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": 76065.0880979941,
-      "required_pages_per_second_upper_estimate": 0.006593129051130872
+      "required_token_positions_per_second": 76111.85832149869,
+      "required_pages_per_second_upper_estimate": 0.006597182975566202
     },
     {
       "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": 9508.136012249262,
-      "required_pages_per_second_upper_estimate": 0.0037929376145880256
+      "required_token_positions_per_second": 9513.982290187336,
+      "required_pages_per_second_upper_estimate": 0.0037952697822671677
     },
     {
       "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": 19016.272024498525,
-      "required_pages_per_second_upper_estimate": 0.0043990798579106715
+      "required_token_positions_per_second": 19027.96458037467,
+      "required_pages_per_second_upper_estimate": 0.004401784724930663
     },
     {
       "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": 28524.408036747787,
-      "required_pages_per_second_upper_estimate": 0.0017369867934349968
+      "required_token_positions_per_second": 28541.94687056201,
+      "required_pages_per_second_upper_estimate": 0.0017380548163951335
     },
     {
       "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": 28524.408036747787,
-      "required_pages_per_second_upper_estimate": 0.0021958878912839946
+      "required_token_positions_per_second": 28541.94687056201,
+      "required_pages_per_second_upper_estimate": 0.0021972380792616127
     },
     {
       "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": 28524.408036747787,
-      "required_pages_per_second_upper_estimate": 0.35417768276379535
+      "required_token_positions_per_second": 28541.94687056201,
+      "required_pages_per_second_upper_estimate": 0.3543954563810672
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08934853077349203,
-    "producer_p95_seconds": 0.21759994499734603,
-    "gpu_update_seconds": 0.969275154260214,
-    "producer_batches_per_second": 11.192125839596688,
-    "gpu_batches_per_second": 1.0316987860513522,
-    "producer_headroom": 10.848249499674807,
+    "producer_batch_seconds": 0.39223933092984,
+    "producer_p95_seconds": 0.29193562000000384,
+    "gpu_update_seconds": 0.9686795412164396,
+    "producer_batches_per_second": 2.549463863374962,
+    "gpu_batches_per_second": 1.0323331478067856,
+    "producer_headroom": 2.4696134855219496,
     "sustainable": true,
-    "buffer_seconds": 60.09505956413327,
+    "buffer_seconds": 60.05813155541926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346725.0642022869,
+    "generation_bytes_per_second": 322176.1943315808,
     "upload_bytes_per_second": 5182151.64904447,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.39912743140654,
+    "estimated_full_shards_per_hour": 69.13151142559593,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -170,3 +170,4 @@
 {"step": 172000, "loss": 0.6820664405822754, "lr": 0.0003, "grad_norm": 0.14046266448044453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25173218574, "tokens_per_second": 149847.75509250868, "data_wait_s": 1.0988002031808719e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172100, "loss": 0.5721213407814503, "lr": 0.0003, "grad_norm": 0.14904676025056426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25187867206, "tokens_per_second": 149166.60639444797, "data_wait_s": 9.974999557016417e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 172200, "loss": 0.6270482540130615, "lr": 0.0003, "grad_norm": 0.13754360075202146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25202419943, "tokens_per_second": 140488.59696334766, "data_wait_s": 1.0435996955493465e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 172300, "loss": 0.6193122211843729, "lr": 0.0003, "grad_norm": 0.138339238126607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25217023351, "tokens_per_second": 149087.81708934403, "data_wait_s": 1.4073000784264877e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c97ce4f4e2d65b6a9e313283d3809c9cafecf153c3379c51472c9d957eb71e7
-size 366810
+oid sha256:2a25fa62dbafe6293e19fce617c1b8d128f8bb30379adbbc98df347e47e6ee21
+size 368928