Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74233.28955853292,
-      "required_pages_per_second_upper_estimate": 0.006434353396381331
+      "required_token_positions_per_second": 75992.82542611643,
+      "required_pages_per_second_upper_estimate": 0.006586865505880583
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9279.161194816616,
-      "required_pages_per_second_upper_estimate": 0.0037015961364355414
+      "required_token_positions_per_second": 9499.103178264553,
+      "required_pages_per_second_upper_estimate": 0.003789334282058622
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18558.32238963323,
-      "required_pages_per_second_upper_estimate": 0.004293141269522889
+      "required_token_positions_per_second": 18998.206356529106,
+      "required_pages_per_second_upper_estimate": 0.004394900683570844
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27837.483584449845,
-      "required_pages_per_second_upper_estimate": 0.0016951566983041204
+      "required_token_positions_per_second": 28497.309534793658,
+      "required_pages_per_second_upper_estimate": 0.001735336636840842
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27837.483584449845,
-      "required_pages_per_second_upper_estimate": 0.0021430065454175118
+      "required_token_positions_per_second": 28497.309534793658,
+      "required_pages_per_second_upper_estimate": 0.0021938017735901112
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27837.483584449845,
-      "required_pages_per_second_upper_estimate": 0.34564838005450715
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28497.309534793658,
+      "required_pages_per_second_upper_estimate": 0.353841210062377
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.5426407134218607,
-    "producer_p95_seconds": 0.4942385039976216,
-    "gpu_update_seconds": 0.9931932215110244,
-    "producer_batches_per_second": 1.842839977291896,
-    "gpu_batches_per_second": 1.0068534282570112,
-    "producer_headroom": 1.8302961737758412,
+    "producer_batch_seconds": 0.1355042502342485,
+    "producer_p95_seconds": 0.4894277540006442,
+    "gpu_update_seconds": 0.9701968519604738,
+    "producer_batches_per_second": 7.379842316910968,
+    "gpu_batches_per_second": 1.0307186608359975,
+    "producer_headroom": 7.15989978383171,
     "sustainable": true,
-    "buffer_seconds": 60.584786512172485,
+    "buffer_seconds": 60.152204821549375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 255395.57177607968,
+    "generation_bytes_per_second": 256798.19968139278,
     "upload_bytes_per_second": 5261212.351291741,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.80194439851563,
+    "estimated_full_shards_per_hour": 55.10291569548929,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 389900, "loss": 0.6039017792791128, "lr": 0.0003, "grad_norm": 0.16267038271425097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57070706750, "tokens_per_second": 149413.2835373003, "data_wait_s": 1.9758997950702906e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390000, "loss": 0.6361445300281048, "lr": 0.0003, "grad_norm": 0.16592915181678689, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57085381676, "tokens_per_second": 140165.60097290107, "data_wait_s": 1.0950003343168646e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390100, "loss": 0.5821457048878074, "lr": 0.0003, "grad_norm": 0.15114542171762976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57100091088, "tokens_per_second": 132205.08004367616, "data_wait_s": 8.513001375831664e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 390200, "loss": 0.5980184301733971, "lr": 0.0003, "grad_norm": 0.16292192317862475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57114717086, "tokens_per_second": 148854.8882911339, "data_wait_s": 8.458999218419194e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcc932db9472659efa3de4fa268a236cb3a9ba7a78a54724924a09c7de7eea3d
-size 749064
+oid sha256:3fb0dd554c052c460071594af63c667fb9754e0c33e85a64fb9f244b41c3f2e1
+size 751182