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": 75394.94298198726,
-      "required_pages_per_second_upper_estimate": 0.006535042570942668
+      "required_token_positions_per_second": 75217.03630188268,
+      "required_pages_per_second_upper_estimate": 0.006519622070811561
     },
     {
       "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": 9424.367872748408,
-      "required_pages_per_second_upper_estimate": 0.0037595212512958383
+      "required_token_positions_per_second": 9402.129537735334,
+      "required_pages_per_second_upper_estimate": 0.003750650046966385
     },
     {
       "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": 18848.735745496815,
-      "required_pages_per_second_upper_estimate": 0.0043603232883014315
+      "required_token_positions_per_second": 18804.25907547067,
+      "required_pages_per_second_upper_estimate": 0.004350034393453541
     },
     {
       "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": 28273.103618245223,
-      "required_pages_per_second_upper_estimate": 0.001721683672840576
+      "required_token_positions_per_second": 28206.388613206,
+      "required_pages_per_second_upper_estimate": 0.0017176210790602674
     },
     {
       "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": 28273.103618245223,
-      "required_pages_per_second_upper_estimate": 0.0021765417814925128
+      "required_token_positions_per_second": 28206.388613206,
+      "required_pages_per_second_upper_estimate": 0.0021714058757256395
     },
     {
       "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": 28273.103618245223,
-      "required_pages_per_second_upper_estimate": 0.3510573229477783
+      "required_token_positions_per_second": 28206.388613206,
+      "required_pages_per_second_upper_estimate": 0.3502289458659498
     }
   ],
   "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": 0.3295790421640987,
-    "producer_p95_seconds": 0.21509005100051581,
-    "gpu_update_seconds": 0.9778905200261838,
-    "producer_batches_per_second": 3.0341735124713907,
-    "gpu_batches_per_second": 1.0226093611923186,
-    "producer_headroom": 2.967089513960321,
+    "producer_batch_seconds": 0.10269297482037132,
+    "producer_p95_seconds": 0.20321003699973517,
+    "gpu_update_seconds": 0.9802034701831851,
+    "producer_batches_per_second": 9.737764455155592,
+    "gpu_batches_per_second": 1.020196347410518,
+    "producer_headroom": 9.544990510769983,
     "sustainable": true,
-    "buffer_seconds": 60.6292122416234,
+    "buffer_seconds": 60.77261515135748,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 224589.6405330118,
-    "upload_bytes_per_second": 4203699.583367449,
+    "generation_bytes_per_second": 282587.78223521425,
+    "upload_bytes_per_second": 4566987.638922346,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.19170867913023,
+    "estimated_full_shards_per_hour": 60.63675976078339,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 10096800,
       "pages": 18,
       "shards": 3
+    },
+    {
+      "seconds": 1.2823710859993298,
+      "bytes": 6954773,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -50,3 +50,4 @@
 {"step": 90000, "loss": 0.6308812461793423, "lr": 0.0003, "grad_norm": 0.13849090878893888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13170804190, "tokens_per_second": 148330.89333950818, "data_wait_s": 7.241998901008628e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 90100, "loss": 0.6027867775410414, "lr": 0.0003, "grad_norm": 0.14209351892134955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13185465865, "tokens_per_second": 133230.76583391437, "data_wait_s": 1.3549000868806615e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 90200, "loss": 0.6176210697740316, "lr": 0.0003, "grad_norm": 0.14878402689218964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13200068105, "tokens_per_second": 148392.5689643031, "data_wait_s": 1.0420000762678683e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 90300, "loss": 0.6494478732347488, "lr": 0.0003, "grad_norm": 0.1417464911937885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13214693795, "tokens_per_second": 149297.36775440862, "data_wait_s": 1.1100000847363845e-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 sha256:773cbcb0e21b5e7dca80bfce04155bb1c9faa2593bd05fba94e4cf283a0a6361
-size 111084
+oid sha256:2dd03cf9cf30486181d97d24f43f02c34d15ab8d047d8e578fd63536e3304ea5
+size 113202