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": 75855.05922464776,
-      "required_pages_per_second_upper_estimate": 0.006574924280702525
+      "required_token_positions_per_second": 75676.2474639588,
+      "required_pages_per_second_upper_estimate": 0.00655942533048027
     },
     {
       "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": 9481.88240308097,
-      "required_pages_per_second_upper_estimate": 0.003782464657364357
+      "required_token_positions_per_second": 9459.53093299485,
+      "required_pages_per_second_upper_estimate": 0.0037735483217627456
     },
     {
       "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": 18963.76480616194,
-      "required_pages_per_second_upper_estimate": 0.004386933237043983
+      "required_token_positions_per_second": 18919.0618659897,
+      "required_pages_per_second_upper_estimate": 0.004376591998580013
     },
     {
       "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": 28445.647209242907,
-      "required_pages_per_second_upper_estimate": 0.0017321906722660757
+      "required_token_positions_per_second": 28378.592798984548,
+      "required_pages_per_second_upper_estimate": 0.001728107410488643
     },
     {
       "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": 28445.647209242907,
-      "required_pages_per_second_upper_estimate": 0.002189824664758744
+      "required_token_positions_per_second": 28378.592798984548,
+      "required_pages_per_second_upper_estimate": 0.0021846626306385683
     },
     {
       "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": 28445.647209242907,
-      "required_pages_per_second_upper_estimate": 0.3531997368817178
+      "required_token_positions_per_second": 28378.592798984548,
+      "required_pages_per_second_upper_estimate": 0.3523671455229838
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.27217878266401385,
-    "producer_p95_seconds": 0.4064009000067017,
-    "gpu_update_seconds": 0.971958901009511,
-    "producer_batches_per_second": 3.6740556711006818,
-    "gpu_batches_per_second": 1.028850087139862,
-    "producer_headroom": 3.57103111233078,
+    "producer_batch_seconds": 0.4109166388745962,
+    "producer_p95_seconds": 0.39920570499816677,
+    "gpu_update_seconds": 0.974255495888764,
+    "producer_batches_per_second": 2.4335836162263087,
+    "gpu_batches_per_second": 1.0264247974169758,
+    "producer_headroom": 2.3709322128133343,
     "sustainable": true,
-    "buffer_seconds": 60.26145186258968,
+    "buffer_seconds": 60.40384074510337,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448659.6084735135,
-    "upload_bytes_per_second": 3722250.3003265085,
+    "generation_bytes_per_second": 389503.5030871587,
+    "upload_bytes_per_second": 3809519.977311894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.2719077172666,
+    "estimated_full_shards_per_hour": 83.5783845850093,
     "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": 1.009647611001128,
-      "bytes": 2308548,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.98430517000088,
       "bytes": 10007399,
@@ -562,6 +556,12 @@
       "bytes": 890694,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2476149669964798,
+      "bytes": 13021486,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 38100, "loss": 0.5695996237918735, "lr": 0.0003, "grad_norm": 0.1426385781318986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5575418628, "tokens_per_second": 147514.8865444126, "data_wait_s": 1.2013995728921145e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38200, "loss": 0.6496881563216448, "lr": 0.0003, "grad_norm": 0.13409868118504056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5589924159, "tokens_per_second": 147902.87036193, "data_wait_s": 1.016599708236754e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38300, "loss": 0.6396598033607006, "lr": 0.0003, "grad_norm": 0.12919586373820427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5604630828, "tokens_per_second": 149840.88248913095, "data_wait_s": 1.0178999218624085e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 38400, "loss": 0.6338763236999512, "lr": 0.0003, "grad_norm": 0.13404358026008098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5619227380, "tokens_per_second": 147564.20487816597, "data_wait_s": 9.408002370037138e-06, "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 sha256:6b525aa4bfdbccb8162dc3fe4eb9b52cc348b82acf78061c08f173455c507f90
-size 811066
+oid sha256:d5441295386f6f4a35deb282e4351e2d261d75e8757690889c283b5228e3fdf6
+size 813184