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": 75651.10325525622,
-      "required_pages_per_second_upper_estimate": 0.006557245894197344
+      "required_token_positions_per_second": 75750.68154243374,
+      "required_pages_per_second_upper_estimate": 0.00656587708775103
     },
     {
       "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": 9456.387906907028,
-      "required_pages_per_second_upper_estimate": 0.0037722945216638855
+      "required_token_positions_per_second": 9468.835192804218,
+      "required_pages_per_second_upper_estimate": 0.003777259930111783
     },
     {
       "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": 18912.775813814056,
-      "required_pages_per_second_upper_estimate": 0.004375137831039917
+      "required_token_positions_per_second": 18937.670385608435,
+      "required_pages_per_second_upper_estimate": 0.004380896752095047
     },
     {
       "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": 28369.16372072108,
-      "required_pages_per_second_upper_estimate": 0.0017275332290929482
+      "required_token_positions_per_second": 28406.50557841265,
+      "required_pages_per_second_upper_estimate": 0.0017298071523087787
     },
     {
       "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": 28369.16372072108,
-      "required_pages_per_second_upper_estimate": 0.0021839367540924873
+      "required_token_positions_per_second": 28406.50557841265,
+      "required_pages_per_second_upper_estimate": 0.002186811433666462
     },
     {
       "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": 28369.16372072108,
-      "required_pages_per_second_upper_estimate": 0.3522500679280465
+      "required_token_positions_per_second": 28406.50557841265,
+      "required_pages_per_second_upper_estimate": 0.3527137288254175
     }
   ],
   "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": 0.14290578087508266,
-    "producer_p95_seconds": 0.37482676500076195,
-    "gpu_update_seconds": 0.9745793098513391,
-    "producer_batches_per_second": 6.997617548265062,
-    "gpu_batches_per_second": 1.026083757259877,
-    "producer_headroom": 6.819733280791783,
+    "producer_batch_seconds": 0.4461538195157573,
+    "producer_p95_seconds": 0.38899926399608376,
+    "gpu_update_seconds": 0.973298173676488,
+    "producer_batches_per_second": 2.241379444168766,
+    "gpu_batches_per_second": 1.0274343742192076,
+    "producer_headroom": 2.1815305195254817,
     "sustainable": true,
-    "buffer_seconds": 60.423917210783024,
+    "buffer_seconds": 60.34448676794226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452293.9372542591,
-    "upload_bytes_per_second": 6209362.779494619,
+    "generation_bytes_per_second": 404335.8024049325,
+    "upload_bytes_per_second": 6299080.344080756,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.05175007077054,
+    "estimated_full_shards_per_hour": 86.76105074034673,
     "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": 0.9876751570009219,
-      "bytes": 1005909,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.0930767579993699,
       "bytes": 7383476,
@@ -562,6 +556,12 @@
       "bytes": 843076,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.0585707080026623,
+      "bytes": 7490877,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 33500, "loss": 0.6309784017503262, "lr": 0.0003, "grad_norm": 0.12907676687920205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4902127146, "tokens_per_second": 148909.81059540564, "data_wait_s": 1.2658005289267749e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 33600, "loss": 0.6542370337992907, "lr": 0.0003, "grad_norm": 0.13428067431129173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4916809725, "tokens_per_second": 149745.1375648214, "data_wait_s": 9.159004548564553e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 33700, "loss": 0.6094903405755758, "lr": 0.0003, "grad_norm": 0.13040393853626536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4931552385, "tokens_per_second": 151119.10008874087, "data_wait_s": 1.1153002560604364e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 33800, "loss": 0.6564355380833149, "lr": 0.0003, "grad_norm": 0.13291179917537557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4946206923, "tokens_per_second": 148956.07547159746, "data_wait_s": 1.4435994671657681e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21daa93569c4ddaad723cfbe98c7c599c46ff3091487b60b7d86bf816163dc02
-size 712855
+oid sha256:4b2ae7f384a18891323b2124bb3b8672978798ac623afeaee60c1f22b8693057
+size 714973