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": 75339.21571701854,
-      "required_pages_per_second_upper_estimate": 0.0065302122728546404
+      "required_token_positions_per_second": 75739.36095564146,
+      "required_pages_per_second_upper_estimate": 0.00656489584797964
     },
     {
       "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": 9417.401964627317,
-      "required_pages_per_second_upper_estimate": 0.0037567424463967277
+      "required_token_positions_per_second": 9467.420119455182,
+      "required_pages_per_second_upper_estimate": 0.003776695436195621
     },
     {
       "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": 18834.803929254635,
-      "required_pages_per_second_upper_estimate": 0.0043571004078053945
+      "required_token_positions_per_second": 18934.840238910365,
+      "required_pages_per_second_upper_estimate": 0.004380242047465329
     },
     {
       "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": 28252.205893881954,
-      "required_pages_per_second_upper_estimate": 0.001720411111068739
+      "required_token_positions_per_second": 28402.260358365547,
+      "required_pages_per_second_upper_estimate": 0.0017295486406808586
     },
     {
       "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": 28252.205893881954,
-      "required_pages_per_second_upper_estimate": 0.002174933016822425
+      "required_token_positions_per_second": 28402.260358365547,
+      "required_pages_per_second_upper_estimate": 0.002186484624875715
     },
     {
       "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": 28252.205893881954,
-      "required_pages_per_second_upper_estimate": 0.35079784315137086
+      "required_token_positions_per_second": 28402.260358365547,
+      "required_pages_per_second_upper_estimate": 0.35266101740027
     }
   ],
   "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": 0.30392081617935673,
-    "producer_p95_seconds": 0.39768118700158084,
-    "gpu_update_seconds": 0.9786138506794333,
-    "producer_batches_per_second": 3.2903307268359567,
-    "gpu_batches_per_second": 1.0218535117868184,
-    "producer_headroom": 3.219963222597794,
+    "producer_batch_seconds": 0.09887767583592222,
+    "producer_p95_seconds": 0.3873123610028415,
+    "gpu_update_seconds": 0.9734436502993541,
+    "producer_batches_per_second": 10.113506325324654,
+    "gpu_batches_per_second": 1.027280828933939,
+    "producer_headroom": 9.844928514649638,
     "sustainable": true,
-    "buffer_seconds": 60.67405874212486,
+    "buffer_seconds": 60.35350631855995,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400930.69358753716,
-    "upload_bytes_per_second": 5910843.1311028255,
+    "generation_bytes_per_second": 400793.7440945206,
+    "upload_bytes_per_second": 5886818.6496439045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.03039365501009,
+    "estimated_full_shards_per_hour": 86.00100748182976,
     "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": 1.1067068009979266,
-      "bytes": 10335265,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.2287768099959067,
       "bytes": 2576063,
@@ -562,6 +556,12 @@
       "bytes": 2367982,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3454053969981032,
+      "bytes": 10041564,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -318,3 +318,4 @@
 {"step": 116800, "loss": 0.6324906665831804, "lr": 0.0003, "grad_norm": 0.13395251782172857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17092217704, "tokens_per_second": 150176.06867900243, "data_wait_s": 2.231699909316376e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116900, "loss": 0.6175306066870689, "lr": 0.0003, "grad_norm": 0.1364368309102881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17106677430, "tokens_per_second": 146819.77538763385, "data_wait_s": 1.5719000657554716e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 117000, "loss": 0.6302737966179848, "lr": 0.0003, "grad_norm": 0.13048163314792993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17121284753, "tokens_per_second": 148935.38785399648, "data_wait_s": 1.483599771745503e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 117100, "loss": 0.6070490796118975, "lr": 0.0003, "grad_norm": 0.14627181813312434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17135712327, "tokens_per_second": 135482.06160661468, "data_wait_s": 1.779400190571323e-05, "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 sha256:25cf9930724d0922095aa21ac7571d8a244f9be9ef0af368f9376a246baaddfb
-size 682623
+oid sha256:8e57c81f81da6754cce7efd3d3ba1792d6bde64c97b42feea4461cf8351e6ff4
+size 684741