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": 75305.139740342,
-      "required_pages_per_second_upper_estimate": 0.006527258653561092
+      "required_token_positions_per_second": 75130.70525507626,
+      "required_pages_per_second_upper_estimate": 0.006512139114478418
     },
     {
       "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": 9413.14246754275,
-      "required_pages_per_second_upper_estimate": 0.003755043269324537
+      "required_token_positions_per_second": 9391.338156884533,
+      "required_pages_per_second_upper_estimate": 0.0037463452037994785
     },
     {
       "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": 18826.2849350855,
-      "required_pages_per_second_upper_estimate": 0.004355129688433542
+      "required_token_positions_per_second": 18782.676313769065,
+      "required_pages_per_second_upper_estimate": 0.004345041601377507
     },
     {
       "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": 28239.42740262825,
-      "required_pages_per_second_upper_estimate": 0.0017196329679949507
+      "required_token_positions_per_second": 28174.0144706536,
+      "required_pages_per_second_upper_estimate": 0.0017156496636328252
     },
     {
       "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": 28239.42740262825,
-      "required_pages_per_second_upper_estimate": 0.002173949293192587
+      "required_token_positions_per_second": 28174.0144706536,
+      "required_pages_per_second_upper_estimate": 0.0021689136246145937
     },
     {
       "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": 28239.42740262825,
-      "required_pages_per_second_upper_estimate": 0.350639177056859
+      "required_token_positions_per_second": 28174.0144706536,
+      "required_pages_per_second_upper_estimate": 0.34982696736473423
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.42236046854662845,
-    "producer_p95_seconds": 0.3775126930049737,
-    "gpu_update_seconds": 0.979056678657259,
-    "producer_batches_per_second": 2.367645825001258,
-    "gpu_batches_per_second": 1.021391326773302,
-    "producer_headroom": 2.3180594576624576,
+    "producer_batch_seconds": 0.1079035382262532,
+    "producer_p95_seconds": 0.3644019900020794,
+    "gpu_update_seconds": 0.9813298005081419,
+    "producer_batches_per_second": 9.267536694702171,
+    "gpu_batches_per_second": 1.0190254076480614,
+    "producer_headroom": 9.094509935813967,
     "sustainable": true,
-    "buffer_seconds": 60.70151407675006,
+    "buffer_seconds": 60.8424476315048,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329894.98637985304,
+    "generation_bytes_per_second": 343073.2580542373,
     "upload_bytes_per_second": 6060256.6675672885,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.78778451487248,
+    "estimated_full_shards_per_hour": 73.61553484173145,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -282,3 +282,4 @@
 {"step": 113200, "loss": 0.5982640460133553, "lr": 0.0003, "grad_norm": 0.13941891006974785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16564794409, "tokens_per_second": 149804.71258791542, "data_wait_s": 1.2853997759521008e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 113300, "loss": 0.6383073944598436, "lr": 0.0003, "grad_norm": 0.13710993194788226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16579491546, "tokens_per_second": 149803.03924101498, "data_wait_s": 1.314899418503046e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 113400, "loss": 0.6115688942372799, "lr": 0.0003, "grad_norm": 0.13229779074199938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16593919775, "tokens_per_second": 146876.21633626384, "data_wait_s": 1.2367003364488482e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 113500, "loss": 0.659650132060051, "lr": 0.0003, "grad_norm": 0.14444335248204354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16608529611, "tokens_per_second": 148318.61076288426, "data_wait_s": 1.6183999832719564e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ae0b53a8e853fb86d18680bc7b4bfce448ef477a65a492ca005ad72dd2df24e
-size 605592
+oid sha256:659244cb146c08c55d5e86c4a89a314b3448442fa699436572b029a97445b22b
+size 607710