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": 74722.80854342718,
-      "required_pages_per_second_upper_estimate": 0.00647678366131744
+      "required_token_positions_per_second": 74896.59510806805,
+      "required_pages_per_second_upper_estimate": 0.00649184704028249
     },
     {
       "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": 9340.351067928397,
-      "required_pages_per_second_upper_estimate": 0.0037260056916899624
+      "required_token_positions_per_second": 9362.074388508507,
+      "required_pages_per_second_upper_estimate": 0.003734671449061954
     },
     {
       "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": 18680.702135856794,
-      "required_pages_per_second_upper_estimate": 0.004321451670001739
+      "required_token_positions_per_second": 18724.148777017013,
+      "required_pages_per_second_upper_estimate": 0.004331502285799386
     },
     {
       "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": 28021.053203785188,
-      "required_pages_per_second_upper_estimate": 0.0017063351250062835
+      "required_token_positions_per_second": 28086.223165525524,
+      "required_pages_per_second_upper_estimate": 0.0017103036337559059
     },
     {
       "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": 28021.053203785188,
-      "required_pages_per_second_upper_estimate": 0.0021571382428671763
+      "required_token_positions_per_second": 28086.223165525524,
+      "required_pages_per_second_upper_estimate": 0.0021621552069239474
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28021.053203785188,
-      "required_pages_per_second_upper_estimate": 0.3479277003586574
+      "required_token_positions_per_second": 28086.223165525524,
+      "required_pages_per_second_upper_estimate": 0.348736893173641
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3923941665077848,
-    "producer_p95_seconds": 0.28253563799989934,
-    "gpu_update_seconds": 0.9866866815793064,
-    "producer_batches_per_second": 2.548457865466664,
-    "gpu_batches_per_second": 1.0134929544193139,
-    "producer_headroom": 2.514529434421985,
+    "producer_batch_seconds": 0.10985037101556827,
+    "producer_p95_seconds": 0.2576749749987357,
+    "gpu_update_seconds": 0.9843972198418113,
+    "producer_batches_per_second": 9.103291966654146,
+    "gpu_batches_per_second": 1.0158500855586488,
+    "producer_headroom": 8.961255303382636,
     "sustainable": true,
-    "buffer_seconds": 60.18788757633769,
+    "buffer_seconds": 60.048230410350484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318505.2603439513,
-    "upload_bytes_per_second": 5180435.991327726,
+    "generation_bytes_per_second": 323878.2889645178,
+    "upload_bytes_per_second": 5247245.840583565,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3438144468203,
+    "estimated_full_shards_per_hour": 69.49674131108905,
     "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": 1.5218778089993066,
-      "bytes": 10001472,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.655338631999257,
       "bytes": 6858339,
@@ -562,6 +556,12 @@
       "bytes": 861803,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.207310141000562,
+      "bytes": 13710377,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -151,3 +151,4 @@
 {"step": 210100, "loss": 0.5807416588068008, "lr": 0.0003, "grad_norm": 0.1436917269689581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30750746073, "tokens_per_second": 125072.04918072904, "data_wait_s": 1.2976999641978182e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210200, "loss": 0.5917802564799786, "lr": 0.0003, "grad_norm": 0.16181771319320384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30765397393, "tokens_per_second": 144146.832485573, "data_wait_s": 1.1764999726437964e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210300, "loss": 0.6302842348814011, "lr": 0.0003, "grad_norm": 0.17129778440251459, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30780130925, "tokens_per_second": 149987.96672474072, "data_wait_s": 1.647099998081103e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 210400, "loss": 0.6081433296203613, "lr": 0.0003, "grad_norm": 0.14212834516230086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30794933317, "tokens_per_second": 149589.68693309152, "data_wait_s": 1.61259995365981e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a26ef4560e3c58084b1a85422ae6aa3f50a185e222b50d0e474f97b76e5d04fc
-size 326568
+oid sha256:bc9347b30072605f8346ef8e7f72ed0b072b7e1071e699b2237d17cb47eaafc2
+size 328686