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": 75339.49909667733,
-      "required_pages_per_second_upper_estimate": 0.006530236835485246
+      "required_token_positions_per_second": 74859.9297558378,
+      "required_pages_per_second_upper_estimate": 0.006488668980478657
     },
     {
       "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": 9417.437387084667,
-      "required_pages_per_second_upper_estimate": 0.0037567565769445776
+      "required_token_positions_per_second": 9357.491219479725,
+      "required_pages_per_second_upper_estimate": 0.00373284315441189
     },
     {
       "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": 18834.874774169333,
-      "required_pages_per_second_upper_estimate": 0.004357116796529582
+      "required_token_positions_per_second": 18714.98243895945,
+      "required_pages_per_second_upper_estimate": 0.0043293818148118085
     },
     {
       "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": 28252.312161253998,
-      "required_pages_per_second_upper_estimate": 0.0017204175821941548
+      "required_token_positions_per_second": 28072.473658439176,
+      "required_pages_per_second_upper_estimate": 0.0017094663609124358
     },
     {
       "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": 28252.312161253998,
-      "required_pages_per_second_upper_estimate": 0.0021749411975788917
+      "required_token_positions_per_second": 28072.473658439176,
+      "required_pages_per_second_upper_estimate": 0.002161096731807369
     },
     {
       "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": 28252.312161253998,
-      "required_pages_per_second_upper_estimate": 0.35079916263647765
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28072.473658439176,
+      "required_pages_per_second_upper_estimate": 0.3485661703122686
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 688946,
-    "producer_batch_seconds": 0.0931812604063964,
-    "producer_p95_seconds": 0.30755716299972846,
-    "gpu_update_seconds": 0.9786101697516011,
-    "producer_batches_per_second": 10.731771556197529,
-    "gpu_batches_per_second": 1.02185735536943,
-    "producer_headroom": 10.50222078434587,
+    "producer_batch_seconds": 0.11710105349220612,
+    "producer_p95_seconds": 0.314315744999476,
+    "gpu_update_seconds": 0.9848793639063022,
+    "producer_batches_per_second": 8.539632822915268,
+    "gpu_batches_per_second": 1.0153527798914632,
+    "producer_headroom": 8.410508142626169,
     "sustainable": true,
-    "buffer_seconds": 60.67383052459927,
+    "buffer_seconds": 60.07764119828443,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380605.3204424056,
-    "upload_bytes_per_second": 4102950.8987672143,
+    "generation_bytes_per_second": 396524.67483674356,
+    "upload_bytes_per_second": 4066696.241870003,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.6690417285359,
+    "estimated_full_shards_per_hour": 85.08496459795694,
     "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": [
-    {
-      "seconds": 1.0740993799990974,
-      "bytes": 5965207,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9706554090007558,
       "bytes": 2646898,
@@ -562,6 +556,12 @@
       "bytes": 10839288,
       "pages": 47,
       "shards": 3
+    },
+    {
+      "seconds": 1.171269268001197,
+      "bytes": 2613392,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -183,3 +183,4 @@
 {"step": 103300, "loss": 0.6290736962109804, "lr": 0.0003, "grad_norm": 0.14952531139442538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15116529417, "tokens_per_second": 146797.9021008877, "data_wait_s": 9.146002412308007e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 103400, "loss": 0.6758065614849329, "lr": 0.0003, "grad_norm": 0.14666754645000032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15131179122, "tokens_per_second": 148702.89277946352, "data_wait_s": 1.7318001482635736e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103500, "loss": 0.5845393426716328, "lr": 0.0003, "grad_norm": 0.13813541500181312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15145877141, "tokens_per_second": 148842.16556128208, "data_wait_s": 1.5707002603448927e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 103600, "loss": 0.5744451247155666, "lr": 0.0003, "grad_norm": 0.13890033496057527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15160483280, "tokens_per_second": 144457.2000252497, "data_wait_s": 1.4171997463563457e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5928e087019214320f1060d2e74d7a63d505624e3751f0a1122342fb775c07af
-size 394344
+oid sha256:6fd54c1eb56156430d8bc52b46dec47e4d5a1dc95d9722b3b2b416fc45f8a044
+size 396462