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": 76682.1991780959,
-      "required_pages_per_second_upper_estimate": 0.005305052466356355
+      "required_token_positions_per_second": 75973.96795348493,
+      "required_pages_per_second_upper_estimate": 0.00525605538691492
     },
     {
       "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": 9585.274897261988,
-      "required_pages_per_second_upper_estimate": 0.004013154436182412
+      "required_token_positions_per_second": 9496.745994185616,
+      "required_pages_per_second_upper_estimate": 0.003976089233158053
     },
     {
       "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": 19170.549794523977,
-      "required_pages_per_second_upper_estimate": 0.003992719960803731
+      "required_token_positions_per_second": 18993.491988371232,
+      "required_pages_per_second_upper_estimate": 0.003955843489110462
     },
     {
       "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": 28755.824691785965,
-      "required_pages_per_second_upper_estimate": 0.001723867782670597
+      "required_token_positions_per_second": 28490.23798255685,
+      "required_pages_per_second_upper_estimate": 0.0017079462649797355
     },
     {
       "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": 28755.824691785965,
-      "required_pages_per_second_upper_estimate": 0.0022356388862807284
+      "required_token_positions_per_second": 28490.23798255685,
+      "required_pages_per_second_upper_estimate": 0.0022149906878306405
     },
     {
       "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": 28755.824691785965,
-      "required_pages_per_second_upper_estimate": 0.3688488435472347
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28490.23798255685,
+      "required_pages_per_second_upper_estimate": 0.36544218240603443
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1447656134140658,
-    "producer_p95_seconds": 0.2542012909998448,
-    "gpu_update_seconds": 0.9614747723753369,
-    "producer_batches_per_second": 6.907717768167433,
-    "gpu_batches_per_second": 1.0400688907619344,
-    "producer_headroom": 6.641596368781853,
+    "producer_batch_seconds": 0.10759828768742352,
+    "producer_p95_seconds": 0.24315363999994588,
+    "gpu_update_seconds": 0.97043766418966,
+    "producer_batches_per_second": 9.29382819645822,
+    "gpu_batches_per_second": 1.0304628899940989,
+    "producer_headroom": 9.019080926350915,
     "sustainable": true,
-    "buffer_seconds": 60.57291065964622,
+    "buffer_seconds": 60.16713517975892,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408985.2196992628,
-    "upload_bytes_per_second": 5380139.543360519,
+    "generation_bytes_per_second": 393200.02115253615,
+    "upload_bytes_per_second": 5468981.310514635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.75870745881475,
+    "estimated_full_shards_per_hour": 84.37157131130279,
     "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": 1.1523795839993909,
-      "bytes": 2573171,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.718842794000011,
       "bytes": 5798754,
@@ -562,6 +556,12 @@
       "bytes": 5882576,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.4825863509995543,
+      "bytes": 11209775,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -139,3 +139,4 @@
 {"step": 411000, "loss": 0.556839108467102, "lr": 0.0003, "grad_norm": 0.16034621578071848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60159845688, "tokens_per_second": 152360.1507626238, "data_wait_s": 9.575000149197876e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 411100, "loss": 0.5723373889923096, "lr": 0.0003, "grad_norm": 0.17086279167399485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60174470644, "tokens_per_second": 151938.0017538998, "data_wait_s": 1.5191999409580603e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411200, "loss": 0.5532742142677307, "lr": 0.0003, "grad_norm": 0.15798192539545453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60189103437, "tokens_per_second": 151911.07878424195, "data_wait_s": 1.0320000001229346e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 411300, "loss": 0.5872836112976074, "lr": 0.0003, "grad_norm": 0.16727288171828153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60203851133, "tokens_per_second": 153061.57260279963, "data_wait_s": 1.3950999345979653e-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 sha256:043f3624c93248868fa7b44f4ad2e8933264769bb7b48022b89e124f906bb52d
-size 301152
+oid sha256:8c0449ede27b514cb8e7216de90525022f2e7f5ea621079c010e0fa872e28d2e
+size 303270