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": 75807.49004186563,
-      "required_pages_per_second_upper_estimate": 0.006570801104501961
+      "required_token_positions_per_second": 76087.37832570166,
+      "required_pages_per_second_upper_estimate": 0.006595061111574498
     },
     {
       "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": 9475.936255233204,
-      "required_pages_per_second_upper_estimate": 0.00378009265008505
+      "required_token_positions_per_second": 9510.922290712708,
+      "required_pages_per_second_upper_estimate": 0.0037940491027256694
     },
     {
       "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": 18951.872510466408,
-      "required_pages_per_second_upper_estimate": 0.004384182163732079
+      "required_token_positions_per_second": 19021.844581425416,
+      "required_pages_per_second_upper_estimate": 0.004400368970882055
     },
     {
       "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": 28427.808765699614,
-      "required_pages_per_second_upper_estimate": 0.0017311044046453693
+      "required_token_positions_per_second": 28532.766872138123,
+      "required_pages_per_second_upper_estimate": 0.0017374958026548483
     },
     {
       "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": 28427.808765699614,
-      "required_pages_per_second_upper_estimate": 0.0021884514119948095
+      "required_token_positions_per_second": 28532.766872138123,
+      "required_pages_per_second_upper_estimate": 0.00219653137757108
     },
     {
       "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": 28427.808765699614,
-      "required_pages_per_second_upper_estimate": 0.3529782431143402
+      "required_token_positions_per_second": 28532.766872138123,
+      "required_pages_per_second_upper_estimate": 0.3542814715241209
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10208410401577339,
-    "producer_p95_seconds": 0.24833947200022521,
-    "gpu_update_seconds": 0.9725688049991207,
-    "producer_batches_per_second": 9.79584441320547,
-    "gpu_batches_per_second": 1.0282048888056863,
-    "producer_headroom": 9.527132694908557,
+    "producer_batch_seconds": 0.08090664140618742,
+    "producer_p95_seconds": 0.22481439199691522,
+    "gpu_update_seconds": 0.9689911996231222,
+    "producer_batches_per_second": 12.359924755491383,
+    "gpu_batches_per_second": 1.0320011166137921,
+    "producer_headroom": 11.97665831607512,
     "sustainable": true,
-    "buffer_seconds": 60.29926590994548,
+    "buffer_seconds": 60.077454376633575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313660.5287236144,
-    "upload_bytes_per_second": 5127761.842112498,
+    "generation_bytes_per_second": 356361.56605114666,
+    "upload_bytes_per_second": 5084773.509371153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.30424782067608,
+    "estimated_full_shards_per_hour": 76.4668964018898,
     "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": 0.9423913660011749,
-      "bytes": 5379592,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.2426582910011348,
       "bytes": 7714336,
@@ -562,6 +556,12 @@
       "bytes": 6964981,
       "pages": 21,
       "shards": 3
+    },
+    {
+      "seconds": 0.8043329940010153,
+      "bytes": 1177926,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -178,3 +178,4 @@
 {"step": 254900, "loss": 0.6106672212481499, "lr": 0.0003, "grad_norm": 0.15820504109110067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37310604627, "tokens_per_second": 151035.21677677392, "data_wait_s": 1.3172000763006508e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255000, "loss": 0.6075802594423294, "lr": 0.0003, "grad_norm": 0.1462900680741144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37325264340, "tokens_per_second": 148534.50880404576, "data_wait_s": 1.8441001884639263e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255100, "loss": 0.6156003326177597, "lr": 0.0003, "grad_norm": 0.14337446662061135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37339925108, "tokens_per_second": 132571.0288182981, "data_wait_s": 1.1971998901572078e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 255200, "loss": 0.592559328302741, "lr": 0.0003, "grad_norm": 0.1478354914599657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37354605934, "tokens_per_second": 149664.23562165003, "data_wait_s": 9.801999112823978e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53e4d094418ea1b86cc3fcf0d51a3497ef2755daf9da65137c3c20e8835547bb
-size 384537
+oid sha256:306777079c4941361fa3372c340572c6287d5aeaf3837a0c2b84ff8047d44fca
+size 386655