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": 75171.43605832849,
-      "required_pages_per_second_upper_estimate": 0.0065156695572730445
+      "required_token_positions_per_second": 75777.23166798048,
+      "required_pages_per_second_upper_estimate": 0.006568178385342743
     },
     {
       "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": 9396.42950729106,
-      "required_pages_per_second_upper_estimate": 0.003748376219599115
+      "required_token_positions_per_second": 9472.15395849756,
+      "required_pages_per_second_upper_estimate": 0.0037785838353668263
     },
     {
       "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": 18792.85901458212,
-      "required_pages_per_second_upper_estimate": 0.004347397190001194
+      "required_token_positions_per_second": 18944.30791699512,
+      "required_pages_per_second_upper_estimate": 0.004382432228164794
     },
     {
       "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": 28189.288521873183,
-      "required_pages_per_second_upper_estimate": 0.0017165797732153442
+      "required_token_positions_per_second": 28416.46187549268,
+      "required_pages_per_second_upper_estimate": 0.0017304134385642991
     },
     {
       "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": 28189.288521873183,
-      "required_pages_per_second_upper_estimate": 0.002170089463358757
+      "required_token_positions_per_second": 28416.46187549268,
+      "required_pages_per_second_upper_estimate": 0.002187577896976477
     },
     {
       "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": 28189.288521873183,
-      "required_pages_per_second_upper_estimate": 0.3500166199619204
+      "required_token_positions_per_second": 28416.46187549268,
+      "required_pages_per_second_upper_estimate": 0.3528373527135687
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.26885645669534597,
-    "producer_p95_seconds": 0.3314132560008147,
-    "gpu_update_seconds": 0.9807980779134182,
-    "producer_batches_per_second": 3.719456888971603,
-    "gpu_batches_per_second": 1.0195778545237697,
-    "producer_headroom": 3.6480361675851705,
+    "producer_batch_seconds": 0.2666393610469697,
+    "producer_p95_seconds": 0.31981465599892545,
+    "gpu_update_seconds": 0.9729571584647057,
+    "producer_batches_per_second": 3.7503840245995996,
+    "gpu_batches_per_second": 1.0277944833439194,
+    "producer_headroom": 3.6489629837258533,
     "sustainable": true,
-    "buffer_seconds": 60.80948083063193,
+    "buffer_seconds": 60.32334382481175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387622.008966943,
-    "upload_bytes_per_second": 4084200.961932447,
+    "generation_bytes_per_second": 387669.2845871445,
+    "upload_bytes_per_second": 4119237.686600681,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.1746597457525,
+    "estimated_full_shards_per_hour": 83.18480399332763,
     "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": 1.215623644999141,
-      "bytes": 7281440,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.1079499599982228,
       "bytes": 10281570,
@@ -562,6 +556,12 @@
       "bytes": 7503350,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1287752040007035,
+      "bytes": 10564318,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -270,3 +270,4 @@
 {"step": 342000, "loss": 0.5790649708360434, "lr": 0.0003, "grad_norm": 0.1369144824027103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50059709616, "tokens_per_second": 145901.03387296785, "data_wait_s": 1.6900001355679706e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342100, "loss": 0.5834092274308205, "lr": 0.0003, "grad_norm": 0.1502677618109866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50074220963, "tokens_per_second": 148094.98289689445, "data_wait_s": 1.0193001799052581e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342200, "loss": 0.5539342202246189, "lr": 0.0003, "grad_norm": 0.1576419914149281, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50088846883, "tokens_per_second": 148006.31993308643, "data_wait_s": 1.4042001566849649e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 342300, "loss": 0.63454519957304, "lr": 0.0003, "grad_norm": 0.16010137369153826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50103505821, "tokens_per_second": 132307.1079772074, "data_wait_s": 1.2348999007372186e-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:bc7fac1603aeef9c6ef113e43ba1c3e78548f94417cd5651c80f20d832213e51
-size 580176
+oid sha256:e28bce42ee117ced7766c463ab0ad077801b3190b73b56c4c45450211920f027
+size 582294