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": 75963.54011130637,
-      "required_pages_per_second_upper_estimate": 0.006584327129015799
+      "required_token_positions_per_second": 75130.22880526984,
+      "required_pages_per_second_upper_estimate": 0.006512097817016749
     },
     {
       "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": 9495.442513913296,
-      "required_pages_per_second_upper_estimate": 0.003787873988317096
+      "required_token_positions_per_second": 9391.27860065873,
+      "required_pages_per_second_upper_estimate": 0.003746321445930561
     },
     {
       "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": 18990.885027826593,
-      "required_pages_per_second_upper_estimate": 0.004393207023026501
+      "required_token_positions_per_second": 18782.55720131746,
+      "required_pages_per_second_upper_estimate": 0.004345014046808136
     },
     {
       "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": 28486.327541739887,
-      "required_pages_per_second_upper_estimate": 0.001734667891081932
+      "required_token_positions_per_second": 28173.83580197619,
+      "required_pages_per_second_upper_estimate": 0.001715638783647773
     },
     {
       "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": 28486.327541739887,
-      "required_pages_per_second_upper_estimate": 0.0021929563493647297
+      "required_token_positions_per_second": 28173.83580197619,
+      "required_pages_per_second_upper_estimate": 0.00216889987020522
     },
     {
       "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": 28486.327541739887,
-      "required_pages_per_second_upper_estimate": 0.3537048504630156
+      "required_token_positions_per_second": 28173.83580197619,
+      "required_pages_per_second_upper_estimate": 0.3498247488977264
     }
   ],
   "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": 0.4375525440468948,
-    "producer_p95_seconds": 0.3004658540012315,
-    "gpu_update_seconds": 0.9705708803456142,
-    "producer_batches_per_second": 2.2854398028430265,
-    "gpu_batches_per_second": 1.030321453332606,
-    "producer_headroom": 2.2181813214222634,
+    "producer_batch_seconds": 0.07342292135936646,
+    "producer_p95_seconds": 0.2679734049997933,
+    "gpu_update_seconds": 0.981336023760765,
+    "producer_batches_per_second": 13.619725032534836,
+    "gpu_batches_per_second": 1.0190189453839769,
+    "producer_headroom": 13.365526808142691,
     "sustainable": true,
-    "buffer_seconds": 60.17539458142808,
+    "buffer_seconds": 60.84283347316743,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360303.57940239366,
-    "upload_bytes_per_second": 6449756.368457114,
+    "generation_bytes_per_second": 359667.21072525106,
+    "upload_bytes_per_second": 6411073.52546754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.31276070169312,
+    "estimated_full_shards_per_hour": 77.17621079748295,
     "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": 1.1864866850010003,
-      "bytes": 11535625,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9566465869993408,
       "bytes": 1137762,
@@ -562,6 +556,12 @@
       "bytes": 2508470,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.3975003769992327,
+      "bytes": 10375015,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -210,3 +210,4 @@
 {"step": 61000, "loss": 0.6203593257814646, "lr": 0.0003, "grad_norm": 0.12988660709056526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8927876567, "tokens_per_second": 148434.45286251933, "data_wait_s": 1.029099803417921e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 61100, "loss": 0.5894524808973074, "lr": 0.0003, "grad_norm": 0.1339999396949194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8942332762, "tokens_per_second": 147009.9088900999, "data_wait_s": 8.252001862274483e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 61200, "loss": 0.62699830904603, "lr": 0.0003, "grad_norm": 0.13177566842895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8956860158, "tokens_per_second": 147476.69550709138, "data_wait_s": 2.0006998965982348e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 61300, "loss": 0.6554252617061138, "lr": 0.0003, "grad_norm": 0.14056016071280364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8971511367, "tokens_per_second": 149383.62697999607, "data_wait_s": 1.0161998943658546e-05, "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 sha256:3d244f6f6ae723a36399f99949bc1ab6fa249eff196cbacef1db4157a2842255
-size 452313
+oid sha256:33de6fe31c89eacc5b75ac13e049ece55655616e392960cfa6c33af618fab7c1
+size 454431