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": 75757.21789836236,
-      "required_pages_per_second_upper_estimate": 0.00656644364251668
+      "required_token_positions_per_second": 75935.55275741976,
+      "required_pages_per_second_upper_estimate": 0.006581901256114208
     },
     {
       "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": 9469.652237295295,
-      "required_pages_per_second_upper_estimate": 0.0037775858613751773
+      "required_token_positions_per_second": 9491.94409467747,
+      "required_pages_per_second_upper_estimate": 0.003786478416577896
     },
     {
       "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": 18939.30447459059,
-      "required_pages_per_second_upper_estimate": 0.004381274769822085
+      "required_token_positions_per_second": 18983.88818935494,
+      "required_pages_per_second_upper_estimate": 0.0043915884275862365
     },
     {
       "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": 28408.956711885887,
-      "required_pages_per_second_upper_estimate": 0.0017299564134771956
+      "required_token_positions_per_second": 28475.83228403241,
+      "required_pages_per_second_upper_estimate": 0.0017340287849519085
     },
     {
       "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": 28408.956711885887,
-      "required_pages_per_second_upper_estimate": 0.0021870001287064037
+      "required_token_positions_per_second": 28475.83228403241,
+      "required_pages_per_second_upper_estimate": 0.0021921483953736753
     },
     {
       "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": 28408.956711885887,
-      "required_pages_per_second_upper_estimate": 0.3527441636997391
+      "required_token_positions_per_second": 28475.83228403241,
+      "required_pages_per_second_upper_estimate": 0.35357453448765674
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07958720800785102,
-    "producer_p95_seconds": 0.2192232369998237,
-    "gpu_update_seconds": 0.9732141972124054,
-    "producer_batches_per_second": 12.564833282018805,
-    "gpu_batches_per_second": 1.0275230292204096,
-    "producer_headroom": 12.228274135667645,
+    "producer_batch_seconds": 0.07942257610159231,
+    "producer_p95_seconds": 0.2163908549991902,
+    "gpu_update_seconds": 0.9709286009351652,
+    "producer_batches_per_second": 12.59087842631626,
+    "gpu_batches_per_second": 1.0299418505509408,
+    "producer_headroom": 12.224843975008001,
     "sustainable": true,
-    "buffer_seconds": 60.339280227169134,
+    "buffer_seconds": 60.197573257980245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484660.31643388374,
+    "generation_bytes_per_second": 465775.2571183369,
     "upload_bytes_per_second": 4672730.834553713,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.99682159197219,
+    "estimated_full_shards_per_hour": 99.94452748453693,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 250100, "loss": 0.568646140396595, "lr": 0.0003, "grad_norm": 0.15801772858695023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36607938466, "tokens_per_second": 136372.44597009715, "data_wait_s": 1.2935999620822258e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250200, "loss": 0.6084444671869278, "lr": 0.0003, "grad_norm": 0.1512300028949969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36622541903, "tokens_per_second": 149158.94468358142, "data_wait_s": 1.3777998901787214e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250300, "loss": 0.5982154533267021, "lr": 0.0003, "grad_norm": 0.14984847969284623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36637338504, "tokens_per_second": 151302.56202830066, "data_wait_s": 1.4895998901920393e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 250400, "loss": 0.6021800842136145, "lr": 0.0003, "grad_norm": 0.15040273044040206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36652199404, "tokens_per_second": 150873.33745707825, "data_wait_s": 1.5345000065281056e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ae4f083465a29dd4beb5e3c161bad1710ffa61553a48c794ce921d642847404
-size 282090
+oid sha256:309af84cd3cfba8a5a17f9c57267cb71b3ea94a17c1785f645507e5da277d036
+size 284208