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": 74968.14905933106,
-      "required_pages_per_second_upper_estimate": 0.006498049155426136
+      "required_token_positions_per_second": 75316.80132785274,
+      "required_pages_per_second_upper_estimate": 0.006528269450410499
     },
     {
       "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": 9371.018632416382,
-      "required_pages_per_second_upper_estimate": 0.0037382394416851693
+      "required_token_positions_per_second": 9414.600165981592,
+      "required_pages_per_second_upper_estimate": 0.003755624767026325
     },
     {
       "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": 18742.037264832765,
-      "required_pages_per_second_upper_estimate": 0.004335640472628927
+      "required_token_positions_per_second": 18829.200331963184,
+      "required_pages_per_second_upper_estimate": 0.004355804114191962
     },
     {
       "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": 28113.05589724915,
-      "required_pages_per_second_upper_estimate": 0.0017119376063374112
+      "required_token_positions_per_second": 28243.800497944776,
+      "required_pages_per_second_upper_estimate": 0.0017198992665558692
     },
     {
       "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": 28113.05589724915,
-      "required_pages_per_second_upper_estimate": 0.0021642208648897875
+      "required_token_positions_per_second": 28243.800497944776,
+      "required_pages_per_second_upper_estimate": 0.0021742859461755556
     },
     {
       "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": 28113.05589724915,
-      "required_pages_per_second_upper_estimate": 0.3490700658982722
+      "required_token_positions_per_second": 28243.800497944776,
+      "required_pages_per_second_upper_estimate": 0.3506934762648817
     }
   ],
   "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": 0.07015760253899828,
-    "producer_p95_seconds": 0.19595086099980108,
-    "gpu_update_seconds": 0.983457654018514,
-    "producer_batches_per_second": 14.25362275519796,
-    "gpu_batches_per_second": 1.0168205981354583,
-    "producer_headroom": 14.017834396091894,
+    "producer_batch_seconds": 0.08158054893745259,
+    "producer_p95_seconds": 0.18591325700072048,
+    "gpu_update_seconds": 0.978905087578843,
+    "producer_batches_per_second": 12.257823868857454,
+    "gpu_batches_per_second": 1.021549497176822,
+    "producer_headroom": 11.999246147869938,
     "sustainable": true,
-    "buffer_seconds": 60.97437454914787,
+    "buffer_seconds": 60.69211542988826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354323.43589874863,
-    "upload_bytes_per_second": 5935363.1756596,
+    "generation_bytes_per_second": 382012.5966570883,
+    "upload_bytes_per_second": 5941435.383486974,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.02956111642689,
+    "estimated_full_shards_per_hour": 81.97101044449317,
     "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": 6782943,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.0282262380005704,
+      "bytes": 6456198,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 48800, "loss": 0.5843997523188591, "lr": 0.0003, "grad_norm": 0.12371321141977733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7141618043, "tokens_per_second": 150900.76054005697, "data_wait_s": 1.0022999049397185e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 48900, "loss": 0.5825246330350637, "lr": 0.0003, "grad_norm": 0.16094780359559918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7156265642, "tokens_per_second": 149208.90017532525, "data_wait_s": 8.05400122771971e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49000, "loss": 0.5643719397485256, "lr": 0.0003, "grad_norm": 0.12859644528598635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7170937232, "tokens_per_second": 149682.37203272746, "data_wait_s": 9.879000572254881e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 49100, "loss": 0.6665434390306473, "lr": 0.0003, "grad_norm": 0.1404694792377278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7185626821, "tokens_per_second": 151023.4391343018, "data_wait_s": 7.995000487426296e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8a296f40806634ecfdf8f69c00c86d644fd2c4bace4df0090702098e2f9f233
-size 191568
+oid sha256:e6687a2dbb4f4db559e256f8930dfe97598611ef23c9492daf58624cf7e0dfc3
+size 193686