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": 75838.99315849383,
-      "required_pages_per_second_upper_estimate": 0.006573531714807374
+      "required_token_positions_per_second": 75341.38095942288,
+      "required_pages_per_second_upper_estimate": 0.006530399950578493
     },
     {
       "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": 9479.874144811729,
-      "required_pages_per_second_upper_estimate": 0.00378166353311462
+      "required_token_positions_per_second": 9417.67261992786,
+      "required_pages_per_second_upper_estimate": 0.0037568504148427716
     },
     {
       "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": 18959.748289623458,
-      "required_pages_per_second_upper_estimate": 0.004386004086631082
+      "required_token_positions_per_second": 18835.34523985572,
+      "required_pages_per_second_upper_estimate": 0.004357225630486215
     },
     {
       "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": 28439.62243443519,
-      "required_pages_per_second_upper_estimate": 0.0017318237950898362
+      "required_token_positions_per_second": 28253.01785978358,
+      "required_pages_per_second_upper_estimate": 0.0017204605555320924
     },
     {
       "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": 28439.62243443519,
-      "required_pages_per_second_upper_estimate": 0.002189360860916402
+      "required_token_positions_per_second": 28253.01785978358,
+      "required_pages_per_second_upper_estimate": 0.002174995524205198
     },
     {
       "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": 28439.62243443519,
-      "required_pages_per_second_upper_estimate": 0.35312492934223016
+      "required_token_positions_per_second": 28253.01785978358,
+      "required_pages_per_second_upper_estimate": 0.350807925050394
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1054770370468674,
-    "producer_p95_seconds": 0.17982051000035426,
-    "gpu_update_seconds": 0.9721648050616636,
-    "producer_batches_per_second": 9.480736547004659,
-    "gpu_batches_per_second": 1.0286321771714118,
-    "producer_headroom": 9.216838397059773,
+    "producer_batch_seconds": 0.1219927291171814,
+    "producer_p95_seconds": 0.19669981300012296,
+    "gpu_update_seconds": 0.9785857262121089,
+    "producer_batches_per_second": 8.197209843870608,
+    "gpu_batches_per_second": 1.0218828797664778,
+    "producer_headroom": 8.021672547977168,
     "sustainable": true,
-    "buffer_seconds": 60.27421791382314,
+    "buffer_seconds": 60.67231502515075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409428.52328315313,
-    "upload_bytes_per_second": 5341890.859447985,
+    "generation_bytes_per_second": 409850.8988410081,
+    "upload_bytes_per_second": 5415528.205806394,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.85383008833833,
+    "estimated_full_shards_per_hour": 87.94446205065424,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 4951926,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.3130787809996036,
+      "bytes": 11524150,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -61,3 +61,4 @@
 {"step": 321100, "loss": 0.6374973971396685, "lr": 0.0003, "grad_norm": 0.19649614827835046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47000203249, "tokens_per_second": 150947.57448954, "data_wait_s": 1.3704000593861565e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 321200, "loss": 0.6152104809880257, "lr": 0.0003, "grad_norm": 0.15699181575060958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47014798356, "tokens_per_second": 148378.56027400636, "data_wait_s": 1.3146000128472224e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 321300, "loss": 0.5986273214221001, "lr": 0.0003, "grad_norm": 0.16807234766414308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47029273715, "tokens_per_second": 147159.05066639223, "data_wait_s": 1.2904999493912328e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 321400, "loss": 0.5764908324927092, "lr": 0.0003, "grad_norm": 0.1483505579466562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47043747942, "tokens_per_second": 147907.25459283905, "data_wait_s": 1.2944000445713755e-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 sha256:3ff68d5665a0a84cd0e5bd7c3d132bdb6b190deba3549e035c0a8021a0b2aba3
-size 134382
+oid sha256:66bbc2a16cfda64437a4beed1e8203e332b6a0ce57e45245a595420d0f7a96c5
+size 136500