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": 75580.83972197319,
-      "required_pages_per_second_upper_estimate": 0.006551155629213668
+      "required_token_positions_per_second": 75591.65894578813,
+      "required_pages_per_second_upper_estimate": 0.006552093412112875
     },
     {
       "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": 9447.604965246648,
-      "required_pages_per_second_upper_estimate": 0.0037687908749188797
+      "required_token_positions_per_second": 9448.957368223517,
+      "required_pages_per_second_upper_estimate": 0.003769330368686579
     },
     {
       "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": 18895.209930493296,
-      "required_pages_per_second_upper_estimate": 0.004371074273082647
+      "required_token_positions_per_second": 18897.914736447034,
+      "required_pages_per_second_upper_estimate": 0.004371699982337085
     },
     {
       "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": 28342.814895739946,
-      "required_pages_per_second_upper_estimate": 0.001725928723893198
+      "required_token_positions_per_second": 28346.87210467055,
+      "required_pages_per_second_upper_estimate": 0.0017261757866305417
     },
     {
       "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": 28342.814895739946,
-      "required_pages_per_second_upper_estimate": 0.0021819083486072236
+      "required_token_positions_per_second": 28346.87210467055,
+      "required_pages_per_second_upper_estimate": 0.0021822206837817746
     },
     {
       "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": 28342.814895739946,
-      "required_pages_per_second_upper_estimate": 0.35192290370562535
+      "required_token_positions_per_second": 28346.87210467055,
+      "required_pages_per_second_upper_estimate": 0.3519732806619386
     }
   ],
   "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": 0.14598630674180413,
-    "producer_p95_seconds": 0.3995833680019132,
-    "gpu_update_seconds": 0.9754853250005038,
-    "producer_batches_per_second": 6.849957522171109,
-    "gpu_batches_per_second": 1.0251307470970756,
-    "producer_headroom": 6.682033039754729,
+    "producer_batch_seconds": 0.11450798038288212,
+    "producer_p95_seconds": 0.3899314090012922,
+    "gpu_update_seconds": 0.9753457065001749,
+    "producer_batches_per_second": 8.73301578332169,
+    "gpu_batches_per_second": 1.0252774922117531,
+    "producer_headroom": 8.517709449061073,
     "sustainable": true,
-    "buffer_seconds": 60.48009015003123,
+    "buffer_seconds": 60.47143380301085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403567.290453633,
-    "upload_bytes_per_second": 5725374.426390219,
+    "generation_bytes_per_second": 422182.298265576,
+    "upload_bytes_per_second": 5678758.337398992,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.59614596564047,
+    "estimated_full_shards_per_hour": 90.59049330687962,
     "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": 1.2462770620004449,
-      "bytes": 11030793,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8822051489987643,
       "bytes": 2343640,
@@ -562,6 +556,12 @@
       "bytes": 2665706,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9742672610009322,
+      "bytes": 6080743,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -363,3 +363,4 @@
 {"step": 231300, "loss": 0.5928157931193709, "lr": 0.0003, "grad_norm": 0.13617362397365723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33855910977, "tokens_per_second": 147701.33501886614, "data_wait_s": 1.848700048867613e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231400, "loss": 0.582163080573082, "lr": 0.0003, "grad_norm": 0.13887501363023447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33870486465, "tokens_per_second": 148673.70855027012, "data_wait_s": 1.2397998943924904e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231500, "loss": 0.5812797918915749, "lr": 0.0003, "grad_norm": 0.22347676210351913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33885191009, "tokens_per_second": 149381.63819913988, "data_wait_s": 1.534100010758266e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 231600, "loss": 0.6095680668950081, "lr": 0.0003, "grad_norm": 0.16434959095671053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33899904451, "tokens_per_second": 149145.79090702962, "data_wait_s": 1.3511002180166543e-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 sha256:56a38b01fcca828f979bd7fe683560e3ae0fa8a4b8684f53e37b113c327a5259
-size 778716
+oid sha256:d8452ce73db4f41bd1fc044c1fa8e84f257b4d6eb41056a34cf6b9669f1b7e93
+size 780834