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": 76050.5832987335,
-      "required_pages_per_second_upper_estimate": 0.0065918718118930415
+      "required_token_positions_per_second": 75758.29940357714,
+      "required_pages_per_second_upper_estimate": 0.006566537384647648
     },
     {
       "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": 9506.322912341688,
-      "required_pages_per_second_upper_estimate": 0.0037922143419266345
+      "required_token_positions_per_second": 9469.787425447143,
+      "required_pages_per_second_upper_estimate": 0.0037776397899501927
     },
     {
       "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": 19012.645824683375,
-      "required_pages_per_second_upper_estimate": 0.004398241000402374
+      "required_token_positions_per_second": 18939.574850894285,
+      "required_pages_per_second_upper_estimate": 0.004381337316621484
     },
     {
       "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": 28518.96873702506,
-      "required_pages_per_second_upper_estimate": 0.001736655568619683
+      "required_token_positions_per_second": 28409.362276341428,
+      "required_pages_per_second_upper_estimate": 0.0017299811102247057
     },
     {
       "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": 28518.96873702506,
-      "required_pages_per_second_upper_estimate": 0.0021954691589343937
+      "required_token_positions_per_second": 28409.362276341428,
+      "required_pages_per_second_upper_estimate": 0.0021870313501808655
     },
     {
       "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": 28518.96873702506,
-      "required_pages_per_second_upper_estimate": 0.35411014486540426
+      "required_token_positions_per_second": 28409.362276341428,
+      "required_pages_per_second_upper_estimate": 0.35274919945294
     }
   ],
   "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": 0.07525077460159935,
-    "producer_p95_seconds": 0.20565995700053463,
-    "gpu_update_seconds": 0.9694600199237108,
-    "producer_batches_per_second": 13.288899752784026,
-    "gpu_batches_per_second": 1.0315020521204088,
-    "producer_headroom": 12.883057019098198,
+    "producer_batch_seconds": 0.15147967911720173,
+    "producer_p95_seconds": 0.2591939209996781,
+    "gpu_update_seconds": 0.9732003038668886,
+    "producer_batches_per_second": 6.601545539493039,
+    "gpu_batches_per_second": 1.0275376980736917,
+    "producer_headroom": 6.424626125025729,
     "sustainable": true,
-    "buffer_seconds": 60.106521235270066,
+    "buffer_seconds": 60.33841883974709,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439646.60693551594,
-    "upload_bytes_per_second": 4771007.009329749,
+    "generation_bytes_per_second": 443595.31889399345,
+    "upload_bytes_per_second": 4655238.699287984,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.33792739914998,
+    "estimated_full_shards_per_hour": 95.18522906412937,
     "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": 1.2944532310002614,
-      "bytes": 11460253,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.090632098000242,
       "bytes": 1122627,
@@ -562,6 +556,12 @@
       "bytes": 7164779,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2709880140009773,
+      "bytes": 1216081,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 250400, "loss": 0.6021800842136145, "lr": 0.0003, "grad_norm": 0.15040273044040206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36652199404, "tokens_per_second": 150873.33745707825, "data_wait_s": 1.5345000065281056e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250500, "loss": 0.5950481034815311, "lr": 0.0003, "grad_norm": 0.13582030029360265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36666699586, "tokens_per_second": 148404.6526536854, "data_wait_s": 1.2508999134297483e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250600, "loss": 0.595927158370614, "lr": 0.0003, "grad_norm": 0.15122073928134755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36681307733, "tokens_per_second": 149336.46746914863, "data_wait_s": 1.5302000974770635e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 250700, "loss": 0.5758683197200298, "lr": 0.0003, "grad_norm": 0.14111654771792698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36695727555, "tokens_per_second": 147699.5653436398, "data_wait_s": 1.1117001122329384e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1380589ef26e0ed56d1dfbb81e3d5550f9177e973c32244426633a3a5f2795d3
-size 288444
+oid sha256:9b19515bedf2fd12351a0b99b171a2a7bdb0eea98f3de0b663a422f2f10fe157
+size 290562