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": 76491.19287319155,
-      "required_pages_per_second_upper_estimate": 0.005291838206987384
+      "required_token_positions_per_second": 76795.94880016235,
+      "required_pages_per_second_upper_estimate": 0.005312921929146398
     },
     {
       "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": 9561.399109148944,
-      "required_pages_per_second_upper_estimate": 0.004003158142282632
+      "required_token_positions_per_second": 9599.493600020294,
+      "required_pages_per_second_upper_estimate": 0.004019107510107037
     },
     {
       "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": 19122.79821829789,
-      "required_pages_per_second_upper_estimate": 0.003982774566769586
+      "required_token_positions_per_second": 19198.98720004059,
+      "required_pages_per_second_upper_estimate": 0.003998642722428026
     },
     {
       "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": 28684.197327446836,
-      "required_pages_per_second_upper_estimate": 0.0017195738315471188
+      "required_token_positions_per_second": 28798.480800060883,
+      "required_pages_per_second_upper_estimate": 0.0017264249512295722
     },
     {
       "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": 28684.197327446836,
-      "required_pages_per_second_upper_estimate": 0.002230070173758841
+      "required_token_positions_per_second": 28798.480800060883,
+      "required_pages_per_second_upper_estimate": 0.002238955211074712
     },
     {
       "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": 28684.197327446836,
-      "required_pages_per_second_upper_estimate": 0.36793008462496424
+      "required_token_positions_per_second": 28798.480800060883,
+      "required_pages_per_second_upper_estimate": 0.3693959903036247
     }
   ],
   "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": 0.43800370014017176,
-    "producer_p95_seconds": 0.29702584000187926,
-    "gpu_update_seconds": 0.9638756728794591,
-    "producer_batches_per_second": 2.283085735759711,
-    "gpu_batches_per_second": 1.0374782019475852,
-    "producer_headroom": 2.2006107997968867,
+    "producer_batch_seconds": 0.07315630215589408,
+    "producer_p95_seconds": 0.2592390179997892,
+    "gpu_update_seconds": 0.9600506426693712,
+    "producer_batches_per_second": 13.66936231780862,
+    "gpu_batches_per_second": 1.041611718752202,
+    "producer_headroom": 13.12328007809265,
     "sustainable": true,
-    "buffer_seconds": 60.72416739140592,
+    "buffer_seconds": 60.483190488170386,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310084.48527229973,
-    "upload_bytes_per_second": 4895949.201999769,
+    "generation_bytes_per_second": 309329.2355705279,
+    "upload_bytes_per_second": 4867212.241845977,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.53691214205497,
+    "estimated_full_shards_per_hour": 66.37485313736799,
     "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": 1.1313278150009864,
-      "bytes": 6410101,
-      "pages": 20,
-      "shards": 2
-    },
     {
       "seconds": 1.2548322069997084,
       "bytes": 7827361,
@@ -562,6 +556,12 @@
       "bytes": 7230636,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.5622936799991294,
+      "bytes": 5942154,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -266,3 +266,4 @@
 {"step": 911600, "loss": 0.5138594508171082, "lr": 3e-05, "grad_norm": 0.3166163818687345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133431940195, "tokens_per_second": 150725.50715979826, "data_wait_s": 1.540899756946601e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911700, "loss": 0.5337302088737488, "lr": 3e-05, "grad_norm": 0.3233251101419872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133446371612, "tokens_per_second": 149147.513570988, "data_wait_s": 1.6968002455541864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911800, "loss": 0.5464526414871216, "lr": 3e-05, "grad_norm": 0.3241595484439124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133460996453, "tokens_per_second": 152485.41767541075, "data_wait_s": 1.4429999282583594e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 911900, "loss": 0.5535483956336975, "lr": 3e-05, "grad_norm": 0.3318998368641891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133475655567, "tokens_per_second": 152193.08449663408, "data_wait_s": 1.050200080499053e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e25463b8979b91782d9669c882e5d7f9ebb1c67cb58982c672bb14be1822da37
-size 571704
+oid sha256:2e466915f8d612df7a5c1920cf20c898584ed94ac544038be46da8cb1eddfa0d
+size 573822