Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76302.79692573685,
-      "required_pages_per_second_upper_estimate": 0.005278804538203643
+      "required_token_positions_per_second": 76927.36162945612,
+      "required_pages_per_second_upper_estimate": 0.005322013373596714
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9537.849615717107,
-      "required_pages_per_second_upper_estimate": 0.0039932984611520655
+      "required_token_positions_per_second": 9615.920203682015,
+      "required_pages_per_second_upper_estimate": 0.0040259849860337085
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19075.699231434213,
-      "required_pages_per_second_upper_estimate": 0.003972965089889693
+      "required_token_positions_per_second": 19231.84040736403,
+      "required_pages_per_second_upper_estimate": 0.00400548517911616
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28613.54884715132,
-      "required_pages_per_second_upper_estimate": 0.0017153385630273858
+      "required_token_positions_per_second": 28847.760611046047,
+      "required_pages_per_second_upper_estimate": 0.0017293792006522182
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28613.54884715132,
-      "required_pages_per_second_upper_estimate": 0.0022245775651656883
+      "required_token_positions_per_second": 28847.760611046047,
+      "required_pages_per_second_upper_estimate": 0.002242786499619828
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28613.54884715132,
-      "required_pages_per_second_upper_estimate": 0.3670238817761614
+      "required_token_positions_per_second": 28847.760611046047,
+      "required_pages_per_second_upper_estimate": 0.3700280988064038
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1027402259061887,
-    "producer_p95_seconds": 0.2915463490062393,
-    "gpu_update_seconds": 0.9662555367630518,
-    "producer_batches_per_second": 9.73328597615789,
-    "gpu_batches_per_second": 1.0349229183720818,
-    "producer_headroom": 9.404841465360727,
+    "producer_batch_seconds": 0.11758859633584962,
+    "producer_p95_seconds": 0.2860190520004835,
+    "gpu_update_seconds": 0.9584106153949902,
+    "producer_batches_per_second": 8.504226014773227,
+    "gpu_batches_per_second": 1.0433941193231353,
+    "producer_headroom": 8.150540488276892,
     "sustainable": true,
-    "buffer_seconds": 60.87409881607226,
+    "buffer_seconds": 60.37986876988438,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453416.456890981,
-    "upload_bytes_per_second": 5069528.452454102,
+    "generation_bytes_per_second": 451012.4654282901,
+    "upload_bytes_per_second": 5097647.062415216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.2926166538913,
+    "estimated_full_shards_per_hour": 96.77677604805496,
     "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": 1.076318840001477,
-      "bytes": 1024576,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0973008010005287,
       "bytes": 11080849,
@@ -562,6 +556,12 @@
       "bytes": 6085922,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.8856082540005445,
+      "bytes": 2405776,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 754700, "loss": 0.5171409249305725, "lr": 5e-05, "grad_norm": 0.23123891530112975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110467115769, "tokens_per_second": 152356.0597015072, "data_wait_s": 9.814000804908574e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 754800, "loss": 0.5421836972236633, "lr": 5e-05, "grad_norm": 0.232307074208228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110481629052, "tokens_per_second": 150601.5728082544, "data_wait_s": 1.2577998859342188e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 754900, "loss": 0.5689280033111572, "lr": 5e-05, "grad_norm": 0.24612306966984038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110496164840, "tokens_per_second": 150521.54136494003, "data_wait_s": 1.1061994882766157e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 755000, "loss": 0.5770891308784485, "lr": 5e-05, "grad_norm": 0.24166446750022272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110510795957, "tokens_per_second": 150614.04831445645, "data_wait_s": 1.3443997886497527e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4b18d729836b5a1eb9c434cbf9afc1805ef8fbedeb5581007725b0c5630e793
-size 744045
+oid sha256:38323ccc6a7582e7553ffd613fabd0d8a1863cd2fd1282587ce7f5c30703138f
+size 746163