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": 74918.41585133543,
-      "required_pages_per_second_upper_estimate": 0.006493738407004739
+      "required_token_positions_per_second": 75454.68869670457,
+      "required_pages_per_second_upper_estimate": 0.0065402211780702345
     },
     {
       "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": 9364.80198141693,
-      "required_pages_per_second_upper_estimate": 0.0037357595266542722
+      "required_token_positions_per_second": 9431.836087088072,
+      "required_pages_per_second_upper_estimate": 0.003762500433655685
     },
     {
       "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": 18729.60396283386,
-      "required_pages_per_second_upper_estimate": 0.004332764247030122
+      "required_token_positions_per_second": 18863.672174176143,
+      "required_pages_per_second_upper_estimate": 0.004363778568204227
     },
     {
       "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": 28094.405944250786,
-      "required_pages_per_second_upper_estimate": 0.0017108019220485522
+      "required_token_positions_per_second": 28295.508261264218,
+      "required_pages_per_second_upper_estimate": 0.0017230479980523555
     },
     {
       "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": 28094.405944250786,
-      "required_pages_per_second_upper_estimate": 0.0021627851398815993
+      "required_token_positions_per_second": 28295.508261264218,
+      "required_pages_per_second_upper_estimate": 0.0021782665529321366
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28094.405944250786,
-      "required_pages_per_second_upper_estimate": 0.34883849589941
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28295.508261264218,
+      "required_pages_per_second_upper_estimate": 0.3513355136305576
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05631654773426931,
-    "producer_p95_seconds": 0.1868064440022863,
-    "gpu_update_seconds": 0.9841105042357324,
-    "producer_batches_per_second": 17.756770260821366,
-    "gpu_batches_per_second": 1.01614604833083,
-    "producer_headroom": 17.47462413497497,
+    "producer_batch_seconds": 0.10550399664049337,
+    "producer_p95_seconds": 0.2610298020008486,
+    "gpu_update_seconds": 0.9771162173414415,
+    "producer_batches_per_second": 9.478313920253814,
+    "gpu_batches_per_second": 1.0234197143107717,
+    "producer_headroom": 9.261414244533137,
     "sustainable": true,
-    "buffer_seconds": 60.03074075837968,
+    "buffer_seconds": 60.58120547516937,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436192.70452006755,
-    "upload_bytes_per_second": 5979393.035195284,
+    "generation_bytes_per_second": 436899.68538193393,
+    "upload_bytes_per_second": 5948721.523509742,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.59680034352799,
+    "estimated_full_shards_per_hour": 93.74850197881234,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 5410849,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.4046865109994542,
+      "bytes": 7189609,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -89,3 +89,4 @@
 {"step": 93900, "loss": 0.6738225184381008, "lr": 0.0003, "grad_norm": 0.1397822452711128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13741760457, "tokens_per_second": 151721.3559334485, "data_wait_s": 9.974999557016417e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94000, "loss": 0.6277541741728783, "lr": 0.0003, "grad_norm": 0.14381173255763774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13756436345, "tokens_per_second": 149511.49512423552, "data_wait_s": 1.3134998880559579e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 94100, "loss": 0.5834298804402351, "lr": 0.0003, "grad_norm": 0.13784254449059272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13771106008, "tokens_per_second": 148703.5214312035, "data_wait_s": 1.162699845735915e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 94200, "loss": 0.6251903437077999, "lr": 0.0003, "grad_norm": 0.1315070496077938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13785653862, "tokens_per_second": 148819.15678524968, "data_wait_s": 6.8160006776452065e-06, "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:c28faf7482b4df3c4b2c53d58bf299e20bdf85a0b0d9304fde886a9d22ad07a8
-size 193686
+oid sha256:e1a44019ac451e3a74a2b293db41ed9c96b4dc4a2d62c48be9c3fd1ae1df5e47
+size 195804