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": 76011.06830459811,
-      "required_pages_per_second_upper_estimate": 0.0065884467523512225
+      "required_token_positions_per_second": 76014.04676912409,
+      "required_pages_per_second_upper_estimate": 0.006588704918107485
     },
     {
       "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": 9501.383538074764,
-      "required_pages_per_second_upper_estimate": 0.0037902439516813323
+      "required_token_positions_per_second": 9501.755846140511,
+      "required_pages_per_second_upper_estimate": 0.0037903924709352605
     },
     {
       "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": 19002.76707614953,
-      "required_pages_per_second_upper_estimate": 0.004395955725789104
+      "required_token_positions_per_second": 19003.511692281023,
+      "required_pages_per_second_upper_estimate": 0.004396127979626329
     },
     {
       "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": 28504.150614224294,
-      "required_pages_per_second_upper_estimate": 0.001735753222685798
+      "required_token_positions_per_second": 28505.267538421533,
+      "required_pages_per_second_upper_estimate": 0.0017358212375093616
     },
     {
       "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": 28504.150614224294,
-      "required_pages_per_second_upper_estimate": 0.0021943284188220015
+      "required_token_positions_per_second": 28505.267538421533,
+      "required_pages_per_second_upper_estimate": 0.002194414402738549
     },
     {
       "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": 28504.150614224294,
-      "required_pages_per_second_upper_estimate": 0.3539261533732855
+      "required_token_positions_per_second": 28505.267538421533,
+      "required_pages_per_second_upper_estimate": 0.35394002183371037
     }
   ],
   "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": 0.19919595935922985,
+    "producer_batch_seconds": 0.09376374821076183,
     "producer_p95_seconds": 0.3290863529982744,
-    "gpu_update_seconds": 0.9699640018812891,
-    "producer_batches_per_second": 5.020182152372884,
-    "gpu_batches_per_second": 1.0309660957112374,
-    "producer_headroom": 4.869395970688625,
+    "gpu_update_seconds": 0.9699259957035645,
+    "producer_batches_per_second": 10.66510265515627,
+    "gpu_batches_per_second": 1.0310064937218435,
+    "producer_headroom": 10.344360312083175,
     "sustainable": true,
-    "buffer_seconds": 60.13776811663992,
+    "buffer_seconds": 60.135411733621,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 504548.5576676655,
-    "upload_bytes_per_second": 5636187.997427757,
+    "generation_bytes_per_second": 505013.24698927026,
+    "upload_bytes_per_second": 5704924.515516374,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.26437518618081,
+    "estimated_full_shards_per_hour": 108.36408669718342,
     "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": 0.8660049680001976,
-      "bytes": 1103360,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 0.9722155079998629,
       "bytes": 6032233,
@@ -562,6 +556,12 @@
       "bytes": 7247693,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.8721044880003319,
+      "bytes": 6235096,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 134600, "loss": 0.6050839927047491, "lr": 0.0003, "grad_norm": 0.1471052218670691, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19698342227, "tokens_per_second": 148477.49888373193, "data_wait_s": 1.4622000890085474e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134700, "loss": 0.6082310862839222, "lr": 0.0003, "grad_norm": 0.1380898777569113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19712934063, "tokens_per_second": 148904.92096440226, "data_wait_s": 1.4512999769067392e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134800, "loss": 0.6221358515322208, "lr": 0.0003, "grad_norm": 0.16071592332279125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19727550637, "tokens_per_second": 148767.60806423056, "data_wait_s": 1.568999869050458e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 134900, "loss": 0.6453112438321114, "lr": 0.0003, "grad_norm": 0.1575750167411725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19742319510, "tokens_per_second": 150326.26741005166, "data_wait_s": 8.969000191427767e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd936465ff4b930bd8a53c12ac7c19670a26adc7e0f93ab406848e5398f0fc74
-size 421878
+oid sha256:2b5aa7710976842e000838bfa9bf45e3e2c0f0a10ed5b59a85faf3a95c90d57b
+size 423996