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": 75665.69836686787,
-      "required_pages_per_second_upper_estimate": 0.00655851096148087
+      "required_token_positions_per_second": 75962.00820263596,
+      "required_pages_per_second_upper_estimate": 0.006584194347054835
     },
     {
       "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": 9458.212295858484,
-      "required_pages_per_second_upper_estimate": 0.003773022297693667
+      "required_token_positions_per_second": 9495.251025329495,
+      "required_pages_per_second_upper_estimate": 0.003787797600658008
     },
     {
       "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": 18916.42459171697,
-      "required_pages_per_second_upper_estimate": 0.004375981911591458
+      "required_token_positions_per_second": 18990.50205065899,
+      "required_pages_per_second_upper_estimate": 0.004393118427998945
     },
     {
       "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": 28374.636887575452,
-      "required_pages_per_second_upper_estimate": 0.0017278665162388923
+      "required_token_positions_per_second": 28485.753075988483,
+      "required_pages_per_second_upper_estimate": 0.0017346329091316604
     },
     {
       "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": 28374.636887575452,
-      "required_pages_per_second_upper_estimate": 0.002184358093627632
+      "required_token_positions_per_second": 28485.753075988483,
+      "required_pages_per_second_upper_estimate": 0.0021929121254009635
     },
     {
       "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": 28374.636887575452,
-      "required_pages_per_second_upper_estimate": 0.3523180263428667
+      "required_token_positions_per_second": 28485.753075988483,
+      "required_pages_per_second_upper_estimate": 0.3536977175209963
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.13129274778125932,
-    "producer_p95_seconds": 0.1682472989996313,
-    "gpu_update_seconds": 0.9743913238271736,
-    "producer_batches_per_second": 7.61656692314836,
-    "gpu_batches_per_second": 1.026281716130478,
-    "producer_headroom": 7.421516727264794,
+    "producer_batch_seconds": 0.06604437228899229,
+    "producer_p95_seconds": 0.17501575400001457,
+    "gpu_update_seconds": 0.9705904536294443,
+    "producer_batches_per_second": 15.141335519463652,
+    "gpu_batches_per_second": 1.0303006754914816,
+    "producer_headroom": 14.696035710391845,
     "sustainable": true,
-    "buffer_seconds": 60.412262077284765,
+    "buffer_seconds": 60.176608125025545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351443.47014520573,
-    "upload_bytes_per_second": 5528607.3444076935,
+    "generation_bytes_per_second": 350497.3563680549,
+    "upload_bytes_per_second": 5442543.546901493,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.41158750788811,
+    "estimated_full_shards_per_hour": 75.20857351571308,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 6780784,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.4607127819999732,
+      "bytes": 6008702,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 359800, "loss": 0.5995389968156815, "lr": 0.0003, "grad_norm": 0.15052040963303098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52664547186, "tokens_per_second": 147907.74579080864, "data_wait_s": 1.2000999959127512e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359900, "loss": 0.589072197675705, "lr": 0.0003, "grad_norm": 0.1709890691884968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52679165005, "tokens_per_second": 148789.96262994417, "data_wait_s": 8.371000149054453e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 360000, "loss": 0.6037846226245165, "lr": 0.0003, "grad_norm": 0.14638077382152287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52693835160, "tokens_per_second": 149850.34213180692, "data_wait_s": 9.698000212665647e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 360100, "loss": 0.6022424828261137, "lr": 0.0003, "grad_norm": 0.18071048781891982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52708490102, "tokens_per_second": 127690.04584187643, "data_wait_s": 1.3590999515145086e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:723c0eb296df1f08f8c761bc6c5c24fd9083f155cb9624eb7041d9d5d2e16498
-size 106848
+oid sha256:2545948532812d726755c01d62362e65ba0bb3f151dbc004f83f296181165ca8
+size 108966