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": 76308.55642027364,
-      "required_pages_per_second_upper_estimate": 0.006614232268241242
+      "required_token_positions_per_second": 76026.3914961327,
+      "required_pages_per_second_upper_estimate": 0.006589774927757167
     },
     {
       "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": 9538.569552534205,
-      "required_pages_per_second_upper_estimate": 0.0038050780088296652
+      "required_token_positions_per_second": 9503.298937016587,
+      "required_pages_per_second_upper_estimate": 0.003791008032956992
     },
     {
       "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": 19077.13910506841,
-      "required_pages_per_second_upper_estimate": 0.004413160385776484
+      "required_token_positions_per_second": 19006.597874033174,
+      "required_pages_per_second_upper_estimate": 0.004396841913459746
     },
     {
       "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": 28615.708657602616,
-      "required_pages_per_second_upper_estimate": 0.0017425465222277185
+      "required_token_positions_per_second": 28509.896811049763,
+      "required_pages_per_second_upper_estimate": 0.0017361031359245042
     },
     {
       "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": 28615.708657602616,
-      "required_pages_per_second_upper_estimate": 0.0022029164658136937
+      "required_token_positions_per_second": 28509.896811049763,
+      "required_pages_per_second_upper_estimate": 0.002194770776960117
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28615.708657602616,
-      "required_pages_per_second_upper_estimate": 0.35531133091129063
+      "required_token_positions_per_second": 28509.896811049763,
+      "required_pages_per_second_upper_estimate": 0.35399750190657414
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.31743760281244704,
-    "producer_p95_seconds": 0.36861487699934514,
-    "gpu_update_seconds": 0.9661826072811405,
-    "producer_batches_per_second": 3.15022540222128,
-    "gpu_batches_per_second": 1.0350010365162983,
-    "producer_headroom": 3.0436929926414362,
+    "producer_batch_seconds": 0.2516266434532213,
+    "producer_p95_seconds": 0.387721731000056,
+    "gpu_update_seconds": 0.9697685047139241,
+    "producer_batches_per_second": 3.9741419520461285,
+    "gpu_batches_per_second": 1.0311739297978069,
+    "producer_headroom": 3.8539976983566495,
     "sustainable": true,
-    "buffer_seconds": 60.86950425871185,
+    "buffer_seconds": 60.125647292263295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478942.11030320724,
+    "generation_bytes_per_second": 468736.8953248167,
     "upload_bytes_per_second": 5618687.659578934,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.76982766935504,
+    "estimated_full_shards_per_hour": 100.58002610023857,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -319,3 +319,4 @@
 {"step": 312100, "loss": 0.5786622334271669, "lr": 0.0003, "grad_norm": 0.15417092769651447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45682679327, "tokens_per_second": 150751.61326526327, "data_wait_s": 1.0034003935288638e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 312200, "loss": 0.5382980741560459, "lr": 0.0003, "grad_norm": 0.17255988027451766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45697309374, "tokens_per_second": 149458.23799435908, "data_wait_s": 1.0853000276256353e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 312300, "loss": 0.6069875601679087, "lr": 0.0003, "grad_norm": 0.19347847657333106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45712039954, "tokens_per_second": 150738.92517254225, "data_wait_s": 1.0047995601780713e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 312400, "loss": 0.5539985727518797, "lr": 0.0003, "grad_norm": 0.14900928106209144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45726875311, "tokens_per_second": 152208.859573901, "data_wait_s": 1.097100175684318e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:633e483d8148db1ad30325bfcaa73c9910d364084b22943494bf4ef7cd414d58
-size 684741
+oid sha256:07a0309234b23637f97fe682faaeba2b0033b94c9173f4e5f406de915c08c80c
+size 686859