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": 74906.53301748993,
-      "required_pages_per_second_upper_estimate": 0.006492708433083777
+      "required_token_positions_per_second": 74998.94668657484,
+      "required_pages_per_second_upper_estimate": 0.006500718615699754
     },
     {
       "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": 9363.316627186241,
-      "required_pages_per_second_upper_estimate": 0.0037351669966436255
+      "required_token_positions_per_second": 9374.868335821855,
+      "required_pages_per_second_upper_estimate": 0.0037397751459318073
     },
     {
       "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": 18726.633254372482,
-      "required_pages_per_second_upper_estimate": 0.004332077025910265
+      "required_token_positions_per_second": 18749.73667164371,
+      "required_pages_per_second_upper_estimate": 0.004337421594889708
     },
     {
       "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": 28089.94988155872,
-      "required_pages_per_second_upper_estimate": 0.0017105305712097597
+      "required_token_positions_per_second": 28124.605007465565,
+      "required_pages_per_second_upper_estimate": 0.0017126408865560927
     },
     {
       "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": 28089.94988155872,
-      "required_pages_per_second_upper_estimate": 0.0021624420998403934
+      "required_token_positions_per_second": 28124.605007465565,
+      "required_pages_per_second_upper_estimate": 0.0021651099473642306
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28089.94988155872,
-      "required_pages_per_second_upper_estimate": 0.34878316651425706
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28124.605007465565,
+      "required_pages_per_second_upper_estimate": 0.34921346719477464
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11322434253145275,
-    "producer_p95_seconds": 0.3297246180009097,
-    "gpu_update_seconds": 0.9842666190782752,
-    "producer_batches_per_second": 8.832023022983849,
-    "gpu_batches_per_second": 1.0159848770818403,
-    "producer_headroom": 8.6930654404538,
+    "producer_batch_seconds": 0.10296336389077965,
+    "producer_p95_seconds": 0.32298557399917627,
+    "gpu_update_seconds": 0.9830538061836228,
+    "producer_batches_per_second": 9.712192397489742,
+    "gpu_batches_per_second": 1.0172383176890034,
+    "producer_headroom": 9.547607702739937,
     "sustainable": true,
-    "buffer_seconds": 60.04026376377479,
+    "buffer_seconds": 60.949335983384614,
     "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": 421008.8963510563,
+    "generation_bytes_per_second": 422269.92770907824,
     "upload_bytes_per_second": 4130836.3573462437,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.33870857142226,
+    "estimated_full_shards_per_hour": 90.60929654554616,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 109200, "loss": 0.573639590293169, "lr": 0.0003, "grad_norm": 0.13428051258734908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15979730827, "tokens_per_second": 148993.70073355455, "data_wait_s": 1.369700112263672e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109300, "loss": 0.6230886615812778, "lr": 0.0003, "grad_norm": 0.14055154783394633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15994395784, "tokens_per_second": 148527.6360667483, "data_wait_s": 1.4556004316546023e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109400, "loss": 0.5503679495304823, "lr": 0.0003, "grad_norm": 0.13035026920431542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16009047196, "tokens_per_second": 148694.1929978418, "data_wait_s": 1.4409997675102204e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 109500, "loss": 0.5849812235683203, "lr": 0.0003, "grad_norm": 0.13039552184598213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16023487806, "tokens_per_second": 146737.0379731941, "data_wait_s": 1.657399843679741e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17f805c54da43d41a690389bf5b0d91282bf285d534fa641178aa75b1a60f09b
-size 520089
+oid sha256:2d2b0fcc4e499aca03266e27af341d95feb7497522e5df755952deaec483994f
+size 522207