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": 73651.7859130299,
-      "required_pages_per_second_upper_estimate": 0.0063839501341966445
+      "required_token_positions_per_second": 74086.4868488775,
+      "required_pages_per_second_upper_estimate": 0.006421628909576464
     },
     {
       "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": 9206.473239128738,
-      "required_pages_per_second_upper_estimate": 0.003672599824129862
+      "required_token_positions_per_second": 9260.810856109687,
+      "required_pages_per_second_upper_estimate": 0.0036942759119633346
     },
     {
       "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": 18412.946478257476,
-      "required_pages_per_second_upper_estimate": 0.004259511110954763
+      "required_token_positions_per_second": 18521.621712219374,
+      "required_pages_per_second_upper_estimate": 0.004284651213713055
     },
     {
       "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": 27619.419717386216,
-      "required_pages_per_second_upper_estimate": 0.0016818777528926356
+      "required_token_positions_per_second": 27782.43256832906,
+      "required_pages_per_second_upper_estimate": 0.0016918043802527242
     },
     {
       "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": 27619.419717386216,
-      "required_pages_per_second_upper_estimate": 0.002126219385291535
+      "required_token_positions_per_second": 27782.43256832906,
+      "required_pages_per_second_upper_estimate": 0.0021387685658055674
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27619.419717386216,
-      "required_pages_per_second_upper_estimate": 0.34294075663839246
+      "required_token_positions_per_second": 27782.43256832906,
+      "required_pages_per_second_upper_estimate": 0.3449648306781859
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.7940945405781292,
-    "producer_p95_seconds": 0.16886102900025435,
-    "gpu_update_seconds": 1.0010347893947893,
-    "producer_batches_per_second": 1.2592959010547589,
-    "gpu_batches_per_second": 0.9989662802874065,
-    "producer_headroom": 1.260599007098072,
+    "producer_batch_seconds": 0.09215708364845554,
+    "producer_p95_seconds": 0.16051031899996815,
+    "gpu_update_seconds": 0.9951612383833406,
+    "producer_batches_per_second": 10.85103781945425,
+    "gpu_batches_per_second": 1.0048622890744017,
+    "producer_headroom": 10.798532234152557,
     "sustainable": true,
-    "buffer_seconds": 60.062087363687354,
+    "buffer_seconds": 60.70483554138377,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 257587.39132622755,
-    "upload_bytes_per_second": 4761158.233335389,
+    "generation_bytes_per_second": 269815.4975116403,
+    "upload_bytes_per_second": 4917752.353212473,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.272257851029586,
+    "estimated_full_shards_per_hour": 57.89612478267581,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 6981450,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.187183233000269,
+      "bytes": 9995831,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 317500, "loss": 0.536181753501296, "lr": 0.0003, "grad_norm": 0.1482592861132939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46473593505, "tokens_per_second": 148694.15235386262, "data_wait_s": 9.55299992710934e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 317600, "loss": 0.5978003963828087, "lr": 0.0003, "grad_norm": 0.165373911410232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46488188880, "tokens_per_second": 149238.6228716883, "data_wait_s": 1.0754999948403565e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317700, "loss": 0.5958874635398388, "lr": 0.0003, "grad_norm": 0.15210044562063937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46502784833, "tokens_per_second": 143218.58467287413, "data_wait_s": 1.0818999726325274e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 317800, "loss": 0.6032504867762327, "lr": 0.0003, "grad_norm": 0.16004447989947035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46517260924, "tokens_per_second": 146172.61111936995, "data_wait_s": 1.1180000001331791e-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 sha256:4560f2b82e1207ac6a4ea1b62fdd55580c42a565a0d0613d81b8b1d8771e3126
-size 57351
+oid sha256:8e0da94bfafa7f7a287e31362c0894a9a5edbd3038e3a76655f3f345d288942e
+size 59469