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": 75313.37723911558,
-      "required_pages_per_second_upper_estimate": 0.006527972659077036
+      "required_token_positions_per_second": 73995.94003887287,
+      "required_pages_per_second_upper_estimate": 0.006413780541574055
     },
     {
       "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": 9414.172154889447,
-      "required_pages_per_second_upper_estimate": 0.0037554540270023327
+      "required_token_positions_per_second": 9249.492504859109,
+      "required_pages_per_second_upper_estimate": 0.0036897608524250475
     },
     {
       "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": 18828.344309778895,
-      "required_pages_per_second_upper_estimate": 0.004355606088525102
+      "required_token_positions_per_second": 18498.985009718217,
+      "required_pages_per_second_upper_estimate": 0.004279414610981769
     },
     {
       "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": 28242.516464668344,
-      "required_pages_per_second_upper_estimate": 0.0017198210756661364
+      "required_token_positions_per_second": 27748.477514577327,
+      "required_pages_per_second_upper_estimate": 0.0016897366956276426
     },
     {
       "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": 28242.516464668344,
-      "required_pages_per_second_upper_estimate": 0.0021741870977395046
+      "required_token_positions_per_second": 27748.477514577327,
+      "required_pages_per_second_upper_estimate": 0.0021361546117741547
     },
     {
       "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": 28242.516464668344,
-      "required_pages_per_second_upper_estimate": 0.35067753286897135
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27748.477514577327,
+      "required_pages_per_second_upper_estimate": 0.3445432225508441
     }
   ],
   "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": 0.1069119173516242,
-    "producer_p95_seconds": 0.42255482300242875,
-    "gpu_update_seconds": 0.9789495930572586,
-    "producer_batches_per_second": 9.353494210669565,
-    "gpu_batches_per_second": 1.0215030550010251,
-    "producer_headroom": 9.156599351198395,
+    "producer_batch_seconds": 0.1674437009844496,
+    "producer_p95_seconds": 0.4507439489971148,
+    "gpu_update_seconds": 0.9963789899995579,
+    "producer_batches_per_second": 5.9721565763340925,
+    "gpu_batches_per_second": 1.0036341693640527,
+    "producer_headroom": 5.95053133764698,
     "sustainable": true,
-    "buffer_seconds": 60.69487476955003,
+    "buffer_seconds": 60.77911838997303,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293580.2974843921,
+    "generation_bytes_per_second": 310253.8469820631,
     "upload_bytes_per_second": 5381315.738470292,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.995497640598515,
+    "estimated_full_shards_per_hour": 66.57325322243136,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -411,3 +411,4 @@
 {"step": 356100, "loss": 0.5713396035134792, "lr": 0.0003, "grad_norm": 0.14518442927444247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52122825287, "tokens_per_second": 148726.7354203944, "data_wait_s": 1.1285999789834023e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356200, "loss": 0.5745569840073586, "lr": 0.0003, "grad_norm": 0.13863634003555855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52137511384, "tokens_per_second": 148395.937101354, "data_wait_s": 1.2736003554891795e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356300, "loss": 0.5990432072430849, "lr": 0.0003, "grad_norm": 0.15338934049619438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52152238141, "tokens_per_second": 147623.61823800617, "data_wait_s": 1.6524994862265885e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 356400, "loss": 0.6035524979233742, "lr": 0.0003, "grad_norm": 0.14347886779950328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52166934931, "tokens_per_second": 149216.09437987188, "data_wait_s": 1.5997000446077436e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c804eab5074e9f9e0911f5ce47934fedc2018f23eee4ae62afce9d2ef38f765
-size 881163
+oid sha256:1f353dea7b217b14d8893cf8a47f8fc0c442d3a01eaa8dce52b040f42ebafc72
+size 883281