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": 76316.37602292965,
-      "required_pages_per_second_upper_estimate": 0.005279743971655287
+      "required_token_positions_per_second": 76674.00783042473,
+      "required_pages_per_second_upper_estimate": 0.005304485769917917
     },
     {
       "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": 9539.547002866206,
-      "required_pages_per_second_upper_estimate": 0.0039940091217059185
+      "required_token_positions_per_second": 9584.250978803091,
+      "required_pages_per_second_upper_estimate": 0.004012725742905521
     },
     {
       "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": 19079.09400573241,
-      "required_pages_per_second_upper_estimate": 0.003973672131849789
+      "required_token_positions_per_second": 19168.501957606182,
+      "required_pages_per_second_upper_estimate": 0.003992293450378846
     },
     {
       "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": 28618.641008598614,
-      "required_pages_per_second_upper_estimate": 0.0017156438303308709
+      "required_token_positions_per_second": 28752.752936409273,
+      "required_pages_per_second_upper_estimate": 0.001723683635626063
     },
     {
       "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": 28618.641008598614,
-      "required_pages_per_second_upper_estimate": 0.0022249734583201643
+      "required_token_positions_per_second": 28752.752936409273,
+      "required_pages_per_second_upper_estimate": 0.0022354000708114147
     },
     {
       "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": 28618.641008598614,
-      "required_pages_per_second_upper_estimate": 0.3670891985556703
+      "required_token_positions_per_second": 28752.752936409273,
+      "required_pages_per_second_upper_estimate": 0.36880944236745644
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.30830784196086825,
-    "producer_p95_seconds": 0.3614205930061871,
-    "gpu_update_seconds": 0.9660836093402555,
-    "producer_batches_per_second": 3.2435113996449183,
-    "gpu_batches_per_second": 1.0351070966651699,
-    "producer_headroom": 3.1335031999052263,
+    "producer_batch_seconds": 0.11232317992943308,
+    "producer_p95_seconds": 0.344571679001092,
+    "gpu_update_seconds": 0.96157749002843,
+    "producer_batches_per_second": 8.902881850640703,
+    "gpu_batches_per_second": 1.0399577884985993,
+    "producer_headroom": 8.560810783958752,
     "sustainable": true,
-    "buffer_seconds": 60.8632673884361,
+    "buffer_seconds": 60.579381871791085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329731.63951439736,
+    "generation_bytes_per_second": 316186.9393089774,
     "upload_bytes_per_second": 5348072.927121419,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.75273408006612,
+    "estimated_full_shards_per_hour": 67.84635672046653,
     "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

@@ -400,3 +400,4 @@
 {"step": 480200, "loss": 0.6237391829490662, "lr": 0.0003, "grad_norm": 0.15932251469272685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70285977100, "tokens_per_second": 152124.1980707504, "data_wait_s": 8.499999239575118e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 480300, "loss": 0.5954119563102722, "lr": 0.0003, "grad_norm": 0.17127996462042347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70300565452, "tokens_per_second": 151434.82687737868, "data_wait_s": 1.6024998330976814e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 480400, "loss": 0.5549145340919495, "lr": 0.0003, "grad_norm": 0.14781810418308025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70315062178, "tokens_per_second": 150693.91314482232, "data_wait_s": 1.0480995115358382e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 480500, "loss": 0.5556878447532654, "lr": 0.0003, "grad_norm": 0.1685063295264039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70329520877, "tokens_per_second": 150014.02147498846, "data_wait_s": 8.021997928153723e-06, "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 sha256:4811f250a595a612aee6d42038bb9c27beb085848e15511c8e59347d8186357f
-size 857865
+oid sha256:e8ac145228343fecf0fabd4f994c7e3d1d6a00ba62ee37f5abeba7824f9fc0d1
+size 859983