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": 75493.51950415203,
-      "required_pages_per_second_upper_estimate": 0.006543586934043999
+      "required_token_positions_per_second": 73070.51818891916,
+      "required_pages_per_second_upper_estimate": 0.006333567321080303
     },
     {
       "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": 9436.689938019004,
-      "required_pages_per_second_upper_estimate": 0.003764436707363573
+      "required_token_positions_per_second": 9133.814773614895,
+      "required_pages_per_second_upper_estimate": 0.003643615275895522
     },
     {
       "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": 18873.37987603801,
-      "required_pages_per_second_upper_estimate": 0.004366024274180253
+      "required_token_positions_per_second": 18267.62954722979,
+      "required_pages_per_second_upper_estimate": 0.0042258945964521455
     },
     {
       "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": 28310.06981405701,
-      "required_pages_per_second_upper_estimate": 0.0017239347202188734
+      "required_token_positions_per_second": 27401.444320844686,
+      "required_pages_per_second_upper_estimate": 0.001668604194871778
     },
     {
       "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": 28310.06981405701,
-      "required_pages_per_second_upper_estimate": 0.0021793875415750835
+      "required_token_positions_per_second": 27401.444320844686,
+      "required_pages_per_second_upper_estimate": 0.0021094390358712516
     },
     {
       "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": 28310.06981405701,
-      "required_pages_per_second_upper_estimate": 0.35151631938186184
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27401.444320844686,
+      "required_pages_per_second_upper_estimate": 0.3402342317300705
     }
   ],
   "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": 0.05003785724218801,
-    "producer_p95_seconds": 0.17836868799986405,
-    "gpu_update_seconds": 0.9766136283518358,
-    "producer_batches_per_second": 19.98486855981671,
-    "gpu_batches_per_second": 1.023946390844076,
-    "producer_headroom": 19.517494996337124,
+    "producer_batch_seconds": 0.46286813790620585,
+    "producer_p95_seconds": 0.18341375500040158,
+    "gpu_update_seconds": 1.0089979081492342,
+    "producer_batches_per_second": 2.16044250641991,
+    "gpu_batches_per_second": 0.9910823322064773,
+    "producer_headroom": 2.1798819696543776,
     "sustainable": true,
-    "buffer_seconds": 60.550044957813824,
+    "buffer_seconds": 60.53987448895405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294131.9337213493,
-    "upload_bytes_per_second": 5174791.147971014,
+    "generation_bytes_per_second": 301733.5344106243,
+    "upload_bytes_per_second": 5205925.937086028,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.113865935615145,
+    "estimated_full_shards_per_hour": 64.74499248732612,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 6395911,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.171668709000187,
+      "bytes": 7290583,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 318800, "loss": 0.5670101344585419, "lr": 0.0003, "grad_norm": 0.14808665536061574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46663481719, "tokens_per_second": 146857.02865334632, "data_wait_s": 1.0927999937848654e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 318900, "loss": 0.5861462354660034, "lr": 0.0003, "grad_norm": 0.14167617553776982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46678326630, "tokens_per_second": 150603.40771212755, "data_wait_s": 1.1279000318609178e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 319000, "loss": 0.6967397872358561, "lr": 0.0003, "grad_norm": 0.16118698416761207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46693053166, "tokens_per_second": 147345.79801610066, "data_wait_s": 1.3175999811210204e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 319100, "loss": 0.5913281161338091, "lr": 0.0003, "grad_norm": 0.13908416392842843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46707670445, "tokens_per_second": 148139.4046867686, "data_wait_s": 1.4523999197990634e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b7c78c0f28ff24b18e1dbecca3f77e974a1d2e59617028b0b00c14806547c888
-size 84885
+oid sha256:6b2bb69a7d76ec28af442487171e345d8603d6381b6eaf3a6dd708f5abc9f93c
+size 87003