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": 77122.3085353717,
-      "required_pages_per_second_upper_estimate": 0.005335500252887096
+      "required_token_positions_per_second": 76731.9993940171,
+      "required_pages_per_second_upper_estimate": 0.0053084977608462034
     },
     {
       "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": 9640.288566921463,
-      "required_pages_per_second_upper_estimate": 0.004036187510852775
+      "required_token_positions_per_second": 9591.499924252137,
+      "required_pages_per_second_upper_estimate": 0.004015760724989842
     },
     {
       "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": 19280.577133842926,
-      "required_pages_per_second_upper_estimate": 0.004015635753967806
+      "required_token_positions_per_second": 19182.999848504274,
+      "required_pages_per_second_upper_estimate": 0.003995312978717825
     },
     {
       "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": 28920.865700764392,
-      "required_pages_per_second_upper_estimate": 0.0017337617391558243
+      "required_token_positions_per_second": 28774.49977275641,
+      "required_pages_per_second_upper_estimate": 0.0017249873252595776
     },
     {
       "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": 28920.865700764392,
-      "required_pages_per_second_upper_estimate": 0.002248470098790122
+      "required_token_positions_per_second": 28774.49977275641,
+      "required_pages_per_second_upper_estimate": 0.0022370907916831806
     },
     {
       "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": 28920.865700764392,
-      "required_pages_per_second_upper_estimate": 0.370965812403181
+      "required_token_positions_per_second": 28774.49977275641,
+      "required_pages_per_second_upper_estimate": 0.36908838743418393
     }
   ],
   "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": 0.11381871815619604,
-    "producer_p95_seconds": 0.21427506300096866,
-    "gpu_update_seconds": 0.9559879806526418,
-    "producer_batches_per_second": 8.785901090782598,
-    "gpu_batches_per_second": 1.0460382559593602,
-    "producer_headroom": 8.399215841991099,
+    "producer_batch_seconds": 0.07816782107026654,
+    "producer_p95_seconds": 0.21202306899976975,
+    "gpu_update_seconds": 0.9608507608593433,
+    "producer_batches_per_second": 12.792988039171272,
+    "gpu_batches_per_second": 1.0407443494197197,
+    "producer_headroom": 12.292152291102195,
     "sustainable": true,
-    "buffer_seconds": 60.22724278111643,
+    "buffer_seconds": 60.533597934138626,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 74,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 217027.9765285454,
+    "generation_bytes_per_second": 224440.61553011162,
     "upload_bytes_per_second": 4568907.797879928,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 46.56915161030075,
+    "estimated_full_shards_per_hour": 48.159731382632366,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -117,3 +117,4 @@
 {"step": 451900, "loss": 0.5504519939422607, "lr": 0.0003, "grad_norm": 0.15542979430587375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66144122324, "tokens_per_second": 144212.62021890015, "data_wait_s": 1.4871999155730009e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452000, "loss": 0.578105628490448, "lr": 0.0003, "grad_norm": 0.17458674679668898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66158787132, "tokens_per_second": 129967.99176219804, "data_wait_s": 1.2552000043797307e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452100, "loss": 0.5985599756240845, "lr": 0.0003, "grad_norm": 0.15233343228801152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66173491630, "tokens_per_second": 153415.9382148165, "data_wait_s": 1.2500000593718141e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 452200, "loss": 0.6027845144271851, "lr": 0.0003, "grad_norm": 0.2114486650853434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66188068098, "tokens_per_second": 151629.75374200332, "data_wait_s": 1.3932000001659617e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:debd1a68170f5ee873de8280df42f476bfd734a9e94692de83e36a5350010778
-size 253773
+oid sha256:517440383b1d71c9d9accef9f31b03f87d89c8a1e862b4b14db6e230db96c4b6
+size 255891