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": 75778.10298895219,
-      "required_pages_per_second_upper_estimate": 0.0065682539092362284
+      "required_token_positions_per_second": 74710.17853230194,
+      "required_pages_per_second_upper_estimate": 0.006475688923963575
     },
     {
       "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": 9472.262873619024,
-      "required_pages_per_second_upper_estimate": 0.003778627283237204
+      "required_token_positions_per_second": 9338.772316537743,
+      "required_pages_per_second_upper_estimate": 0.0037253759041557936
     },
     {
       "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": 18944.525747238047,
-      "required_pages_per_second_upper_estimate": 0.004382482619357813
+      "required_token_positions_per_second": 18677.544633075486,
+      "required_pages_per_second_upper_estimate": 0.0043207212373034334
     },
     {
       "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": 28416.78862085707,
-      "required_pages_per_second_upper_estimate": 0.00173043333564269
+      "required_token_positions_per_second": 28016.31694961323,
+      "required_pages_per_second_upper_estimate": 0.0017060467119764165
     },
     {
       "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": 28416.78862085707,
-      "required_pages_per_second_upper_estimate": 0.002187603050739117
+      "required_token_positions_per_second": 28016.31694961323,
+      "required_pages_per_second_upper_estimate": 0.002156773632910228
     },
     {
       "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": 28416.78862085707,
-      "required_pages_per_second_upper_estimate": 0.352841409797448
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28016.31694961323,
+      "required_pages_per_second_upper_estimate": 0.34786889193306464
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13012236482020967,
-    "producer_p95_seconds": 0.32464972799789393,
-    "gpu_update_seconds": 0.9729459710907374,
-    "producer_batches_per_second": 7.685073979263303,
-    "gpu_batches_per_second": 1.027806301390953,
-    "producer_headroom": 7.477161765658493,
+    "producer_batch_seconds": 0.0765494699532212,
+    "producer_p95_seconds": 0.2569278779992601,
+    "gpu_update_seconds": 0.9868534843364444,
+    "producer_batches_per_second": 13.063447736621722,
+    "gpu_batches_per_second": 1.0133216489298766,
+    "producer_headroom": 12.891708916332183,
     "sustainable": true,
-    "buffer_seconds": 60.32265020762572,
+    "buffer_seconds": 60.198062544523104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415684.7691357818,
+    "generation_bytes_per_second": 416090.1794565348,
     "upload_bytes_per_second": 5361320.188986288,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.19627481036272,
+    "estimated_full_shards_per_hour": 89.28326642772706,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 259100, "loss": 0.560224512591958, "lr": 0.0003, "grad_norm": 0.17704939871046363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37924740155, "tokens_per_second": 147939.7315335934, "data_wait_s": 1.4435001503443345e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 259200, "loss": 0.43400827050209045, "lr": 0.0003, "grad_norm": 0.17114195344436686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37939434987, "tokens_per_second": 150179.4439340094, "data_wait_s": 9.211998985847458e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 259300, "loss": 0.5869931597262621, "lr": 0.0003, "grad_norm": 0.14652908417966715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37954120623, "tokens_per_second": 148289.87017773432, "data_wait_s": 9.011000656755641e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 259400, "loss": 0.6004841439425945, "lr": 0.0003, "grad_norm": 0.17627090721865235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37968785192, "tokens_per_second": 149266.07668842596, "data_wait_s": 9.916999260894954e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5792e5e35f9c7864d8fcbb21c31201281234b0f7a358ca8c6eaa5ee6cb4fd4fe
-size 473493
+oid sha256:f4e61ad060cf13633a5af3142946bd6e8d3a6ca8b1039dae5a2d27277efd33e7
+size 475611