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": 74629.83811563866,
-      "required_pages_per_second_upper_estimate": 0.0064687252202681265
+      "required_token_positions_per_second": 74936.89336873229,
+      "required_pages_per_second_upper_estimate": 0.006495339991381858
     },
     {
       "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": 9328.729764454833,
-      "required_pages_per_second_upper_estimate": 0.0037213697799803862
+      "required_token_positions_per_second": 9367.111671091536,
+      "required_pages_per_second_upper_estimate": 0.003736680896398411
     },
     {
       "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": 18657.459528909665,
-      "required_pages_per_second_upper_estimate": 0.004316074901940432
+      "required_token_positions_per_second": 18734.22334218307,
+      "required_pages_per_second_upper_estimate": 0.004333832859144263
     },
     {
       "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": 27986.189293364496,
-      "required_pages_per_second_upper_estimate": 0.0017042120957784649
+      "required_token_positions_per_second": 28101.33501327461,
+      "required_pages_per_second_upper_estimate": 0.0017112238659980875
     },
     {
       "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": 27986.189293364496,
-      "required_pages_per_second_upper_estimate": 0.0021544543225335104
+      "required_token_positions_per_second": 28101.33501327461,
+      "required_pages_per_second_upper_estimate": 0.002163318558795944
     },
     {
       "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": 27986.189293364496,
-      "required_pages_per_second_upper_estimate": 0.3474948072732346
+      "required_token_positions_per_second": 28101.33501327461,
+      "required_pages_per_second_upper_estimate": 0.34892453174658367
     }
   ],
   "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": 0.09331656832790713,
-    "producer_p95_seconds": 0.29208238200226333,
-    "gpu_update_seconds": 0.9879158505711715,
-    "producer_batches_per_second": 10.716210614240314,
-    "gpu_batches_per_second": 1.0122319622889357,
-    "producer_headroom": 10.586714323867037,
+    "producer_batch_seconds": 0.08864658730473707,
+    "producer_p95_seconds": 0.22232287799852202,
+    "gpu_update_seconds": 0.9838678478065024,
+    "producer_batches_per_second": 11.280750115764043,
+    "gpu_batches_per_second": 1.0163966657000365,
+    "producer_headroom": 11.098767338039721,
     "sustainable": true,
-    "buffer_seconds": 60.26286688484146,
+    "buffer_seconds": 60.015938716196644,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281328.0916156472,
-    "upload_bytes_per_second": 5201474.588990371,
+    "generation_bytes_per_second": 291625.76022796764,
+    "upload_bytes_per_second": 5265005.528183636,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.366459477921126,
+    "estimated_full_shards_per_hour": 62.5760994446685,
     "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": 1.3869329340013792,
-      "bytes": 2462257,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.3128174869998475,
       "bytes": 5972623,
@@ -562,6 +556,12 @@
       "bytes": 1141925,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0775915699996403,
+      "bytes": 5849379,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -192,3 +192,4 @@
 {"step": 214200, "loss": 0.5742165800184011, "lr": 0.0003, "grad_norm": 0.13650238462357717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31351757875, "tokens_per_second": 149042.78802969292, "data_wait_s": 1.4353998267324641e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214300, "loss": 0.6546969227492809, "lr": 0.0003, "grad_norm": 0.13723219802122189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31366426304, "tokens_per_second": 149064.15503127806, "data_wait_s": 1.2709999282378703e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214400, "loss": 0.6502564568072557, "lr": 0.0003, "grad_norm": 0.142935596254749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31381178428, "tokens_per_second": 150158.68425512163, "data_wait_s": 1.8796003132592887e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 214500, "loss": 0.6368815768510103, "lr": 0.0003, "grad_norm": 0.14979569600535872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31395891452, "tokens_per_second": 149599.7179484442, "data_wait_s": 1.168600283563137e-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 sha256:89f3f9be55ea4faf5543e50113a2a8c548908c435094c368c65a0ef18ce64b9e
-size 413406
+oid sha256:4650df9ab7f7c55733f24b3dda52a195ec36bf72fac28d3b8f518ebe3e59dcbc
+size 415524