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": 77261.42380576664,
-      "required_pages_per_second_upper_estimate": 0.005345124570084917
+      "required_token_positions_per_second": 77434.2773987804,
+      "required_pages_per_second_upper_estimate": 0.005357082982725202
     },
     {
       "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": 9657.67797572083,
-      "required_pages_per_second_upper_estimate": 0.004043468093184921
+      "required_token_positions_per_second": 9679.28467484755,
+      "required_pages_per_second_upper_estimate": 0.004052514366910094
     },
     {
       "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": 19315.35595144166,
-      "required_pages_per_second_upper_estimate": 0.0040228792644945235
+      "required_token_positions_per_second": 19358.5693496951,
+      "required_pages_per_second_upper_estimate": 0.004031879475736768
     },
     {
       "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": 28973.033927162487,
-      "required_pages_per_second_upper_estimate": 0.0017368891446721206
+      "required_token_positions_per_second": 29037.854024542645,
+      "required_pages_per_second_upper_estimate": 0.0017407750105355032
     },
     {
       "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": 28973.033927162487,
-      "required_pages_per_second_upper_estimate": 0.0022525259489287967
+      "required_token_positions_per_second": 29037.854024542645,
+      "required_pages_per_second_upper_estimate": 0.0022575654263866274
     },
     {
       "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": 28973.033927162487,
-      "required_pages_per_second_upper_estimate": 0.37163497039753834
+      "required_token_positions_per_second": 29037.854024542645,
+      "required_pages_per_second_upper_estimate": 0.37246641300833294
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1218054964294879,
-    "producer_p95_seconds": 0.30949456399685005,
-    "gpu_update_seconds": 0.9542666491022793,
-    "producer_batches_per_second": 8.20981014250774,
-    "gpu_batches_per_second": 1.0479251275738748,
-    "producer_headroom": 7.834348014456767,
+    "producer_batch_seconds": 0.11042765967147261,
+    "producer_p95_seconds": 0.31774244099506177,
+    "gpu_update_seconds": 0.9521364759472946,
+    "producer_batches_per_second": 9.055702194314778,
+    "gpu_batches_per_second": 1.0502696044756454,
+    "producer_headroom": 8.622264374523056,
     "sustainable": true,
-    "buffer_seconds": 60.118798893443596,
+    "buffer_seconds": 60.93673446062685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286416.33806753746,
+    "generation_bytes_per_second": 304844.88287171576,
     "upload_bytes_per_second": 5540130.38895154,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.458278718181546,
+    "estimated_full_shards_per_hour": 65.41261543859105,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -368,3 +368,4 @@
 {"step": 521800, "loss": 0.5179608464241028, "lr": 0.0003, "grad_norm": 0.15108013226515238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76373767291, "tokens_per_second": 152419.8019010753, "data_wait_s": 8.96599522093311e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 521900, "loss": 0.580098032951355, "lr": 0.0003, "grad_norm": 0.17006697372490467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76388434426, "tokens_per_second": 153243.27918139452, "data_wait_s": 1.3111995940562338e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522000, "loss": 0.6028584837913513, "lr": 0.0003, "grad_norm": 0.17199760238003273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76403168891, "tokens_per_second": 153939.4876438651, "data_wait_s": 1.3580000086221844e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 522100, "loss": 0.5927833318710327, "lr": 0.0003, "grad_norm": 0.15929691437666202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76417899080, "tokens_per_second": 154297.78805688574, "data_wait_s": 1.1994001397397369e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:943054ce911cdd3ffc3b961b550f4d606aa582d4564a4343656f9b1388b426b4
-size 789306
+oid sha256:f5e58c00d5a79f8c014df779624f4934dd1f8fd746ade2b93a5de929261e1244
+size 791424