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": 76101.21084401335,
-      "required_pages_per_second_upper_estimate": 0.006596260079203557
+      "required_token_positions_per_second": 75799.18487431516,
+      "required_pages_per_second_upper_estimate": 0.006570081233627941
     },
     {
       "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": 9512.651355501668,
-      "required_pages_per_second_upper_estimate": 0.003794738852521808
+      "required_token_positions_per_second": 9474.898109289396,
+      "required_pages_per_second_upper_estimate": 0.0037796785181464
     },
     {
       "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": 19025.302711003336,
-      "required_pages_per_second_upper_estimate": 0.004401168948298906
+      "required_token_positions_per_second": 18949.79621857879,
+      "required_pages_per_second_upper_estimate": 0.0043837018501453676
     },
     {
       "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": 28537.954066505004,
-      "required_pages_per_second_upper_estimate": 0.0017378116755766859
+      "required_token_positions_per_second": 28424.694327868187,
+      "required_pages_per_second_upper_estimate": 0.001730914751721628
     },
     {
       "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": 28537.954066505004,
-      "required_pages_per_second_upper_estimate": 0.0021969307021525155
+      "required_token_positions_per_second": 28424.694327868187,
+      "required_pages_per_second_upper_estimate": 0.0021882116539492305
     },
     {
       "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": 28537.954066505004,
-      "required_pages_per_second_upper_estimate": 0.35434587911773474
+      "required_token_positions_per_second": 28424.694327868187,
+      "required_pages_per_second_upper_estimate": 0.35293957221982675
     }
   ],
   "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": 0.0774612383515958,
-    "producer_p95_seconds": 0.18188421099966945,
-    "gpu_update_seconds": 0.9688150711704473,
-    "producer_batches_per_second": 12.909682588096642,
-    "gpu_batches_per_second": 1.0321887321507885,
-    "producer_headroom": 12.507095055374732,
+    "producer_batch_seconds": 0.06945363478124733,
+    "producer_p95_seconds": 0.1868763400007083,
+    "gpu_update_seconds": 0.9726753674495384,
+    "producer_batches_per_second": 14.39809454393023,
+    "gpu_batches_per_second": 1.0280922427614363,
+    "producer_headroom": 14.00467190109053,
     "sustainable": true,
-    "buffer_seconds": 60.066534412567734,
+    "buffer_seconds": 60.30587278187138,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326227.1896180361,
-    "upload_bytes_per_second": 5879873.250275586,
+    "generation_bytes_per_second": 352213.3794002739,
+    "upload_bytes_per_second": 5881452.689870524,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.00076071172535,
+    "estimated_full_shards_per_hour": 75.57679211145556,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 6312525,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0277426739994553,
+      "bytes": 6105617,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -67,3 +67,4 @@
 {"step": 46700, "loss": 0.7140990644693375, "lr": 0.0003, "grad_norm": 0.14737092949938116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6833976873, "tokens_per_second": 151543.23811024037, "data_wait_s": 9.436000254936516e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 46800, "loss": 0.6093469150364399, "lr": 0.0003, "grad_norm": 0.12029286405930982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6848580225, "tokens_per_second": 149620.3313044925, "data_wait_s": 1.3033999493927695e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46900, "loss": 0.5917779915034771, "lr": 0.0003, "grad_norm": 0.12117127875398802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6863103323, "tokens_per_second": 148304.48442120262, "data_wait_s": 1.211200014950009e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 47000, "loss": 0.6305246166884899, "lr": 0.0003, "grad_norm": 0.12255809164608776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6877677278, "tokens_per_second": 148189.86050842697, "data_wait_s": 9.967000551114324e-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 sha256:8a002ba32b6fbe5c997ad4d52243322fd356decd8f8eecaffa26a48627bcb6e0
-size 147090
+oid sha256:fc6482c258a64c3fc879b2dc2ec2d21ac6356c9c022b0989ec32c59dc0212527
+size 149208