "required_token_positions_per_second": 18751.22375854971,
+      "required_pages_per_second_upper_estimate": 0.004337765606273575
     },
     {
       "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": 27730.839094951963,
-      "required_pages_per_second_upper_estimate": 0.0016886626084140937
+      "required_token_positions_per_second": 28126.835637824563,
+      "required_pages_per_second_upper_estimate": 0.0017127767202417426
     },
     {
       "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": 27730.839094951963,
-      "required_pages_per_second_upper_estimate": 0.0021347967574051234
+      "required_token_positions_per_second": 28126.835637824563,
+      "required_pages_per_second_upper_estimate": 0.002165281667464046
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27730.839094951963,
-      "required_pages_per_second_upper_estimate": 0.3443242124110901
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28126.835637824563,
+      "required_pages_per_second_upper_estimate": 0.3492411641583938
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11231119485910313,
-    "producer_p95_seconds": 0.35895233799965354,
-    "gpu_update_seconds": 0.9970127447399512,
-    "producer_batches_per_second": 8.903831904330838,
-    "gpu_batches_per_second": 1.0029962056912602,
-    "producer_headroom": 8.877233885640035,
+    "producer_batch_seconds": 0.11643662714055836,
+    "producer_p95_seconds": 0.34709825800018734,
+    "gpu_update_seconds": 0.9829758439950269,
+    "producer_batches_per_second": 8.588362824979754,
+    "gpu_batches_per_second": 1.01731899731715,
+    "producer_headroom": 8.442153196419987,
     "sustainable": true,
-    "buffer_seconds": 60.817777429137024,
+    "buffer_seconds": 60.944502327691666,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348485.00132539845,
+    "generation_bytes_per_second": 381817.239823324,
     "upload_bytes_per_second": 6341487.696620985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.77676896878685,
+    "estimated_full_shards_per_hour": 81.9290914156417,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -286,3 +286,4 @@
 {"step": 28600, "loss": 0.6345416530966759, "lr": 0.0003, "grad_norm": 0.13952403465690785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4184940693, "tokens_per_second": 149597.94950770514, "data_wait_s": 1.918100315378979e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28700, "loss": 0.661632664501667, "lr": 0.0003, "grad_norm": 0.12816804777186405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4199582893, "tokens_per_second": 148808.9875211151, "data_wait_s": 1.673099905019626e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28800, "loss": 0.6906627379357815, "lr": 0.0003, "grad_norm": 0.15367915047919362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4214172372, "tokens_per_second": 148736.3276621741, "data_wait_s": 1.518800127087161e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 28900, "loss": 0.6480767019093037, "lr": 0.0003, "grad_norm": 0.14121009485425134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4228921148, "tokens_per_second": 150899.2919792501, "data_wait_s": 1.1834999895654619e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ae79d60b2468d465b44ad3acaf89ec48d8bc83a95d1bd8a3bce25e9c1f71b9c
-size 608290
+oid sha256:92484f08b6a069b4d4dc4ada66f98a1651ac2525d3cbf44cdad8a9b2807cd967
+size 610408