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": 75889.48278542777,
-      "required_pages_per_second_upper_estimate": 0.006577908027705219
+      "required_token_positions_per_second": 74353.30766112472,
+      "required_pages_per_second_upper_estimate": 0.006444756261330888
     },
     {
       "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": 9486.185348178471,
-      "required_pages_per_second_upper_estimate": 0.0037841811664985126
+      "required_token_positions_per_second": 9294.16345764059,
+      "required_pages_per_second_upper_estimate": 0.003707580763379843
     },
     {
       "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": 18972.370696356942,
-      "required_pages_per_second_upper_estimate": 0.004388924058282103
+      "required_token_positions_per_second": 18588.32691528118,
+      "required_pages_per_second_upper_estimate": 0.004300082288469926
     },
     {
       "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": 28458.556044535413,
-      "required_pages_per_second_upper_estimate": 0.0017329767526079647
+      "required_token_positions_per_second": 27882.49037292177,
+      "required_pages_per_second_upper_estimate": 0.0016978973755896907
     },
     {
       "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": 28458.556044535413,
-      "required_pages_per_second_upper_estimate": 0.0021908184226335043
+      "required_token_positions_per_second": 27882.49037292177,
+      "required_pages_per_second_upper_estimate": 0.002146471292580845
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28458.556044535413,
-      "required_pages_per_second_upper_estimate": 0.3533600214129582
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27882.49037292177,
+      "required_pages_per_second_upper_estimate": 0.3462072137392971
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07277678164834356,
-    "producer_p95_seconds": 0.17719575699993584,
-    "gpu_update_seconds": 0.971518019281549,
-    "producer_batches_per_second": 13.740646087264297,
-    "gpu_batches_per_second": 1.0293169865645042,
-    "producer_headroom": 13.349285270347776,
+    "producer_batch_seconds": 0.4502374353360068,
+    "producer_p95_seconds": 0.18879999700038752,
+    "gpu_update_seconds": 0.9915900491747504,
+    "producer_batches_per_second": 2.2210503203797614,
+    "gpu_batches_per_second": 1.0084812779557932,
+    "producer_headroom": 2.2023713964049625,
     "sustainable": true,
-    "buffer_seconds": 60.23411719545604,
+    "buffer_seconds": 60.48699299965977,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389645.90411545493,
-    "upload_bytes_per_second": 5705294.197757101,
+    "generation_bytes_per_second": 347871.3618436244,
+    "upload_bytes_per_second": 5692622.209798694,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.60894053075539,
+    "estimated_full_shards_per_hour": 74.64509622079419,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 5051955,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0979891799997858,
+      "bytes": 6035254,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -18,3 +18,4 @@
 {"step": 282000, "loss": 0.5302728898823261, "lr": 0.0003, "grad_norm": 0.14995796687799284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41277296948, "tokens_per_second": 149942.32708424868, "data_wait_s": 8.96500023372937e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 282100, "loss": 0.6072414442896843, "lr": 0.0003, "grad_norm": 0.15289263625057586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41291768611, "tokens_per_second": 148103.2368538803, "data_wait_s": 1.7689000742393546e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282200, "loss": 0.6118101850152016, "lr": 0.0003, "grad_norm": 0.16666576891706478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41306225341, "tokens_per_second": 147814.78691748617, "data_wait_s": 9.528999726171605e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 282300, "loss": 0.6148794535547495, "lr": 0.0003, "grad_norm": 0.16453517197819229, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41320742235, "tokens_per_second": 148631.58136930884, "data_wait_s": 1.3801998647977598e-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 sha256:0c8f7a38fc77c56e2e478adbb10cdd8450aedb7ca520bfb2986340fac26f088b
-size 42525
+oid sha256:3b4d7384bca8eebd71e965b63f815951b9399286220724890a35f2f5d52e6120
+size 44643