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": 75644.55660022405,
-      "required_pages_per_second_upper_estimate": 0.006556678446731501
+      "required_token_positions_per_second": 75673.06575542712,
+      "required_pages_per_second_upper_estimate": 0.00655914954804871
     },
     {
       "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": 9455.569575028007,
-      "required_pages_per_second_upper_estimate": 0.003771968076842192
+      "required_token_positions_per_second": 9459.13321942839,
+      "required_pages_per_second_upper_estimate": 0.0037733896678747367
     },
     {
       "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": 18911.139150056013,
-      "required_pages_per_second_upper_estimate": 0.004374759217683792
+      "required_token_positions_per_second": 18918.26643885678,
+      "required_pages_per_second_upper_estimate": 0.00437640799051185
     },
     {
       "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": 28366.70872508402,
-      "required_pages_per_second_upper_estimate": 0.001727383732739018
+      "required_token_positions_per_second": 28377.39965828517,
+      "required_pages_per_second_upper_estimate": 0.0017280347544799764
     },
     {
       "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": 28366.70872508402,
-      "required_pages_per_second_upper_estimate": 0.0021837477617324833
+      "required_token_positions_per_second": 28377.39965828517,
+      "required_pages_per_second_upper_estimate": 0.002184570779364705
     },
     {
       "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": 28366.70872508402,
-      "required_pages_per_second_upper_estimate": 0.35221958509857604
+      "required_token_positions_per_second": 28377.39965828517,
+      "required_pages_per_second_upper_estimate": 0.3523523307086826
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07069071313298991,
-    "producer_p95_seconds": 0.257033590998617,
-    "gpu_update_seconds": 0.9746636547775286,
-    "producer_batches_per_second": 14.146129748595229,
-    "gpu_batches_per_second": 1.025994962568143,
-    "producer_headroom": 13.787718521722947,
+    "producer_batch_seconds": 0.12859424987530588,
+    "producer_p95_seconds": 0.3372333940023964,
+    "gpu_update_seconds": 0.9742964589050284,
+    "producer_batches_per_second": 7.776397474767893,
+    "gpu_batches_per_second": 1.0263816427331152,
+    "producer_headroom": 7.576516522704362,
     "sustainable": true,
-    "buffer_seconds": 60.42914659620677,
+    "buffer_seconds": 60.40638045211176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429697.7371971901,
+    "generation_bytes_per_second": 520051.68271092384,
     "upload_bytes_per_second": 4918652.573586773,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.20313155113962,
+    "estimated_full_shards_per_hour": 111.59098492618357,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 216600, "loss": 0.601808425039053, "lr": 0.0003, "grad_norm": 0.1728315363195944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31703583637, "tokens_per_second": 149834.48276005237, "data_wait_s": 1.4024997653905302e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 216700, "loss": 0.6148982718586922, "lr": 0.0003, "grad_norm": 0.15622664756895163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31718036413, "tokens_per_second": 146886.537276718, "data_wait_s": 1.09370012069121e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 216800, "loss": 0.5722600612789392, "lr": 0.0003, "grad_norm": 0.14606250506051432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31732506474, "tokens_per_second": 146420.55161636823, "data_wait_s": 1.002699718810618e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 216900, "loss": 0.5820107161998749, "lr": 0.0003, "grad_norm": 0.1447592184058356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31747048305, "tokens_per_second": 147250.59297370855, "data_wait_s": 1.1703999916790053e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15f3d3d726a579853e2dd844478d81f5b3730c9fa781662d91a2f4bed8263d8f
-size 465021
+oid sha256:b904098c87416575189e40eb8f70c19861e1311c8c4a33418b65d6d0ad1df0e9
+size 467139