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": 75434.74103291096,
-      "required_pages_per_second_upper_estimate": 0.006538492165129496
+      "required_token_positions_per_second": 75597.47022224982,
+      "required_pages_per_second_upper_estimate": 0.006552597118828028
     },
     {
       "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": 9429.34262911387,
-      "required_pages_per_second_upper_estimate": 0.003761505755989257
+      "required_token_positions_per_second": 9449.683777781227,
+      "required_pages_per_second_upper_estimate": 0.0037696201443199405
     },
     {
       "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": 18858.68525822774,
-      "required_pages_per_second_upper_estimate": 0.004362624932966287
+      "required_token_positions_per_second": 18899.367555562454,
+      "required_pages_per_second_upper_estimate": 0.0043720360664706946
     },
     {
       "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": 28288.02788734161,
-      "required_pages_per_second_upper_estimate": 0.0017225924825269574
+      "required_token_positions_per_second": 28349.051333343683,
+      "required_pages_per_second_upper_estimate": 0.0017263084902232062
     },
     {
       "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": 28288.02788734161,
-      "required_pages_per_second_upper_estimate": 0.0021776906930405736
+      "required_token_positions_per_second": 28349.051333343683,
+      "required_pages_per_second_upper_estimate": 0.0021823884468375235
     },
     {
       "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": 28288.02788734161,
-      "required_pages_per_second_upper_estimate": 0.3512426324216399
+      "required_token_positions_per_second": 28349.051333343683,
+      "required_pages_per_second_upper_estimate": 0.3520003393886497
     }
   ],
   "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": 0.16526897266396645,
-    "producer_p95_seconds": 0.27403312000024016,
-    "gpu_update_seconds": 0.9773746020793478,
-    "producer_batches_per_second": 6.050742519185694,
-    "gpu_batches_per_second": 1.0231491568048905,
-    "producer_headroom": 5.913842061973709,
+    "producer_batch_seconds": 0.08147948204674549,
+    "producer_p95_seconds": 0.24823009600004298,
+    "gpu_update_seconds": 0.9752707303993937,
+    "producer_batches_per_second": 12.27302843464679,
+    "gpu_batches_per_second": 1.0253563126932757,
+    "producer_headroom": 11.969525405670502,
     "sustainable": true,
-    "buffer_seconds": 60.59722532891956,
+    "buffer_seconds": 60.46678528476241,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 495377.05611306115,
-    "upload_bytes_per_second": 5882090.189655369,
+    "generation_bytes_per_second": 526854.8161409865,
+    "upload_bytes_per_second": 5922843.278577422,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.29638445419194,
+    "estimated_full_shards_per_hour": 113.05077899143406,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 2523849,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.5673655419996066,
+      "bytes": 11589450,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -110,3 +110,4 @@
 {"step": 96000, "loss": 0.6212404612451792, "lr": 0.0003, "grad_norm": 0.1366567958379472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14048197658, "tokens_per_second": 149438.04337328838, "data_wait_s": 1.1753996659535915e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 96100, "loss": 0.6387616209685802, "lr": 0.0003, "grad_norm": 0.14333084500028356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14062985718, "tokens_per_second": 150250.28948042548, "data_wait_s": 1.4741999621037394e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 96200, "loss": 0.6147613078355789, "lr": 0.0003, "grad_norm": 0.1378868176610411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14077682953, "tokens_per_second": 150106.42102338027, "data_wait_s": 8.982999133877456e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 96300, "loss": 0.603190841153264, "lr": 0.0003, "grad_norm": 0.13281830211263865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14092219734, "tokens_per_second": 149040.19338791494, "data_wait_s": 1.1356001778040081e-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 sha256:31ef6cc9d322d9f5561cb7521ad863ae8ac648123792e8e3a99d797c7de4503e
-size 238947
+oid sha256:3269ebda6170d6f5b18a3ad92d9b0e5a5799b11e82ee3ea729ce06612e678b5e
+size 241065