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": 75520.71083902438,
-      "required_pages_per_second_upper_estimate": 0.006545943810034926
+      "required_token_positions_per_second": 75033.2361082719,
+      "required_pages_per_second_upper_estimate": 0.006503690735866711
     },
     {
       "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": 9440.088854878048,
-      "required_pages_per_second_upper_estimate": 0.003765792586116981
+      "required_token_positions_per_second": 9379.154513533987,
+      "required_pages_per_second_upper_estimate": 0.003741484966305245
     },
     {
       "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": 18880.177709756095,
-      "required_pages_per_second_upper_estimate": 0.004367596833373144
+      "required_token_positions_per_second": 18758.309027067975,
+      "required_pages_per_second_upper_estimate": 0.004339404658448808
     },
     {
       "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": 28320.266564634145,
-      "required_pages_per_second_upper_estimate": 0.0017245556488308053
+      "required_token_positions_per_second": 28137.463540601962,
+      "required_pages_per_second_upper_estimate": 0.0017134239037605897
     },
     {
       "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": 28320.266564634145,
-      "required_pages_per_second_upper_estimate": 0.0021801725156609234
+      "required_token_positions_per_second": 28137.463540601962,
+      "required_pages_per_second_upper_estimate": 0.0021660998328397684
     },
     {
       "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": 28320.266564634145,
-      "required_pages_per_second_upper_estimate": 0.351642928897701
+      "required_token_positions_per_second": 28137.463540601962,
+      "required_pages_per_second_upper_estimate": 0.34937312714158664
     }
   ],
   "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": 0.5168304311641805,
-    "producer_p95_seconds": 0.36067550600273535,
-    "gpu_update_seconds": 0.9762619972838229,
-    "producer_batches_per_second": 1.9348705875299592,
-    "gpu_batches_per_second": 1.0243151969268716,
-    "producer_headroom": 1.8889406242677218,
+    "producer_batch_seconds": 0.32697384019587616,
+    "producer_p95_seconds": 0.3592587620005361,
+    "gpu_update_seconds": 0.9826045606457855,
+    "producer_batches_per_second": 3.0583486415945154,
+    "gpu_batches_per_second": 1.0177033977358927,
+    "producer_headroom": 3.005147323275614,
     "sustainable": true,
-    "buffer_seconds": 60.52824383159702,
+    "buffer_seconds": 60.9214827600387,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341762.4666184092,
+    "generation_bytes_per_second": 317418.9943673647,
     "upload_bytes_per_second": 6092780.155580305,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.33426951326568,
+    "estimated_full_shards_per_hour": 68.11072705522255,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -259,3 +259,4 @@
 {"step": 110900, "loss": 0.6345473565161228, "lr": 0.0003, "grad_norm": 0.13995303506369336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16227984156, "tokens_per_second": 147334.913506741, "data_wait_s": 1.5177000022958964e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111000, "loss": 0.6521772630512714, "lr": 0.0003, "grad_norm": 0.13849151102313498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16242613239, "tokens_per_second": 148622.19497898594, "data_wait_s": 1.523299579275772e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111100, "loss": 0.643631000071764, "lr": 0.0003, "grad_norm": 0.14187004986412452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16257234626, "tokens_per_second": 148983.6677029596, "data_wait_s": 1.8320002709515393e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 111200, "loss": 0.6151546761393547, "lr": 0.0003, "grad_norm": 0.14512287742372176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16271827997, "tokens_per_second": 148151.61254821444, "data_wait_s": 7.84599978942424e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6d911bb5148989db808f085e982880f9dd5fb02c06f0e36d09e3cfe383b6109
-size 556878
+oid sha256:34608870fc48578ebdd5759fed8a79aefadc1368019f83f85dfd518568dad22e
+size 558996