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": 75227.74145594588,
-      "required_pages_per_second_upper_estimate": 0.006520549966433795
+      "required_token_positions_per_second": 74233.44117817623,
+      "required_pages_per_second_upper_estimate": 0.006434366538387736
     },
     {
       "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": 9403.467681993236,
-      "required_pages_per_second_upper_estimate": 0.0037511838527179014
+      "required_token_positions_per_second": 9279.180147272029,
+      "required_pages_per_second_upper_estimate": 0.0037016036968533705
     },
     {
       "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": 18806.93536398647,
-      "required_pages_per_second_upper_estimate": 0.004350653505700602
+      "required_token_positions_per_second": 18558.360294544058,
+      "required_pages_per_second_upper_estimate": 0.004293150038156905
     },
     {
       "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": 28210.403045979707,
-      "required_pages_per_second_upper_estimate": 0.001717865536954085
+      "required_token_positions_per_second": 27837.540441816083,
+      "required_pages_per_second_upper_estimate": 0.0016951601606194167
     },
     {
       "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": 28210.403045979707,
-      "required_pages_per_second_upper_estimate": 0.0021717149178732205
+      "required_token_positions_per_second": 27837.540441816083,
+      "required_pages_per_second_upper_estimate": 0.002143010922454956
     },
     {
       "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": 28210.403045979707,
-      "required_pages_per_second_upper_estimate": 0.35027879168555703
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27837.540441816083,
+      "required_pages_per_second_upper_estimate": 0.3456490860327065
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4609511448828698,
-    "producer_p95_seconds": 0.2774087679972581,
-    "gpu_update_seconds": 0.9800639840181278,
-    "producer_batches_per_second": 2.1694273050435866,
-    "gpu_batches_per_second": 1.0203415453551687,
-    "producer_headroom": 2.1261775676187273,
+    "producer_batch_seconds": 0.09718630888269786,
+    "producer_p95_seconds": 0.26745798800038756,
+    "gpu_update_seconds": 0.9931911929427728,
+    "producer_batches_per_second": 10.289515174477737,
+    "gpu_batches_per_second": 1.0068554847300377,
+    "producer_headroom": 10.219455850942305,
     "sustainable": true,
-    "buffer_seconds": 60.763967009123924,
+    "buffer_seconds": 60.584662769509144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289347.85124829283,
+    "generation_bytes_per_second": 290175.07457322767,
     "upload_bytes_per_second": 5264266.185710132,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.08731320463742,
+    "estimated_full_shards_per_hour": 62.26481607339499,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 214900, "loss": 0.6138138696551323, "lr": 0.0003, "grad_norm": 0.15233410682666723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31454647483, "tokens_per_second": 147255.4126865786, "data_wait_s": 8.671999239595607e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 215000, "loss": 0.5686859451234341, "lr": 0.0003, "grad_norm": 0.1390351501732773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31469240712, "tokens_per_second": 134576.64699822635, "data_wait_s": 1.1709002137649804e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 215100, "loss": 0.5917164199054241, "lr": 0.0003, "grad_norm": 0.1463686432349559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31483816673, "tokens_per_second": 128251.81969456565, "data_wait_s": 1.2594002328114584e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 215200, "loss": 0.5364100411534309, "lr": 0.0003, "grad_norm": 0.1523177941082317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31498437528, "tokens_per_second": 148927.4257286477, "data_wait_s": 1.3799999578623101e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59fa763dc1078fb88c2f5b6e64d8053e860c97e6c088021dd916708fad6dd502
-size 429015
+oid sha256:1e55b20642867dc028bccad2abcd2ddfa60ef6029bf32c9f6c66fc81d4392d7e
+size 431133