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": 75669.58504882465,
-      "required_pages_per_second_upper_estimate": 0.005234997471304879
+      "required_token_positions_per_second": 76457.95864825443,
+      "required_pages_per_second_upper_estimate": 0.005289538986192207
     },
     {
       "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": 9458.698131103081,
-      "required_pages_per_second_upper_estimate": 0.003960159387415126
+      "required_token_positions_per_second": 9557.244831031803,
+      "required_pages_per_second_upper_estimate": 0.0040014188327861765
     },
     {
       "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": 18917.396262206163,
-      "required_pages_per_second_upper_estimate": 0.003939994756129527
+      "required_token_positions_per_second": 19114.489662063606,
+      "required_pages_per_second_upper_estimate": 0.003981044113617348
     },
     {
       "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": 28376.094393309246,
-      "required_pages_per_second_upper_estimate": 0.0017011035310915109
+      "required_token_positions_per_second": 28671.73449309541,
+      "required_pages_per_second_upper_estimate": 0.001718826703657397
     },
     {
       "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": 28376.094393309246,
-      "required_pages_per_second_upper_estimate": 0.0022061165258312285
+      "required_token_positions_per_second": 28671.73449309541,
+      "required_pages_per_second_upper_estimate": 0.0022291012432062323
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28376.094393309246,
-      "required_pages_per_second_upper_estimate": 0.36397807100100366
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28671.73449309541,
+      "required_pages_per_second_upper_estimate": 0.3677702247674531
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.5250360160703025,
-    "producer_p95_seconds": 0.19251542500023788,
-    "gpu_update_seconds": 0.9743412753278364,
-    "producer_batches_per_second": 1.9046312431757058,
-    "gpu_batches_per_second": 1.0263344326283725,
-    "producer_headroom": 1.8557608345050596,
+    "producer_batch_seconds": 0.0737259678592892,
+    "producer_p95_seconds": 0.19179887999962375,
+    "gpu_update_seconds": 0.9642946437948516,
+    "producer_batches_per_second": 13.563741908530316,
+    "gpu_batches_per_second": 1.0370274339227217,
+    "producer_headroom": 13.079443672211543,
     "sustainable": true,
-    "buffer_seconds": 60.409159070325856,
+    "buffer_seconds": 60.750562559075654,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311087.8943836899,
+    "generation_bytes_per_second": 310288.74049333367,
     "upload_bytes_per_second": 5006494.831519759,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.75222037919066,
+    "estimated_full_shards_per_hour": 66.58074055767067,
     "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": [

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

@@ -104,3 +104,4 @@
 {"step": 407500, "loss": 0.6548508405685425, "lr": 0.0003, "grad_norm": 0.19174734648015654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59647166839, "tokens_per_second": 153215.99214786285, "data_wait_s": 9.211998985847458e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 407600, "loss": 0.6006882190704346, "lr": 0.0003, "grad_norm": 0.16348153630773482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59661852973, "tokens_per_second": 144387.3074612499, "data_wait_s": 1.02319991128752e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 407700, "loss": 0.6368137001991272, "lr": 0.0003, "grad_norm": 0.15450960747937334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59676616688, "tokens_per_second": 153721.03336529367, "data_wait_s": 7.396000000881031e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 407800, "loss": 0.5638266205787659, "lr": 0.0003, "grad_norm": 0.1509126354770479, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59691261272, "tokens_per_second": 152007.03551981243, "data_wait_s": 1.0596999345580116e-05, "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 sha256:0e81f0da9a5385aa59d6a5f89deecadb93545691d9177522ff85b0ea19cfe767
-size 226239
+oid sha256:2fe6f45c9ef1434fcec8fb940665c4c2ed625723f5d40d948ba1922c3cac159d
+size 228357