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": 75010.39590930459,
-      "required_pages_per_second_upper_estimate": 0.006501711005308179
+      "required_token_positions_per_second": 74127.73189660434,
+      "required_pages_per_second_upper_estimate": 0.006425203925778809
     },
     {
       "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": 9376.299488663073,
-      "required_pages_per_second_upper_estimate": 0.003740346054197811
+      "required_token_positions_per_second": 9265.966487075542,
+      "required_pages_per_second_upper_estimate": 0.0036963325702391663
     },
     {
       "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": 18752.598977326146,
-      "required_pages_per_second_upper_estimate": 0.00433808373893448
+      "required_token_positions_per_second": 18531.932974151085,
+      "required_pages_per_second_upper_estimate": 0.004287036542688944
     },
     {
       "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": 28128.89846598922,
-      "required_pages_per_second_upper_estimate": 0.0017129023356541471
+      "required_token_positions_per_second": 27797.89946122663,
+      "required_pages_per_second_upper_estimate": 0.0016927462328816705
     },
     {
       "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": 28128.89846598922,
-      "required_pages_per_second_upper_estimate": 0.0021654404696864354
+      "required_token_positions_per_second": 27797.89946122663,
+      "required_pages_per_second_upper_estimate": 0.0021399592500359224
     },
     {
       "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": 28128.89846598922,
-      "required_pages_per_second_upper_estimate": 0.34926677758035707
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27797.89946122663,
+      "required_pages_per_second_upper_estimate": 0.3451568777236131
     }
   ],
   "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": 0.09876644691394176,
-    "producer_p95_seconds": 0.3008249760023318,
-    "gpu_update_seconds": 0.9829037576224082,
-    "producer_batches_per_second": 10.124895966656881,
-    "gpu_batches_per_second": 1.017393607710837,
-    "producer_headroom": 9.951798291163014,
+    "producer_batch_seconds": 0.1065382271092119,
+    "producer_p95_seconds": 0.2779873849976866,
+    "gpu_update_seconds": 0.9946075255997053,
+    "producer_batches_per_second": 9.386302242244975,
+    "gpu_batches_per_second": 1.0054217108371901,
+    "producer_headroom": 9.33568684769024,
     "sustainable": true,
-    "buffer_seconds": 60.94003297258931,
+    "buffer_seconds": 60.67105906158203,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442816.2377519854,
-    "upload_bytes_per_second": 3911924.9343966153,
+    "generation_bytes_per_second": 442715.176426963,
+    "upload_bytes_per_second": 3969190.513530301,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.01805638713523,
+    "estimated_full_shards_per_hour": 94.99637097937267,
     "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": 1.0145048790000146,
-      "bytes": 1134517,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 3.9685035570000764,
       "bytes": 7473998,
@@ -562,6 +556,12 @@
       "bytes": 821123,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.1538495610002428,
+      "bytes": 7799438,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -262,3 +262,4 @@
 {"step": 341200, "loss": 0.6028633490204811, "lr": 0.0003, "grad_norm": 0.16473557046001944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49942427086, "tokens_per_second": 148579.36652209962, "data_wait_s": 1.2051001249346882e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341300, "loss": 0.5985691305249929, "lr": 0.0003, "grad_norm": 0.16874504542210036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49957132232, "tokens_per_second": 149777.43470337408, "data_wait_s": 1.2324999261181802e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341400, "loss": 0.559636328369379, "lr": 0.0003, "grad_norm": 0.14600848201033748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49971705475, "tokens_per_second": 148040.5523181139, "data_wait_s": 7.499998901039362e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 341500, "loss": 0.6143929120153189, "lr": 0.0003, "grad_norm": 0.19491709326595896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49986554279, "tokens_per_second": 151903.6701715982, "data_wait_s": 1.0818999726325274e-05, "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 sha256:090f57071e9a1afd9af59032c2431f060119672f884ddbe031c4d9862e7ca806
-size 563232
+oid sha256:c5a0258bd7f4643d72b604dd99c4b01ee675edc8cfeca8707a7b2e61199ac802
+size 565350