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": 76872.39957130198,
-      "required_pages_per_second_upper_estimate": 0.005318210970884066
+      "required_token_positions_per_second": 76999.34918319326,
+      "required_pages_per_second_upper_estimate": 0.005326993639598387
     },
     {
       "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": 9609.049946412748,
-      "required_pages_per_second_upper_estimate": 0.004023108552782352
+      "required_token_positions_per_second": 9624.918647899158,
+      "required_pages_per_second_upper_estimate": 0.0040297524467185425
     },
     {
       "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": 19218.099892825496,
-      "required_pages_per_second_upper_estimate": 0.004002623392299519
+      "required_token_positions_per_second": 19249.837295798316,
+      "required_pages_per_second_upper_estimate": 0.004009233456367157
     },
     {
       "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": 28827.149839238245,
-      "required_pages_per_second_upper_estimate": 0.0017281436163531668
+      "required_token_positions_per_second": 28874.755943697473,
+      "required_pages_per_second_upper_estimate": 0.001730997529625185
     },
     {
       "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": 28827.149839238245,
-      "required_pages_per_second_upper_estimate": 0.0022411841027689754
+      "required_token_positions_per_second": 28874.755943697473,
+      "required_pages_per_second_upper_estimate": 0.0022448852679935564
     },
     {
       "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": 28827.149839238245,
-      "required_pages_per_second_upper_estimate": 0.3697637259557759
+      "required_token_positions_per_second": 28874.755943697473,
+      "required_pages_per_second_upper_estimate": 0.3703743659483264
     }
   ],
   "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": 0.06951333892970979,
-    "producer_p95_seconds": 0.18227206300070975,
-    "gpu_update_seconds": 0.9590958576961626,
-    "producer_batches_per_second": 14.385728198312785,
-    "gpu_batches_per_second": 1.0426486486993,
-    "producer_headroom": 13.797292324944673,
+    "producer_batch_seconds": 0.05064199409373771,
+    "producer_p95_seconds": 0.16957250799896428,
+    "gpu_update_seconds": 0.9575145865790343,
+    "producer_batches_per_second": 19.74645781422059,
+    "gpu_batches_per_second": 1.0443705130098913,
+    "producer_headroom": 18.90752139038377,
     "sustainable": true,
-    "buffer_seconds": 60.42303903485824,
+    "buffer_seconds": 60.32341895447916,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301520.7081352791,
+    "generation_bytes_per_second": 335875.8569346891,
     "upload_bytes_per_second": 4317809.417392553,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.69932492297917,
+    "estimated_full_shards_per_hour": 72.07114010839943,
     "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-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -73,3 +73,4 @@
 {"step": 532300, "loss": 0.5710451602935791, "lr": 0.0003, "grad_norm": 0.16897770926561098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77911518820, "tokens_per_second": 152876.33175186967, "data_wait_s": 1.3333999959286302e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532400, "loss": 0.568967878818512, "lr": 0.0003, "grad_norm": 0.14396788905138194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77926188670, "tokens_per_second": 153249.58756346416, "data_wait_s": 9.742998372530565e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 532500, "loss": 0.6018000245094299, "lr": 0.0003, "grad_norm": 0.16666585614226395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77940993934, "tokens_per_second": 154926.7151380044, "data_wait_s": 1.0138999641640112e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 532600, "loss": 0.6273928880691528, "lr": 0.0003, "grad_norm": 0.14651706984761576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77955790238, "tokens_per_second": 154786.6819423123, "data_wait_s": 1.268699998036027e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f169624bf198e371e4d92b84c960118783f78bb0ba700fff3eeb56c32f30de63
-size 159798
+oid sha256:e67f8e3efbd82a56573a7aec785f55be375e18a59c5a56f7a3ed92becffcc9d4
+size 161916