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": 76968.04134411908,
-      "required_pages_per_second_upper_estimate": 0.005324827690647039
+      "required_token_positions_per_second": 76911.82845494506,
+      "required_pages_per_second_upper_estimate": 0.005320938752542108
     },
     {
       "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": 9621.005168014884,
-      "required_pages_per_second_upper_estimate": 0.004028113954413751
+      "required_token_positions_per_second": 9613.978556868133,
+      "required_pages_per_second_upper_estimate": 0.004025172059058932
     },
     {
       "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": 19242.01033602977,
-      "required_pages_per_second_upper_estimate": 0.004007603307058179
+      "required_token_positions_per_second": 19227.957113736265,
+      "required_pages_per_second_upper_estimate": 0.004004676391462836
     },
     {
       "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": 28863.015504044655,
-      "required_pages_per_second_upper_estimate": 0.0017302937081946118
+      "required_token_positions_per_second": 28841.9356706044,
+      "required_pages_per_second_upper_estimate": 0.0017290300043669104
     },
     {
       "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": 28863.015504044655,
-      "required_pages_per_second_upper_estimate": 0.002243972500451286
+      "required_token_positions_per_second": 28841.9356706044,
+      "required_pages_per_second_upper_estimate": 0.002242333636121695
     },
     {
       "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": 28863.015504044655,
-      "required_pages_per_second_upper_estimate": 0.370223772194362
+      "required_token_positions_per_second": 28841.9356706044,
+      "required_pages_per_second_upper_estimate": 0.3699533827247521
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11375020768036848,
-    "producer_p95_seconds": 0.34670877100143116,
-    "gpu_update_seconds": 0.9579040691755027,
-    "producer_batches_per_second": 8.791192740587713,
-    "gpu_batches_per_second": 1.043945873265504,
-    "producer_headroom": 8.42111929911511,
+    "producer_batch_seconds": 0.4839515449065175,
+    "producer_p95_seconds": 0.38270848899992416,
+    "gpu_update_seconds": 0.9586041767709352,
+    "producer_batches_per_second": 2.0663225699448176,
+    "gpu_batches_per_second": 1.0431834371601707,
+    "producer_headroom": 1.980785446105155,
     "sustainable": true,
-    "buffer_seconds": 60.34795635805667,
+    "buffer_seconds": 60.39206313656892,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326688.6941799735,
+    "generation_bytes_per_second": 306864.18088396674,
     "upload_bytes_per_second": 5159758.16617691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.09978884744076,
+    "estimated_full_shards_per_hour": 65.84590978516819,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 963900, "loss": 0.4930093288421631, "lr": 3e-05, "grad_norm": 0.35177214582856164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141089051260, "tokens_per_second": 153172.66581548544, "data_wait_s": 1.8566002836450934e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964000, "loss": 0.5064939260482788, "lr": 3e-05, "grad_norm": 0.3384078411223471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141103593687, "tokens_per_second": 150458.0840729264, "data_wait_s": 1.1809999705292284e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964100, "loss": 0.520802915096283, "lr": 3e-05, "grad_norm": 0.35308641166995997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141118071464, "tokens_per_second": 150352.54709681298, "data_wait_s": 1.6137993952725083e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 964200, "loss": 0.4938688278198242, "lr": 3e-05, "grad_norm": 0.3483320897937832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141132599406, "tokens_per_second": 151650.17021092563, "data_wait_s": 2.3651002265978605e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:650edfe1eb42a167fa7da0aaeeba5278307b78da8b46b78c970ae81f33d93d54
-size 833784
+oid sha256:d38849919f337605585a28e993788168f9e9d3d644b7959d21ec12c24a18c513
+size 835902