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": 76682.40247189176,
-      "required_pages_per_second_upper_estimate": 0.005305066530692858
+      "required_token_positions_per_second": 76743.33841686371,
+      "required_pages_per_second_upper_estimate": 0.0053092822207568904
     },
     {
       "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": 9585.30030898647,
-      "required_pages_per_second_upper_estimate": 0.004013165075540796
+      "required_token_positions_per_second": 9592.917302107964,
+      "required_pages_per_second_upper_estimate": 0.004016354151499861
     },
     {
       "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": 19170.60061797294,
-      "required_pages_per_second_upper_estimate": 0.003992730545987846
+      "required_token_positions_per_second": 19185.83460421593,
+      "required_pages_per_second_upper_estimate": 0.003995903383575027
     },
     {
       "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": 28755.90092695941,
-      "required_pages_per_second_upper_estimate": 0.0017238723528528405
+      "required_token_positions_per_second": 28778.75190632389,
+      "required_pages_per_second_upper_estimate": 0.0017252422341743159
     },
     {
       "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": 28755.90092695941,
-      "required_pages_per_second_upper_estimate": 0.002235644813230141
+      "required_token_positions_per_second": 28778.75190632389,
+      "required_pages_per_second_upper_estimate": 0.002237421376364893
     },
     {
       "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": 28755.90092695941,
-      "required_pages_per_second_upper_estimate": 0.3688498214101847
+      "required_token_positions_per_second": 28778.75190632389,
+      "required_pages_per_second_upper_estimate": 0.36914292923800224
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13995438833586604,
-    "producer_p95_seconds": 0.35195301200292306,
-    "gpu_update_seconds": 0.9614722233960431,
-    "producer_batches_per_second": 7.145185026997331,
-    "gpu_batches_per_second": 1.040071648110511,
-    "producer_headroom": 6.869896934483241,
+    "producer_batch_seconds": 0.4968400585550512,
+    "producer_p95_seconds": 0.3174719439994078,
+    "gpu_update_seconds": 0.9607087927230292,
+    "producer_batches_per_second": 2.012720155673996,
+    "gpu_batches_per_second": 1.0408981447599785,
+    "producer_headroom": 1.9336379508468722,
     "sustainable": true,
-    "buffer_seconds": 60.572750073950715,
+    "buffer_seconds": 60.52465394155084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361231.98358110926,
-    "upload_bytes_per_second": 5347905.428815578,
+    "generation_bytes_per_second": 332091.84742007346,
+    "upload_bytes_per_second": 5350020.788125196,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.51197462630232,
+    "estimated_full_shards_per_hour": 71.25917975379613,
     "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": 1.1640828560011869,
-      "bytes": 6135465,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.8637312579994614,
       "bytes": 2324435,
@@ -562,6 +556,12 @@
       "bytes": 2441375,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.106896608005627,
+      "bytes": 5997924,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 430600, "loss": 0.6422554850578308, "lr": 0.0003, "grad_norm": 0.1645814127129983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63028317997, "tokens_per_second": 152778.01857043232, "data_wait_s": 1.6288002370856702e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430700, "loss": 0.5648614168167114, "lr": 0.0003, "grad_norm": 0.1504450733490528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63042805057, "tokens_per_second": 149627.51538204597, "data_wait_s": 1.2897995475213975e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430800, "loss": 0.5561989545822144, "lr": 0.0003, "grad_norm": 0.1542304801022447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63057327012, "tokens_per_second": 151158.72180449034, "data_wait_s": 1.9488994439598173e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 430900, "loss": 0.5481613278388977, "lr": 0.0003, "grad_norm": 0.1654212890461418, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63071904651, "tokens_per_second": 151206.1293694662, "data_wait_s": 9.268005669582635e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e851042218c33861fe82646e7b51f16978d9ebded7fd1ae42e4ff29076bda8f9
-size 719412
+oid sha256:71935276b0bfe90d2cde96d209a52c604fd41714f1579b0078f350c9d8cf13bc
+size 721530