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": 75271.10357341544,
-      "required_pages_per_second_upper_estimate": 0.0065243084848757744
+      "required_token_positions_per_second": 75493.51950415203,
+      "required_pages_per_second_upper_estimate": 0.006543586934043999
     },
     {
       "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": 9408.88794667693,
-      "required_pages_per_second_upper_estimate": 0.003753346077340406
+      "required_token_positions_per_second": 9436.689938019004,
+      "required_pages_per_second_upper_estimate": 0.003764436707363573
     },
     {
       "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": 18817.77589335386,
-      "required_pages_per_second_upper_estimate": 0.004353161271382947
+      "required_token_positions_per_second": 18873.37987603801,
+      "required_pages_per_second_upper_estimate": 0.004366024274180253
     },
     {
       "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": 28226.66384003079,
-      "required_pages_per_second_upper_estimate": 0.001718855733998003
+      "required_token_positions_per_second": 28310.06981405701,
+      "required_pages_per_second_upper_estimate": 0.0017239347202188734
     },
     {
       "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": 28226.66384003079,
-      "required_pages_per_second_upper_estimate": 0.0021729667188120317
+      "required_token_positions_per_second": 28310.06981405701,
+      "required_pages_per_second_upper_estimate": 0.0021793875415750835
     },
     {
       "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": 28226.66384003079,
-      "required_pages_per_second_upper_estimate": 0.35048069632629464
+      "required_token_positions_per_second": 28310.06981405701,
+      "required_pages_per_second_upper_estimate": 0.35151631938186184
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.0676052911327858,
-    "producer_p95_seconds": 0.17054571399967244,
-    "gpu_update_seconds": 0.9794993895378407,
-    "producer_batches_per_second": 14.791741640988821,
-    "gpu_batches_per_second": 1.0209296817140765,
-    "producer_headroom": 14.488501907550008,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.728962151346124,
+    "buffer_seconds": 60.550044957813824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294581.1151844781,
-    "upload_bytes_per_second": 5176548.628307393,
+    "generation_bytes_per_second": 294131.9337213493,
+    "upload_bytes_per_second": 5174791.147971014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.21024982119329,
+    "estimated_full_shards_per_hour": 63.113865935615145,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 2537590,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.248541894000482,
+      "bytes": 6395911,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -37,3 +37,4 @@
 {"step": 318700, "loss": 0.5911545790731907, "lr": 0.0003, "grad_norm": 0.16510909979838936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46649008923, "tokens_per_second": 147937.58094382394, "data_wait_s": 1.1021000318578444e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:7f73f889de908bca590f8f3a9ce7b0fe475f024ae9271e9e79960a6c2dd86176
-size 82767
+oid sha256:b7c78c0f28ff24b18e1dbecca3f77e974a1d2e59617028b0b00c14806547c888
+size 84885