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": 77449.40552283586,
-      "required_pages_per_second_upper_estimate": 0.005358129581449443
+      "required_token_positions_per_second": 77282.39318032004,
+      "required_pages_per_second_upper_estimate": 0.005346575280072174
     },
     {
       "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": 9681.175690354483,
-      "required_pages_per_second_upper_estimate": 0.004053306095613952
+      "required_token_positions_per_second": 9660.299147540005,
+      "required_pages_per_second_upper_estimate": 0.004044565523089318
     },
     {
       "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": 19362.351380708966,
-      "required_pages_per_second_upper_estimate": 0.0040326671730580915
+      "required_token_positions_per_second": 19320.59829508001,
+      "required_pages_per_second_upper_estimate": 0.004023971106424494
     },
     {
       "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": 29043.527071063447,
-      "required_pages_per_second_upper_estimate": 0.0017411151010121848
+      "required_token_positions_per_second": 28980.897442620015,
+      "required_pages_per_second_upper_estimate": 0.0017373605504169066
     },
     {
       "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": 29043.527071063447,
-      "required_pages_per_second_upper_estimate": 0.002258006480800525
+      "required_token_positions_per_second": 28980.897442620015,
+      "required_pages_per_second_upper_estimate": 0.0022531373026676705
     },
     {
       "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": 29043.527071063447,
-      "required_pages_per_second_upper_estimate": 0.372539180757859
+      "required_token_positions_per_second": 28980.897442620015,
+      "required_pages_per_second_upper_estimate": 0.37173583513064246
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.37329934762500727,
-    "producer_p95_seconds": 0.20466321599997173,
-    "gpu_update_seconds": 0.9519504959694157,
-    "producer_batches_per_second": 2.6788152895582775,
-    "gpu_batches_per_second": 1.0504747927901998,
-    "producer_headroom": 2.5500995435054565,
+    "producer_batch_seconds": 0.19460904176575866,
+    "producer_p95_seconds": 0.2211941490004392,
+    "gpu_update_seconds": 0.9540077236994109,
+    "producer_batches_per_second": 5.138507393729685,
+    "gpu_batches_per_second": 1.0482095429188374,
+    "producer_headroom": 4.902175741904649,
     "sustainable": true,
-    "buffer_seconds": 60.924831742042606,
+    "buffer_seconds": 60.102486593062885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447238.73747340473,
-    "upload_bytes_per_second": 3255963.4890849297,
+    "generation_bytes_per_second": 430467.8106738023,
+    "upload_bytes_per_second": 3314330.3003355428,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.9670218768273,
+    "estimated_full_shards_per_hour": 92.36837139282753,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 6258608,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2659589719987707,
+      "bytes": 11586393,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -117,3 +117,4 @@
 {"step": 936700, "loss": 0.5473290085792542, "lr": 3e-05, "grad_norm": 0.3401454195545312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137106134044, "tokens_per_second": 152972.75210344046, "data_wait_s": 1.4083998394198716e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 936800, "loss": 0.5142284631729126, "lr": 3e-05, "grad_norm": 0.3525035614871149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137120860452, "tokens_per_second": 153951.09291976836, "data_wait_s": 1.2086999049643055e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 936900, "loss": 0.6139829754829407, "lr": 3e-05, "grad_norm": 0.5582208762462549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137135621824, "tokens_per_second": 153828.90388116034, "data_wait_s": 1.1075999282184057e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 937000, "loss": 0.5273974537849426, "lr": 3e-05, "grad_norm": 0.3514502216498336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137150286192, "tokens_per_second": 153196.9278589522, "data_wait_s": 1.2024000170640647e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e24b9a1a446d28f68dfbb44722f6859adf03d43355a29f93014cf606c8b174d
-size 253773
+oid sha256:6bf2d5da82a5db875068e7c8e65400a92686b35f55b3c44f3606a0ac3ab8379c
+size 255891