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": 75427.17015186643,
-      "required_pages_per_second_upper_estimate": 0.006537835940348779
+      "required_token_positions_per_second": 75363.28276656098,
+      "required_pages_per_second_upper_estimate": 0.006532298343711601
     },
     {
       "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": 9428.396268983304,
-      "required_pages_per_second_upper_estimate": 0.00376112823878383
+      "required_token_positions_per_second": 9420.410345820123,
+      "required_pages_per_second_upper_estimate": 0.003757942534633845
     },
     {
       "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": 18856.79253796661,
-      "required_pages_per_second_upper_estimate": 0.004362187085444613
+      "required_token_positions_per_second": 18840.820691640245,
+      "required_pages_per_second_upper_estimate": 0.004358492279891904
     },
     {
       "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": 28285.18880694991,
-      "required_pages_per_second_upper_estimate": 0.0017224195974266067
+      "required_token_positions_per_second": 28261.23103746037,
+      "required_pages_per_second_upper_estimate": 0.0017209606949613958
     },
     {
       "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": 28285.18880694991,
-      "required_pages_per_second_upper_estimate": 0.0021774721327729434
+      "required_token_positions_per_second": 28261.23103746037,
+      "required_pages_per_second_upper_estimate": 0.0021756277973582878
     },
     {
       "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": 28285.18880694991,
-      "required_pages_per_second_upper_estimate": 0.3512073805449658
+      "required_token_positions_per_second": 28261.23103746037,
+      "required_pages_per_second_upper_estimate": 0.35090990522940224
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1195486120001874,
-    "producer_p95_seconds": 0.40861041500465944,
-    "gpu_update_seconds": 0.9774727044850643,
-    "producer_batches_per_second": 8.364798078947436,
-    "gpu_batches_per_second": 1.0230464701587787,
-    "producer_headroom": 8.17636180070022,
+    "producer_batch_seconds": 0.28977639320362414,
+    "producer_p95_seconds": 0.42413917000521906,
+    "gpu_update_seconds": 0.9783013331355761,
+    "producer_batches_per_second": 3.4509367341642148,
+    "gpu_batches_per_second": 1.0221799420377737,
+    "producer_headroom": 3.376056007599382,
     "sustainable": true,
-    "buffer_seconds": 60.60330767807399,
+    "buffer_seconds": 60.65468265440572,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412434.25987987785,
+    "generation_bytes_per_second": 386078.71496347134,
     "upload_bytes_per_second": 3645106.9026584383,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.49879119202853,
+    "estimated_full_shards_per_hour": 82.8435047786532,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 39500, "loss": 0.6539227645844221, "lr": 0.0003, "grad_norm": 0.13698722597409166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5780151611, "tokens_per_second": 150391.8955422777, "data_wait_s": 8.754999726079404e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 39600, "loss": 0.6563158668577671, "lr": 0.0003, "grad_norm": 0.13424698124505394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5794733535, "tokens_per_second": 148580.15477793117, "data_wait_s": 1.2144999345764518e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 39700, "loss": 0.650486072525382, "lr": 0.0003, "grad_norm": 0.1258898488456938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5809464754, "tokens_per_second": 149928.77250968918, "data_wait_s": 9.103998309001327e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 39800, "loss": 0.6376858409494162, "lr": 0.0003, "grad_norm": 0.12891983290411593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5824231971, "tokens_per_second": 150190.43901381618, "data_wait_s": 9.909002983476967e-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:d380e5ebe02916ce24b0d8af4833e6644082da0f606b1ee78beac13113546731
-size 840718
+oid sha256:205688ce5e4c099c2ddefe0e294dbe3ae81344ca462096b90a3c99c9dabf9f0c
+size 842836