Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77183.69050169231,
-      "required_pages_per_second_upper_estimate": 0.00533974679974294
+      "required_token_positions_per_second": 77511.68334679016,
+      "required_pages_per_second_upper_estimate": 0.005362438105815062
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9647.961312711539,
-      "required_pages_per_second_upper_estimate": 0.004039399929289928
+      "required_token_positions_per_second": 9688.96041834877,
+      "required_pages_per_second_upper_estimate": 0.004056565398661554
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19295.922625423078,
-      "required_pages_per_second_upper_estimate": 0.004018831815176124
+      "required_token_positions_per_second": 19377.92083669754,
+      "required_pages_per_second_upper_estimate": 0.004035909880146345
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28943.88393813462,
-      "required_pages_per_second_upper_estimate": 0.001735141647339356
+      "required_token_positions_per_second": 29066.88125504631,
+      "required_pages_per_second_upper_estimate": 0.0017425151486821834
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28943.88393813462,
-      "required_pages_per_second_upper_estimate": 0.002250259665499131
+      "required_token_positions_per_second": 29066.88125504631,
+      "required_pages_per_second_upper_estimate": 0.0022598221658809937
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28943.88393813462,
-      "required_pages_per_second_upper_estimate": 0.37126106563710853
+      "required_token_positions_per_second": 29066.88125504631,
+      "required_pages_per_second_upper_estimate": 0.3728387431542221
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04815776383593828,
-    "producer_p95_seconds": 0.16547565099972417,
-    "gpu_update_seconds": 0.9552277109421641,
-    "producer_batches_per_second": 20.765083765241993,
-    "gpu_batches_per_second": 1.0468708021605402,
-    "producer_headroom": 19.8353834325944,
+    "producer_batch_seconds": 0.07033839270314957,
+    "producer_p95_seconds": 0.16545128200050385,
+    "gpu_update_seconds": 0.9511856383009795,
+    "producer_batches_per_second": 14.216986791556621,
+    "gpu_batches_per_second": 1.0513194898381912,
+    "producer_headroom": 13.52299365604338,
     "sustainable": true,
-    "buffer_seconds": 60.17934578935634,
+    "buffer_seconds": 60.875880851262686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451354.77205289196,
+    "generation_bytes_per_second": 475133.1842037734,
     "upload_bytes_per_second": 3914007.6726365685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.85022708120412,
+    "estimated_full_shards_per_hour": 101.95252079567815,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 445200, "loss": 0.6336721777915955, "lr": 0.0003, "grad_norm": 0.17824873338584174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65163980123, "tokens_per_second": 153330.52134395167, "data_wait_s": 1.343100029771449e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 445300, "loss": 0.5626630783081055, "lr": 0.0003, "grad_norm": 0.16750663043162037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65178632207, "tokens_per_second": 152824.15368629096, "data_wait_s": 9.791000593395438e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 445400, "loss": 0.5859872698783875, "lr": 0.0003, "grad_norm": 0.14780479011233563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65193310939, "tokens_per_second": 152781.65016439898, "data_wait_s": 1.4901000213285442e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 445500, "loss": 0.5570152997970581, "lr": 0.0003, "grad_norm": 0.15834351221982682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65207806873, "tokens_per_second": 150562.27444808307, "data_wait_s": 1.4928000382496975e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8848dcfe95174b84eb4600ae9b2a245b8331a9d79fb295003320f37b53f5ded3
-size 111084
+oid sha256:9bef00d82d337b15727c5767cf29401dee81e477f4a713254e857e9d354dc9e7
+size 113202