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": 75016.10323630324,
-      "required_pages_per_second_upper_estimate": 0.006502205701947334
+      "required_token_positions_per_second": 75181.58494793368,
+      "required_pages_per_second_upper_estimate": 0.006516549237301897
     },
     {
       "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": 9377.012904537905,
-      "required_pages_per_second_upper_estimate": 0.0037406306464567995
+      "required_token_positions_per_second": 9397.69811849171,
+      "required_pages_per_second_upper_estimate": 0.0037488822875744814
     },
     {
       "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": 18754.02580907581,
-      "required_pages_per_second_upper_estimate": 0.004338413811348384
+      "required_token_positions_per_second": 18795.39623698342,
+      "required_pages_per_second_upper_estimate": 0.004347984131750145
     },
     {
       "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": 28131.038713613718,
-      "required_pages_per_second_upper_estimate": 0.0017130326655054678
+      "required_token_positions_per_second": 28193.09435547513,
+      "required_pages_per_second_upper_estimate": 0.001716811528514037
     },
     {
       "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": 28131.038713613718,
-      "required_pages_per_second_upper_estimate": 0.0021656052318731536
+      "required_token_positions_per_second": 28193.09435547513,
+      "required_pages_per_second_upper_estimate": 0.0021703824469646565
     },
     {
       "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": 28131.038713613718,
-      "required_pages_per_second_upper_estimate": 0.34929335229290537
+      "required_token_positions_per_second": 28193.09435547513,
+      "required_pages_per_second_upper_estimate": 0.3500638756779509
     }
   ],
   "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": 0.12853406414836854,
-    "producer_p95_seconds": 0.3916308669940918,
-    "gpu_update_seconds": 0.9828289769698424,
-    "producer_batches_per_second": 7.780038751794909,
-    "gpu_batches_per_second": 1.017471018287533,
-    "producer_headroom": 7.64644752721232,
+    "producer_batch_seconds": 0.1192259365470818,
+    "producer_p95_seconds": 0.37940353600424714,
+    "gpu_update_seconds": 0.9806656783181634,
+    "producer_batches_per_second": 8.387436735337403,
+    "gpu_batches_per_second": 1.01971550764884,
+    "producer_headroom": 8.225271335410335,
     "sustainable": true,
-    "buffer_seconds": 60.93539657213023,
+    "buffer_seconds": 60.80127205572613,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357172.3266227915,
+    "generation_bytes_per_second": 371681.3214392803,
     "upload_bytes_per_second": 5948293.372819613,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.64086674702462,
+    "estimated_full_shards_per_hour": 79.7541592825299,
     "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

@@ -331,3 +331,4 @@
 {"step": 118100, "loss": 0.6580418385565281, "lr": 0.0003, "grad_norm": 0.15876329008956036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17282311564, "tokens_per_second": 148539.76533340095, "data_wait_s": 2.014799974858761e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118200, "loss": 0.5931661874055862, "lr": 0.0003, "grad_norm": 0.1305205706464004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17296945848, "tokens_per_second": 148410.52471017124, "data_wait_s": 2.3145999875850976e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118300, "loss": 0.6324690375477076, "lr": 0.0003, "grad_norm": 0.14163706276546853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17311707810, "tokens_per_second": 149922.68149478783, "data_wait_s": 1.6352998500224203e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 118400, "loss": 0.6224633827805519, "lr": 0.0003, "grad_norm": 0.13959889518325222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17326463478, "tokens_per_second": 149829.84798189733, "data_wait_s": 1.7182996089104563e-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 sha256:14a305de0ece410247bac1571d2c8ef62c17d3dc1693ae870d8c9ed302ffa954
-size 710157
+oid sha256:b9e4b1fe518d2a555b78fa8331431aef6e88fbc8425fdcaad1a10114a969b84e
+size 712275