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": 75367.5744820617,
-      "required_pages_per_second_upper_estimate": 0.006532670338733949
+      "required_token_positions_per_second": 75581.25634216156,
+      "required_pages_per_second_upper_estimate": 0.006551191740795682
     },
     {
       "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": 9420.946810257712,
-      "required_pages_per_second_upper_estimate": 0.0037581565383188577
+      "required_token_positions_per_second": 9447.657042770195,
+      "required_pages_per_second_upper_estimate": 0.0037688116494216514
     },
     {
       "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": 18841.893620515424,
-      "required_pages_per_second_upper_estimate": 0.0043587404831042734
+      "required_token_positions_per_second": 18895.31408554039,
+      "required_pages_per_second_upper_estimate": 0.004371098367519726
     },
     {
       "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": 28262.840430773136,
-      "required_pages_per_second_upper_estimate": 0.0017210586985703092
+      "required_token_positions_per_second": 28342.971128310586,
+      "required_pages_per_second_upper_estimate": 0.001725938237637061
     },
     {
       "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": 28262.840430773136,
-      "required_pages_per_second_upper_estimate": 0.0021757516929105905
+      "required_token_positions_per_second": 28342.971128310586,
+      "required_pages_per_second_upper_estimate": 0.0021819203758229907
     },
     {
       "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": 28262.840430773136,
-      "required_pages_per_second_upper_estimate": 0.3509298885080539
+      "required_token_positions_per_second": 28342.971128310586,
+      "required_pages_per_second_upper_estimate": 0.3519248435912759
     }
   ],
   "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": 0.48974803266399647,
-    "producer_p95_seconds": 0.2843006750008499,
-    "gpu_update_seconds": 0.9782456249477428,
-    "producer_batches_per_second": 2.0418662930823333,
-    "gpu_batches_per_second": 1.0222381521546997,
-    "producer_headroom": 1.9974467679360581,
+    "producer_batch_seconds": 0.09060472796089414,
+    "producer_p95_seconds": 0.20138274999953865,
+    "gpu_update_seconds": 0.9754799479149732,
+    "producer_batches_per_second": 11.036951630511043,
+    "gpu_batches_per_second": 1.0251363978700299,
+    "producer_headroom": 10.76632500167099,
     "sustainable": true,
-    "buffer_seconds": 60.651228746760054,
+    "buffer_seconds": 60.479756770728336,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244780.56507529906,
-    "upload_bytes_per_second": 6495244.274867937,
+    "generation_bytes_per_second": 244837.86818029176,
+    "upload_bytes_per_second": 6469925.625094634,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.524211065237324,
+    "estimated_full_shards_per_hour": 52.53650697762074,
     "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": 1.1135005340001953,
-      "bytes": 5221142,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.3410022780008148,
       "bytes": 8747781,
@@ -562,6 +556,12 @@
       "bytes": 7362656,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9597948790014925,
+      "bytes": 2508350,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -137,3 +137,4 @@
 {"step": 13700, "loss": 0.6662630550563335, "lr": 0.0003, "grad_norm": 0.16334830063970154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2004188834, "tokens_per_second": 147413.34039166078, "data_wait_s": 8.015000275918283e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 13800, "loss": 0.7241083066910505, "lr": 0.0003, "grad_norm": 0.1441718568705546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2018860478, "tokens_per_second": 149899.95310192794, "data_wait_s": 1.2190999768790789e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 13900, "loss": 0.637822050601244, "lr": 0.0003, "grad_norm": 0.1495436076408611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2033595625, "tokens_per_second": 150288.13934782927, "data_wait_s": 1.3520000720745884e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 14000, "loss": 0.8031466826796532, "lr": 0.0003, "grad_norm": 0.2034141855442637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2048451112, "tokens_per_second": 152011.57959002908, "data_wait_s": 1.369399979012087e-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:d758ee7e5e67b7cff62731d3b0522d6b7a743f28b4616423f13cc4df4fbf6f1c
-size 291209
+oid sha256:5c3f1fd77f340664b78f7ebfc8708311399c9fddcf3737b17f92a357e8cbd7d4
+size 293292