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": 75213.28868775343,
-      "required_pages_per_second_upper_estimate": 0.006519297237117079
+      "required_token_positions_per_second": 75534.8358593945,
+      "required_pages_per_second_upper_estimate": 0.006547168131001071
     },
     {
       "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": 9401.66108596918,
-      "required_pages_per_second_upper_estimate": 0.0037504631745528877
+      "required_token_positions_per_second": 9441.854482424313,
+      "required_pages_per_second_upper_estimate": 0.0037664969213436704
     },
     {
       "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": 18803.32217193836,
-      "required_pages_per_second_upper_estimate": 0.004349817657310279
+      "required_token_positions_per_second": 18883.708964848625,
+      "required_pages_per_second_upper_estimate": 0.004368413727090834
     },
     {
       "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": 28204.983257907537,
-      "required_pages_per_second_upper_estimate": 0.0017175355002959204
+      "required_token_positions_per_second": 28325.563447272936,
+      "required_pages_per_second_upper_estimate": 0.0017248782011929122
     },
     {
       "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": 28204.983257907537,
-      "required_pages_per_second_upper_estimate": 0.0021712976875844915
+      "required_token_positions_per_second": 28325.563447272936,
+      "required_pages_per_second_upper_estimate": 0.0021805802843491676
     },
     {
       "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": 28204.983257907537,
-      "required_pages_per_second_upper_estimate": 0.35021149605656454
+      "required_token_positions_per_second": 28325.563447272936,
+      "required_pages_per_second_upper_estimate": 0.351708698452549
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0828051415781772,
-    "producer_p95_seconds": 0.22265558099752525,
-    "gpu_update_seconds": 0.980252310280972,
-    "producer_batches_per_second": 12.07654477658117,
-    "gpu_batches_per_second": 1.020145517140753,
-    "producer_headroom": 11.838060917455296,
+    "producer_batch_seconds": 0.08020570947653027,
+    "producer_p95_seconds": 0.233655551000993,
+    "gpu_update_seconds": 0.9760794362119505,
+    "producer_batches_per_second": 12.467940331512423,
+    "gpu_batches_per_second": 1.0245067797769436,
+    "producer_headroom": 12.169700169506886,
     "sustainable": true,
-    "buffer_seconds": 60.77564323742027,
+    "buffer_seconds": 60.51692504514093,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326951.81007084047,
-    "upload_bytes_per_second": 6380614.468422342,
+    "generation_bytes_per_second": 355798.2132041172,
+    "upload_bytes_per_second": 6285581.277255736,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.15624739259634,
+    "estimated_full_shards_per_hour": 76.34601399510038,
     "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": 1.1833946610004205,
-      "bytes": 10020680,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 0.9556004539990681,
       "bytes": 6956463,
@@ -562,6 +556,12 @@
       "bytes": 13273679,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.7825120570014406,
+      "bytes": 1093241,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -178,3 +178,4 @@
 {"step": 57800, "loss": 0.5990044809877872, "lr": 0.0003, "grad_norm": 0.13417589388506815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8459756888, "tokens_per_second": 147107.63316854305, "data_wait_s": 1.052000152412802e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57900, "loss": 0.6642866209149361, "lr": 0.0003, "grad_norm": 0.14105749521211802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8474263964, "tokens_per_second": 148700.0902675301, "data_wait_s": 1.85349999810569e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 58000, "loss": 0.6079016998410225, "lr": 0.0003, "grad_norm": 0.130945887154781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8488896618, "tokens_per_second": 149982.79334273018, "data_wait_s": 1.3979999494040385e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 58100, "loss": 0.604543486610055, "lr": 0.0003, "grad_norm": 0.13388629203116947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8503480705, "tokens_per_second": 148663.97225649076, "data_wait_s": 1.293099921895191e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b4cc9637abea16ac85e428f2fea3b3b676125b56541787a3ba4fb9ccc3370b6
-size 383754
+oid sha256:78e408d54e324e127298b3c26bac303dd4d61571fc68d29cf73bdf5454849ec7
+size 385872