Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76874.97278776098,
-      "required_pages_per_second_upper_estimate": 0.0053183889919693845
+      "required_token_positions_per_second": 77121.36225193013,
+      "required_pages_per_second_upper_estimate": 0.0053354347868029275
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9609.371598470123,
-      "required_pages_per_second_upper_estimate": 0.004023243221781917
+      "required_token_positions_per_second": 9640.170281491266,
+      "required_pages_per_second_upper_estimate": 0.004036137987213233
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19218.743196940246,
-      "required_pages_per_second_upper_estimate": 0.004002757375581551
+      "required_token_positions_per_second": 19280.340562982532,
+      "required_pages_per_second_upper_estimate": 0.004015586482496379
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28828.11479541037,
-      "required_pages_per_second_upper_estimate": 0.001728201464002283
+      "required_token_positions_per_second": 28920.5108444738,
+      "required_pages_per_second_upper_estimate": 0.0017337404660630374
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28828.11479541037,
-      "required_pages_per_second_upper_estimate": 0.002241259123866964
+      "required_token_positions_per_second": 28920.5108444738,
+      "required_pages_per_second_upper_estimate": 0.002248442510274377
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28828.11479541037,
-      "required_pages_per_second_upper_estimate": 0.3697761033774627
+      "required_token_positions_per_second": 28920.5108444738,
+      "required_pages_per_second_upper_estimate": 0.37096126068770024
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09166045025776981,
-    "producer_p95_seconds": 0.16821103999973275,
-    "gpu_update_seconds": 0.9590637541238648,
-    "producer_batches_per_second": 10.909830763298402,
-    "gpu_batches_per_second": 1.04268355018122,
-    "producer_headroom": 10.463223248704995,
+    "producer_batch_seconds": 0.07534616505463987,
+    "producer_p95_seconds": 0.1636205799995878,
+    "gpu_update_seconds": 0.9559997106787983,
+    "producer_batches_per_second": 13.272075616254861,
+    "gpu_batches_per_second": 1.0460254211687572,
+    "producer_headroom": 12.688100449246782,
     "sustainable": true,
-    "buffer_seconds": 60.42101650980348,
+    "buffer_seconds": 60.227981772764295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333321.80042543786,
-    "upload_bytes_per_second": 6044035.408153326,
+    "generation_bytes_per_second": 344484.21823540685,
+    "upload_bytes_per_second": 5939941.583462157,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.52309903690674,
+    "estimated_full_shards_per_hour": 73.91829405113845,
     "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": [
@@ -340,6 +340,12 @@
       "bytes": 11936639,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9754752660001031,
+      "bytes": 2583341,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 488600, "loss": 0.5770623087882996, "lr": 0.0003, "grad_norm": 0.16261208145739145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71514909085, "tokens_per_second": 152146.73027920883, "data_wait_s": 1.575599981151754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488700, "loss": 0.5719287991523743, "lr": 0.0003, "grad_norm": 0.15512963706693148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71529513012, "tokens_per_second": 152308.80070706914, "data_wait_s": 1.0300999747414608e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488800, "loss": 0.5863213539123535, "lr": 0.0003, "grad_norm": 0.15694707538758185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71544182605, "tokens_per_second": 153333.76972484827, "data_wait_s": 1.2375000551401172e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 488900, "loss": 0.6243778467178345, "lr": 0.0003, "grad_norm": 0.18731338743514026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71558886946, "tokens_per_second": 152825.46945399616, "data_wait_s": 7.768000614305492e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d464748ca007d95044a7f132936c9325fd92e4687841e5a804877e383e5a5daa
-size 80649
+oid sha256:e129ea8f88b5cf0ed519ac5a28cc5f6daf54ea5d593a5053afe4fe73174974ad
+size 82767