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": 76998.08690004592,
-      "required_pages_per_second_upper_estimate": 0.0053269063119213535
+      "required_token_positions_per_second": 77345.85708109349,
+      "required_pages_per_second_upper_estimate": 0.0053509658599842184
     },
     {
       "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": 9624.76086250574,
-      "required_pages_per_second_upper_estimate": 0.0040296863852692525
+      "required_token_positions_per_second": 9668.232135136686,
+      "required_pages_per_second_upper_estimate": 0.00404788689933643
     },
     {
       "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": 19249.52172501148,
-      "required_pages_per_second_upper_estimate": 0.004009167731294421
+      "required_token_positions_per_second": 19336.464270273373,
+      "required_pages_per_second_upper_estimate": 0.004027275570643368
     },
     {
       "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": 28874.282587517224,
-      "required_pages_per_second_upper_estimate": 0.001730969152644952
+      "required_token_positions_per_second": 29004.696405410057,
+      "required_pages_per_second_upper_estimate": 0.0017387872618972575
     },
     {
       "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": 28874.282587517224,
-      "required_pages_per_second_upper_estimate": 0.002244848466632616
+      "required_token_positions_per_second": 29004.696405410057,
+      "required_pages_per_second_upper_estimate": 0.002254987567344022
     },
     {
       "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": 28874.282587517224,
-      "required_pages_per_second_upper_estimate": 0.37036829424349643
+      "required_token_positions_per_second": 29004.696405410057,
+      "required_pages_per_second_upper_estimate": 0.37204110267197776
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10380318758586782,
-    "producer_p95_seconds": 0.308798060999834,
-    "gpu_update_seconds": 0.957530283781064,
-    "producer_batches_per_second": 9.633615530089406,
-    "gpu_batches_per_second": 1.0443533921989736,
-    "producer_headroom": 9.224478612364173,
+    "producer_batch_seconds": 0.13761712600754095,
+    "producer_p95_seconds": 0.31354038800054695,
+    "gpu_update_seconds": 0.9532249403183892,
+    "producer_batches_per_second": 7.266537450761786,
+    "gpu_batches_per_second": 1.0490703271632689,
+    "producer_headroom": 6.926644727823743,
     "sustainable": true,
-    "buffer_seconds": 60.32440787820703,
+    "buffer_seconds": 60.05317124005852,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332916.4099656605,
-    "upload_bytes_per_second": 5423327.712331371,
+    "generation_bytes_per_second": 319507.01113052515,
+    "upload_bytes_per_second": 5352930.793357215,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.43611168124544,
+    "estimated_full_shards_per_hour": 68.55876684605423,
     "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": 1.1678417209986947,
-      "bytes": 10292515,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9083037529999274,
       "bytes": 719591,
@@ -562,6 +556,12 @@
       "bytes": 6125022,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4454559629994037,
+      "bytes": 6189473,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -321,3 +321,4 @@
 {"step": 636500, "loss": 0.5357717275619507, "lr": 0.0001, "grad_norm": 0.1773941165840124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93166979673, "tokens_per_second": 150792.34890343397, "data_wait_s": 1.7664002371020615e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 636600, "loss": 0.5652539134025574, "lr": 0.0001, "grad_norm": 0.1808153497684379, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93181546916, "tokens_per_second": 150857.92081038535, "data_wait_s": 1.7282000044360757e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 636700, "loss": 0.5259715914726257, "lr": 0.0001, "grad_norm": 0.17011995912858752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93196050485, "tokens_per_second": 151656.2617979774, "data_wait_s": 1.265399987460114e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 636800, "loss": 0.5594694018363953, "lr": 0.0001, "grad_norm": 0.16506547113416456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93210708814, "tokens_per_second": 152369.67231532015, "data_wait_s": 9.011000656755641e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8105a6e95c298978a6f37c465a9f28478ee68373c2b2a11d90acbc51500c04f
-size 689760
+oid sha256:9ff27ec805a4eeb6515b1bc932a791558bb85a725dbb1105bd4168e8cb5064b2
+size 691878