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": 77462.24075273279,
-      "required_pages_per_second_upper_estimate": 0.005359017552435549
+      "required_token_positions_per_second": 77430.37409841757,
+      "required_pages_per_second_upper_estimate": 0.005356812943349197
     },
     {
       "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": 9682.780094091599,
-      "required_pages_per_second_upper_estimate": 0.004053977825954923
+      "required_token_positions_per_second": 9678.796762302196,
+      "required_pages_per_second_upper_estimate": 0.004052310088116126
     },
     {
       "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": 19365.560188183197,
-      "required_pages_per_second_upper_estimate": 0.00403333548303303
+      "required_token_positions_per_second": 19357.593524604392,
+      "required_pages_per_second_upper_estimate": 0.004031676237104612
     },
     {
       "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": 29048.3402822748,
-      "required_pages_per_second_upper_estimate": 0.0017414036456749574
+      "required_token_positions_per_second": 29036.39028690659,
+      "required_pages_per_second_upper_estimate": 0.0017406872616992197
     },
     {
       "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": 29048.3402822748,
-      "required_pages_per_second_upper_estimate": 0.0022583806868011294
+      "required_token_positions_per_second": 29036.39028690659,
+      "required_pages_per_second_upper_estimate": 0.0022574516272237254
     },
     {
       "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": 29048.3402822748,
-      "required_pages_per_second_upper_estimate": 0.3726009194632547
+      "required_token_positions_per_second": 29036.39028690659,
+      "required_pages_per_second_upper_estimate": 0.37244763775357664
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13411027369542694,
-    "producer_p95_seconds": 0.20028289100082475,
-    "gpu_update_seconds": 0.9517927610091623,
-    "producer_batches_per_second": 7.456550288392255,
-    "gpu_batches_per_second": 1.0506488817373698,
-    "producer_headroom": 7.09709058659253,
+    "producer_batch_seconds": 0.10163664038286413,
+    "producer_p95_seconds": 0.19971602099940355,
+    "gpu_update_seconds": 0.9521844735799457,
+    "producer_batches_per_second": 9.838971420474062,
+    "gpu_batches_per_second": 1.0502166625761933,
+    "producer_headroom": 9.368515822572224,
     "sustainable": true,
-    "buffer_seconds": 60.91473670458639,
+    "buffer_seconds": 60.93980630911653,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540071.0352874454,
+    "generation_bytes_per_second": 529811.9739143956,
     "upload_bytes_per_second": 5386298.614575342,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.8866719624283,
+    "estimated_full_shards_per_hour": 113.68531621049786,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -99,3 +99,4 @@
 {"step": 894900, "loss": 0.5160253047943115, "lr": 3e-05, "grad_norm": 0.3158318015444769, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130988146495, "tokens_per_second": 151723.9355317458, "data_wait_s": 1.266899926122278e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895000, "loss": 0.547240674495697, "lr": 3e-05, "grad_norm": 0.4042976704834889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131002760332, "tokens_per_second": 152778.88040975755, "data_wait_s": 1.3955001122667454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895100, "loss": 0.49131640791893005, "lr": 3e-05, "grad_norm": 0.3079410104055874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131017429681, "tokens_per_second": 142054.74784761734, "data_wait_s": 9.436000254936516e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 895200, "loss": 0.5449494123458862, "lr": 3e-05, "grad_norm": 0.315098659898938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131032076146, "tokens_per_second": 152288.34958202927, "data_wait_s": 1.6501999198226258e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a55cc491078b4d9b9f6fa21bfbbe6621a6c29d51ce38c1489b62d30c56e1c04
-size 215649
+oid sha256:85a86c1b7749696556750ac974fa8ffcdfbc2ee082440c5da3c390e886923a8f
+size 217767