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": 75764.4407400881,
-      "required_pages_per_second_upper_estimate": 0.006567069700131343
+      "required_token_positions_per_second": 75478.57431205422,
+      "required_pages_per_second_upper_estimate": 0.006542291522671204
     },
     {
       "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": 9470.555092511013,
-      "required_pages_per_second_upper_estimate": 0.003777946023819616
+      "required_token_positions_per_second": 9434.821789006777,
+      "required_pages_per_second_upper_estimate": 0.0037636914747912788
     },
     {
       "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": 18941.110185022026,
-      "required_pages_per_second_upper_estimate": 0.004381692489150979
+      "required_token_positions_per_second": 18869.643578013554,
+      "required_pages_per_second_upper_estimate": 0.004365159947388894
     },
     {
       "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": 28411.665277533037,
-      "required_pages_per_second_upper_estimate": 0.0017301213509143614
+      "required_token_positions_per_second": 28304.46536702033,
+      "required_pages_per_second_upper_estimate": 0.0017235934387986008
     },
     {
       "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": 28411.665277533037,
-      "required_pages_per_second_upper_estimate": 0.002187208641587711
+      "required_token_positions_per_second": 28304.46536702033,
+      "required_pages_per_second_upper_estimate": 0.0021789560957280974
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28411.665277533037,
-      "required_pages_per_second_upper_estimate": 0.3527777950200906
+      "required_token_positions_per_second": 28304.46536702033,
+      "required_pages_per_second_upper_estimate": 0.35144673090654394
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07240278460128025,
-    "producer_p95_seconds": 0.2475747329990554,
-    "gpu_update_seconds": 0.9731214179079846,
-    "producer_batches_per_second": 13.811623482535474,
-    "gpu_batches_per_second": 1.0276209952811428,
-    "producer_headroom": 13.440386626936135,
+    "producer_batch_seconds": 0.08683917310148104,
+    "producer_p95_seconds": 0.25813985700006015,
+    "gpu_update_seconds": 0.9768070034707235,
+    "producer_batches_per_second": 11.515540329148356,
+    "gpu_batches_per_second": 1.0237436837029923,
+    "producer_headroom": 11.248460442261674,
     "sustainable": true,
-    "buffer_seconds": 60.33352791029505,
+    "buffer_seconds": 60.56203421518485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352665.7753389788,
-    "upload_bytes_per_second": 6422715.098092522,
+    "generation_bytes_per_second": 350409.2671027314,
+    "upload_bytes_per_second": 6515651.99156485,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.67386574866316,
+    "estimated_full_shards_per_hour": 75.1896716099878,
     "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": 0.9868076469992957,
-      "bytes": 927962,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.2888746650005487,
       "bytes": 18374728,
@@ -562,6 +556,12 @@
       "bytes": 4694947,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.8593272979996982,
+      "bytes": 6225685,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -201,3 +201,4 @@
 {"step": 60100, "loss": 0.5657984018325806, "lr": 0.0003, "grad_norm": 0.13197050722056766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8796266252, "tokens_per_second": 121109.6552948848, "data_wait_s": 1.123200127040036e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 60200, "loss": 0.6868634317070246, "lr": 0.0003, "grad_norm": 0.1663107885689551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8810790000, "tokens_per_second": 148013.39657406998, "data_wait_s": 1.0452997230459005e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 60300, "loss": 0.6200925186276436, "lr": 0.0003, "grad_norm": 0.16018672564577316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8825286552, "tokens_per_second": 148037.3266823668, "data_wait_s": 1.9115999748464674e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 60400, "loss": 0.6006607227027416, "lr": 0.0003, "grad_norm": 0.1277029163917345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8839879911, "tokens_per_second": 149213.90652760616, "data_wait_s": 9.121002221945673e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bfc6574eeb0e8022a252a6e6a8c4ac0211899b5da2600445aa659b0d106dd57
-size 433251
+oid sha256:823eea114f31e800003104aa48bb27e378e1d2e38de6a7db04e6955ab731e978
+size 435369