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": 76569.64578729637,
-      "required_pages_per_second_upper_estimate": 0.005297265761620994
+      "required_token_positions_per_second": 75506.03497093178,
+      "required_pages_per_second_upper_estimate": 0.005223682697427786
     },
     {
       "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": 9571.205723412046,
-      "required_pages_per_second_upper_estimate": 0.004007263966889199
+      "required_token_positions_per_second": 9438.254371366473,
+      "required_pages_per_second_upper_estimate": 0.003951600012127657
     },
     {
       "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": 19142.41144682409,
-      "required_pages_per_second_upper_estimate": 0.0039868594850359755
+      "required_token_positions_per_second": 18876.508742732945,
+      "required_pages_per_second_upper_estimate": 0.003931478964099655
     },
     {
       "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": 28713.61717023614,
-      "required_pages_per_second_upper_estimate": 0.0017213375062008385
+      "required_token_positions_per_second": 28314.763114099416,
+      "required_pages_per_second_upper_estimate": 0.0016974268145503227
     },
     {
       "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": 28713.61717023614,
-      "required_pages_per_second_upper_estimate": 0.0022323574371314967
+      "required_token_positions_per_second": 28314.763114099416,
+      "required_pages_per_second_upper_estimate": 0.0022013482886404515
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28713.61717023614,
-      "required_pages_per_second_upper_estimate": 0.36830745077969934
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28314.763114099416,
+      "required_pages_per_second_upper_estimate": 0.36319137920369693
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12144105311705289,
-    "producer_p95_seconds": 0.34899410100479145,
-    "gpu_update_seconds": 0.9628880902075712,
-    "producer_batches_per_second": 8.234447695674493,
-    "gpu_batches_per_second": 1.0385422876966197,
-    "producer_headroom": 7.928851615602149,
+    "producer_batch_seconds": 0.14691978431255848,
+    "producer_p95_seconds": 0.35562902899982873,
+    "gpu_update_seconds": 0.9764517502261073,
+    "producer_batches_per_second": 6.806435257709002,
+    "gpu_batches_per_second": 1.0241161427263967,
+    "producer_headroom": 6.646155620190641,
     "sustainable": true,
-    "buffer_seconds": 60.66194968307698,
+    "buffer_seconds": 60.54000851401865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380521.49067997973,
+    "generation_bytes_per_second": 461327.19174209615,
     "upload_bytes_per_second": 5356483.757815193,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.6510538129763,
+    "estimated_full_shards_per_hour": 98.99007620045818,
     "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

@@ -353,3 +353,4 @@
 {"step": 432400, "loss": 0.5923108458518982, "lr": 0.0003, "grad_norm": 0.1772879380050161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63291303295, "tokens_per_second": 152364.13310560162, "data_wait_s": 1.901299401652068e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432500, "loss": 0.5722652673721313, "lr": 0.0003, "grad_norm": 0.1537789846340482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63305966357, "tokens_per_second": 152138.76700763498, "data_wait_s": 1.7398000636603683e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432600, "loss": 0.5397249460220337, "lr": 0.0003, "grad_norm": 0.19410255482679223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63320561635, "tokens_per_second": 151767.36056835818, "data_wait_s": 2.1735999325755984e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 432700, "loss": 0.596798837184906, "lr": 0.0003, "grad_norm": 0.15001806380939847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63335096869, "tokens_per_second": 150830.0575688252, "data_wait_s": 1.727099879644811e-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 sha256:64c1577493e5b7c3c1e8e0c1ad4ec8d246fdf8a119c4ec7916234720c6d0f2c3
-size 757536
+oid sha256:d93e187e1294493453da96e8c8386cebc0c58c8e062fc3d59f67b01df5e79804
+size 759654