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": 75738.98614958476,
-      "required_pages_per_second_upper_estimate": 0.006564863360740582
+      "required_token_positions_per_second": 75772.7178983662,
+      "required_pages_per_second_upper_estimate": 0.006567787143232622
     },
     {
       "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": 9467.373268698095,
-      "required_pages_per_second_upper_estimate": 0.0037766767467281373
+      "required_token_positions_per_second": 9471.589737295775,
+      "required_pages_per_second_upper_estimate": 0.0037783587590935756
     },
     {
       "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": 18934.74653739619,
-      "required_pages_per_second_upper_estimate": 0.004380220371269115
+      "required_token_positions_per_second": 18943.17947459155,
+      "required_pages_per_second_upper_estimate": 0.004382171182874636
     },
     {
       "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": 28402.119806094284,
-      "required_pages_per_second_upper_estimate": 0.0017295400817849593
+      "required_token_positions_per_second": 28414.769211887324,
+      "required_pages_per_second_upper_estimate": 0.0017303103642314524
     },
     {
       "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": 28402.119806094284,
-      "required_pages_per_second_upper_estimate": 0.0021864738047727996
+      "required_token_positions_per_second": 28414.769211887324,
+      "required_pages_per_second_upper_estimate": 0.002187447591046544
     },
     {
       "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": 28402.119806094284,
-      "required_pages_per_second_upper_estimate": 0.35265927221145915
+      "required_token_positions_per_second": 28414.769211887324,
+      "required_pages_per_second_upper_estimate": 0.3528163354965708
     }
   ],
   "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": 0.10158735789839568,
-    "producer_p95_seconds": 0.21321011100008036,
-    "gpu_update_seconds": 0.9734484675354241,
-    "producer_batches_per_second": 9.843744543491002,
-    "gpu_batches_per_second": 1.0272757453014427,
-    "producer_headroom": 9.582378040671509,
+    "producer_batch_seconds": 0.10950917097655122,
+    "producer_p95_seconds": 0.21912337499998102,
+    "gpu_update_seconds": 0.9730151173789388,
+    "producer_batches_per_second": 9.131655285876707,
+    "gpu_batches_per_second": 1.0277332614253227,
+    "producer_headroom": 8.88523863985133,
     "sustainable": true,
-    "buffer_seconds": 60.3538049871963,
+    "buffer_seconds": 60.3269372774942,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309363.5859994619,
+    "generation_bytes_per_second": 322863.33400529117,
     "upload_bytes_per_second": 5745043.504389929,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.38222393978016,
+    "estimated_full_shards_per_hour": 69.27895560378124,
     "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

@@ -70,3 +70,4 @@
 {"step": 92000, "loss": 0.6390262767672539, "lr": 0.0003, "grad_norm": 0.1441192939934209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13463728262, "tokens_per_second": 149921.50896801552, "data_wait_s": 9.022000085678883e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 92100, "loss": 0.6326837167143822, "lr": 0.0003, "grad_norm": 0.15781426775344015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13478260656, "tokens_per_second": 147960.2941388051, "data_wait_s": 1.72180007211864e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92200, "loss": 0.6220071911811829, "lr": 0.0003, "grad_norm": 0.13366662895182727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13492718975, "tokens_per_second": 147595.04661729187, "data_wait_s": 3.900800038536545e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 92300, "loss": 0.6198513470590115, "lr": 0.0003, "grad_norm": 0.13410049734005566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13507492070, "tokens_per_second": 141402.48420216204, "data_wait_s": 6.253554176000762, "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 sha256:85a764cf6eb8acb6d84e50a619d41c2b795e5b8df03c27280c4fb4ed392bc17b
-size 153444
+oid sha256:aa5f864227d0e26dbf9962c43382a6c4d78a1bb361d28e3525f7b144f9ce07b3
+size 155562