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": 75036.80319564872,
-      "required_pages_per_second_upper_estimate": 0.006503999921959833
+      "required_token_positions_per_second": 74169.4774415367,
+      "required_pages_per_second_upper_estimate": 0.00642882232381043
     },
     {
       "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": 9379.60039945609,
-      "required_pages_per_second_upper_estimate": 0.0037416628368661603
+      "required_token_positions_per_second": 9271.184680192087,
+      "required_pages_per_second_upper_estimate": 0.003698414185492296
     },
     {
       "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": 18759.20079891218,
-      "required_pages_per_second_upper_estimate": 0.004339610954170317
+      "required_token_positions_per_second": 18542.369360384175,
+      "required_pages_per_second_upper_estimate": 0.004289450816969845
     },
     {
       "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": 28138.801198368274,
-      "required_pages_per_second_upper_estimate": 0.0017135053601004101
+      "required_token_positions_per_second": 27813.554040576262,
+      "required_pages_per_second_upper_estimate": 0.001693699514630833
     },
     {
       "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": 28138.801198368274,
-      "required_pages_per_second_upper_estimate": 0.0021662028094374926
+      "required_token_positions_per_second": 27813.554040576262,
+      "required_pages_per_second_upper_estimate": 0.0021411643828889064
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28138.801198368274,
-      "required_pages_per_second_upper_estimate": 0.3493897363741917
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27813.554040576262,
+      "required_pages_per_second_upper_estimate": 0.3453512552066288
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12581077795277906,
-    "producer_p95_seconds": 0.4192240530028357,
-    "gpu_update_seconds": 0.9825578497495929,
-    "producer_batches_per_second": 7.948444610805388,
-    "gpu_batches_per_second": 1.017751779454871,
-    "producer_headroom": 7.809806645646682,
+    "producer_batch_seconds": 0.1319990560697306,
+    "producer_p95_seconds": 0.40223021999554476,
+    "gpu_update_seconds": 0.9940477207502953,
+    "producer_batches_per_second": 7.575811750288078,
+    "gpu_batches_per_second": 1.0059879210277871,
+    "producer_headroom": 7.530718403207169,
     "sustainable": true,
-    "buffer_seconds": 60.91858668447476,
+    "buffer_seconds": 60.63691096576802,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386117.70338405063,
-    "upload_bytes_per_second": 3778544.794405317,
+    "generation_bytes_per_second": 398870.63640522305,
+    "upload_bytes_per_second": 3821068.2776880506,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.85187078610554,
+    "estimated_full_shards_per_hour": 85.58835333936233,
     "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": 0.9838440579987946,
-      "bytes": 1078556,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.215074095998716,
       "bytes": 7808319,
@@ -562,6 +556,12 @@
       "bytes": 1117608,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.9938163240003632,
+      "bytes": 5819862,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 353200, "loss": 0.5807544263079762, "lr": 0.0003, "grad_norm": 0.16400161316404074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51698876027, "tokens_per_second": 148157.21734976931, "data_wait_s": 1.4037999790161848e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 353300, "loss": 0.5942079722881317, "lr": 0.0003, "grad_norm": 0.15920950746602044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51713385954, "tokens_per_second": 147074.32379345736, "data_wait_s": 1.5839999832678586e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 353400, "loss": 0.6058970261365175, "lr": 0.0003, "grad_norm": 0.16720838608454577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51727941253, "tokens_per_second": 146409.04667974575, "data_wait_s": 1.5572994016110897e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 353500, "loss": 0.5445586293935776, "lr": 0.0003, "grad_norm": 0.16862535399243653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51742517664, "tokens_per_second": 147934.22499500206, "data_wait_s": 7.714996172580868e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f92d5f6d907f284ce5825ed56bc15c6db99b2c76256571a4a84097581cb75839
-size 818958
+oid sha256:d48eee51aac5f3767f34a6953691034e3d6ea996c4c1a728b2bb44c5f081a902
+size 821076