Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74408.26279429169,
-      "required_pages_per_second_upper_estimate": 0.006449519632991289
+      "required_token_positions_per_second": 74739.14368928339,
+      "required_pages_per_second_upper_estimate": 0.0064781995503592074
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9301.032849286461,
-      "required_pages_per_second_upper_estimate": 0.0037103210664139385
+      "required_token_positions_per_second": 9342.392961160424,
+      "required_pages_per_second_upper_estimate": 0.0037268202334292417
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18602.065698572922,
-      "required_pages_per_second_upper_estimate": 0.0043032605141901424
+      "required_token_positions_per_second": 18684.785922320847,
+      "required_pages_per_second_upper_estimate": 0.00432239638212801
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27903.098547859387,
-      "required_pages_per_second_upper_estimate": 0.001699152305320654
+      "required_token_positions_per_second": 28027.178883481272,
+      "required_pages_per_second_upper_estimate": 0.0017067081467608176
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27903.098547859387,
-      "required_pages_per_second_upper_estimate": 0.0021480577669346222
+      "required_token_positions_per_second": 28027.178883481272,
+      "required_pages_per_second_upper_estimate": 0.0021576098146471475
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27903.098547859387,
-      "required_pages_per_second_upper_estimate": 0.34646309830089755
+      "required_token_positions_per_second": 28027.178883481272,
+      "required_pages_per_second_upper_estimate": 0.3480037607991516
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08882598783586104,
-    "producer_p95_seconds": 0.3085070839988475,
-    "gpu_update_seconds": 0.9908576982078947,
-    "producer_batches_per_second": 11.257966551949536,
-    "gpu_batches_per_second": 1.009226654653479,
-    "producer_headroom": 11.155042824166186,
+    "producer_batch_seconds": 0.10241411964079816,
+    "producer_p95_seconds": 0.2988990100020601,
+    "gpu_update_seconds": 0.9864710292442329,
+    "producer_batches_per_second": 9.764278631768224,
+    "gpu_batches_per_second": 1.013714514014803,
+    "producer_headroom": 9.632177991707872,
     "sustainable": true,
-    "buffer_seconds": 60.44231959068158,
+    "buffer_seconds": 60.17473278389821,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295077.8278492008,
-    "upload_bytes_per_second": 2785879.6680428805,
+    "generation_bytes_per_second": 306070.5163131273,
+    "upload_bytes_per_second": 2751869.433531144,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.31683279616373,
+    "estimated_full_shards_per_hour": 65.67560784383167,
     "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": [
-    {
-      "seconds": 1.1623286959984398,
-      "bytes": 7130450,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.6525030769989826,
       "bytes": 8063307,
@@ -562,6 +556,12 @@
       "bytes": 11511318,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.8884790899974178,
+      "bytes": 1189196,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 222300, "loss": 0.5637040808796883, "lr": 0.0003, "grad_norm": 0.1353372322068904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32537770694, "tokens_per_second": 141829.89712486847, "data_wait_s": 1.053700179909356e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 222400, "loss": 0.5519446395337582, "lr": 0.0003, "grad_norm": 0.13743435066114967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32552387626, "tokens_per_second": 148368.14760780317, "data_wait_s": 7.335002010222524e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 222500, "loss": 0.5978420991450548, "lr": 0.0003, "grad_norm": 0.1403259231274147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32566965062, "tokens_per_second": 148842.6613286359, "data_wait_s": 1.2837001122534275e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 222600, "loss": 0.5505469087511301, "lr": 0.0003, "grad_norm": 0.13772615889590778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32581643509, "tokens_per_second": 148935.62317157237, "data_wait_s": 1.1741001799236983e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99cea2146d007b69131d8f8525ef44a15ba5c06ddb365f57e5f3267405321696
-size 586530
+oid sha256:61069ba662ce8d4e6f2869f166e3c90c92ccbce1dc168c394ad86eecc4aa6508
+size 588648