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": 72854.55634173428,
-      "required_pages_per_second_upper_estimate": 0.006314848295516615
+      "required_token_positions_per_second": 74790.6535422747,
+      "required_pages_per_second_upper_estimate": 0.006482664293865965
     },
     {
       "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": 9106.819542716785,
-      "required_pages_per_second_upper_estimate": 0.0036328464746755965
+      "required_token_positions_per_second": 9348.831692784337,
+      "required_pages_per_second_upper_estimate": 0.0037293887397416375
     },
     {
       "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": 18213.63908543357,
-      "required_pages_per_second_upper_estimate": 0.0042134048533152635
+      "required_token_positions_per_second": 18697.663385568674,
+      "required_pages_per_second_upper_estimate": 0.004325375356614782
     },
     {
       "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": 27320.45862815035,
-      "required_pages_per_second_upper_estimate": 0.0016636725910857846
+      "required_token_positions_per_second": 28046.49507835301,
+      "required_pages_per_second_upper_estimate": 0.0017078844016842642
     },
     {
       "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": 27320.45862815035,
-      "required_pages_per_second_upper_estimate": 0.0021032045330648956
+      "required_token_positions_per_second": 28046.49507835301,
+      "required_pages_per_second_upper_estimate": 0.0021590968287990234
     },
     {
       "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": 27320.45862815035,
-      "required_pages_per_second_upper_estimate": 0.33922866046848466
+      "required_token_positions_per_second": 28046.49507835301,
+      "required_pages_per_second_upper_estimate": 0.34824360329231296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.12150057469520448,
-    "producer_p95_seconds": 0.27700270599962096,
-    "gpu_update_seconds": 1.011988867987456,
-    "producer_batches_per_second": 8.230413745026254,
-    "gpu_batches_per_second": 0.988153162187151,
-    "producer_headroom": 8.329087088897518,
+    "producer_batch_seconds": 0.09693332501569785,
+    "producer_p95_seconds": 0.2622789079978247,
+    "gpu_update_seconds": 0.9857916264674162,
+    "producer_batches_per_second": 10.31636952346425,
+    "gpu_batches_per_second": 1.0144131611094116,
+    "producer_headroom": 10.169790691774706,
     "sustainable": true,
-    "buffer_seconds": 60.71933207924736,
+    "buffer_seconds": 60.13328921451239,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421328.90566952585,
-    "upload_bytes_per_second": 5638844.582235521,
+    "generation_bytes_per_second": 438695.2353276823,
+    "upload_bytes_per_second": 5485509.194508932,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.40737512173015,
+    "estimated_full_shards_per_hour": 94.13378519890644,
     "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": [
-    {
-      "seconds": 0.9809586890005448,
-      "bytes": 9894024,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0132985079999344,
       "bytes": 6275693,
@@ -562,6 +556,12 @@
       "bytes": 771728,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 2.6395804720013984,
+      "bytes": 7630392,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -201,3 +201,4 @@
 {"step": 335100, "loss": 0.5736087337136269, "lr": 0.0003, "grad_norm": 0.17493034646593886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49050095421, "tokens_per_second": 131968.2329283222, "data_wait_s": 1.3155000488040969e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 335200, "loss": 0.6403679437935352, "lr": 0.0003, "grad_norm": 0.15703487421252635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49064787693, "tokens_per_second": 149277.68798784644, "data_wait_s": 9.128001693170518e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 335300, "loss": 0.5784935504198074, "lr": 0.0003, "grad_norm": 0.2093674290392367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49079527448, "tokens_per_second": 148808.59507893154, "data_wait_s": 9.47700027609244e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 335400, "loss": 0.600412305444479, "lr": 0.0003, "grad_norm": 0.1508263100708454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49094194312, "tokens_per_second": 148633.22685352326, "data_wait_s": 9.461997251491994e-06, "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 sha256:1d4f436726feaf1d35eddb8a67210cf137aab7dfccc351a2d7bc7f0b37412102
-size 433251
+oid sha256:0f8d95140684b81a483d195e602401222eabaa3776bc877b417288758ef7d73a
+size 435369