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": 75657.63776137683,
-      "required_pages_per_second_upper_estimate": 0.006557812288626324
+      "required_token_positions_per_second": 75945.64833491984,
+      "required_pages_per_second_upper_estimate": 0.006582776315184907
     },
     {
       "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": 9457.204720172103,
-      "required_pages_per_second_upper_estimate": 0.0037726203606877704
+      "required_token_positions_per_second": 9493.20604186498,
+      "required_pages_per_second_upper_estimate": 0.0037869818261787856
     },
     {
       "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": 18914.409440344207,
-      "required_pages_per_second_upper_estimate": 0.004375515741786822
+      "required_token_positions_per_second": 18986.41208372996,
+      "required_pages_per_second_upper_estimate": 0.004392172286130873
     },
     {
       "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": 28371.61416051631,
-      "required_pages_per_second_upper_estimate": 0.0017276824480200138
+      "required_token_positions_per_second": 28479.61812559494,
+      "required_pages_per_second_upper_estimate": 0.0017342593228402909
     },
     {
       "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": 28371.61416051631,
-      "required_pages_per_second_upper_estimate": 0.0021841253957311756
+      "required_token_positions_per_second": 28479.61812559494,
+      "required_pages_per_second_upper_estimate": 0.0021924398399370388
     },
     {
       "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": 28371.61416051631,
-      "required_pages_per_second_upper_estimate": 0.3522804941892088
+      "required_token_positions_per_second": 28479.61812559494,
+      "required_pages_per_second_upper_estimate": 0.3536215419694667
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.27749369498440046,
-    "producer_p95_seconds": 0.19855503200051317,
-    "gpu_update_seconds": 0.9744951360038113,
-    "producer_batches_per_second": 3.603685482137588,
-    "gpu_batches_per_second": 1.026172387171452,
-    "producer_headroom": 3.5117739740306293,
+    "producer_batch_seconds": 0.08361664436721128,
+    "producer_p95_seconds": 0.24965563299883797,
+    "gpu_update_seconds": 0.9707995338305623,
+    "producer_batches_per_second": 11.959341439347828,
+    "gpu_batches_per_second": 1.0300787805843077,
+    "producer_headroom": 11.610123094239396,
     "sustainable": true,
-    "buffer_seconds": 60.418698432236305,
+    "buffer_seconds": 60.18957109749486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 509145.93712588673,
-    "upload_bytes_per_second": 6119414.0332479365,
+    "generation_bytes_per_second": 507691.2834705876,
+    "upload_bytes_per_second": 6164060.617038061,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.25086579639866,
+    "estimated_full_shards_per_hour": 108.93873098457547,
     "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": 1.0832384449995516,
-      "bytes": 7778060,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9103183699999136,
       "bytes": 5940010,
@@ -562,6 +556,12 @@
       "bytes": 7371694,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.1864866850010003,
+      "bytes": 11535625,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 53200, "loss": 0.6077895984053612, "lr": 0.0003, "grad_norm": 0.1264437712783856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7786437542, "tokens_per_second": 148231.86932768498, "data_wait_s": 1.2634000086109154e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 53300, "loss": 0.664090633392334, "lr": 0.0003, "grad_norm": 0.13915791119886245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7800905803, "tokens_per_second": 147873.39977916967, "data_wait_s": 1.1521000487846322e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 53400, "loss": 0.6035120617598295, "lr": 0.0003, "grad_norm": 0.13725951580014492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7815517624, "tokens_per_second": 149912.85062598373, "data_wait_s": 1.0322999514755793e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 53500, "loss": 0.6287303790450096, "lr": 0.0003, "grad_norm": 0.12792731073113495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7830201501, "tokens_per_second": 150569.31614186527, "data_wait_s": 9.946999853127636e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71fb679d43816d85e3663302a6bf7682a5fe82eb4fa559d03e1528e90f595845
-size 285543
+oid sha256:4363e8ff12f041cc9386fe4039dbe167e7d2e8beb19df2d3b3b27ff22e0cd6de
+size 287661