Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74916.16751091513,
-      "required_pages_per_second_upper_estimate": 0.005182874297862759
+      "required_token_positions_per_second": 75225.88506393171,
+      "required_pages_per_second_upper_estimate": 0.00520430127682418
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9364.52093886439,
-      "required_pages_per_second_upper_estimate": 0.003920729363668195
+      "required_token_positions_per_second": 9403.235632991464,
+      "required_pages_per_second_upper_estimate": 0.003936938397644454
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18729.04187772878,
-      "required_pages_per_second_upper_estimate": 0.003900765505082039
+      "required_token_positions_per_second": 18806.471265982927,
+      "required_pages_per_second_upper_estimate": 0.003916892004705095
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28093.562816593174,
-      "required_pages_per_second_upper_estimate": 0.0016841661944681213
+      "required_token_positions_per_second": 28209.70689897439,
+      "required_pages_per_second_upper_estimate": 0.0016911288548651297
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28093.562816593174,
-      "required_pages_per_second_upper_estimate": 0.0021841509384666244
+      "required_token_positions_per_second": 28209.70689897439,
+      "required_pages_per_second_upper_estimate": 0.0021931806300079354
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28093.562816593174,
-      "required_pages_per_second_upper_estimate": 0.3603540592936619
+      "required_token_positions_per_second": 28209.70689897439,
+      "required_pages_per_second_upper_estimate": 0.36184383087664845
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1973813759375318,
-    "producer_p95_seconds": 0.16964342899973417,
-    "gpu_update_seconds": 0.9841400387874618,
-    "producer_batches_per_second": 5.066334122204543,
-    "gpu_batches_per_second": 1.0161155532621953,
-    "producer_headroom": 4.98598225953662,
+    "producer_batch_seconds": 0.09632356767964012,
+    "producer_p95_seconds": 0.17093385699990904,
+    "gpu_update_seconds": 0.9800881696153032,
+    "producer_batches_per_second": 10.381675264831056,
+    "gpu_batches_per_second": 1.0203163664270252,
+    "producer_headroom": 10.174957107848737,
     "sustainable": true,
-    "buffer_seconds": 60.03254236603517,
+    "buffer_seconds": 60.7654665161488,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394478.1711725605,
-    "upload_bytes_per_second": 3566864.0620530853,
+    "generation_bytes_per_second": 420542.66153516225,
+    "upload_bytes_per_second": 3562887.7625868884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.64583255179033,
+    "estimated_full_shards_per_hour": 90.23866543332244,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 5956633,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.6070335669992346,
+      "bytes": 5667674,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 805000, "loss": 0.5351990461349487, "lr": 3e-05, "grad_norm": 0.28247715374620674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117828717221, "tokens_per_second": 150056.90640947784, "data_wait_s": 1.569699998071883e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805100, "loss": 0.52427738904953, "lr": 3e-05, "grad_norm": 0.27281520317803215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117843303051, "tokens_per_second": 137330.71672140184, "data_wait_s": 1.8432000615575816e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805200, "loss": 0.5211919546127319, "lr": 3e-05, "grad_norm": 0.2688684128774359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117857862950, "tokens_per_second": 149835.87029157692, "data_wait_s": 1.3558999853557907e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 805300, "loss": 0.5562940239906311, "lr": 3e-05, "grad_norm": 0.29062744684689046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117872513238, "tokens_per_second": 151925.33232381212, "data_wait_s": 1.9110999346594326e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4302b5e9dace7fdafb368c8be8aa20920f219a59dd2fcb657b7b11ffc98acf6b
-size 111084
+oid sha256:150436d665c19713444aa8d997997132a8e131600faf46d05a896f43f85eac04
+size 113202