Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76603.08111362535,
-      "required_pages_per_second_upper_estimate": 0.005299578895077079
+      "required_token_positions_per_second": 77105.0942351414,
+      "required_pages_per_second_upper_estimate": 0.005334309327654484
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9575.385139203168,
-      "required_pages_per_second_upper_estimate": 0.004009013801004817
+      "required_token_positions_per_second": 9638.136779392675,
+      "required_pages_per_second_upper_estimate": 0.004035286602348905
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19150.770278406337,
-      "required_pages_per_second_upper_estimate": 0.003988600409217344
+      "required_token_positions_per_second": 19276.27355878535,
+      "required_pages_per_second_upper_estimate": 0.004014739432776218
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28726.155417609505,
-      "required_pages_per_second_upper_estimate": 0.0017220891549860767
+      "required_token_positions_per_second": 28914.410338178026,
+      "required_pages_per_second_upper_estimate": 0.0017333747500255397
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28726.155417609505,
-      "required_pages_per_second_upper_estimate": 0.002233332230714854
+      "required_token_positions_per_second": 28914.410338178026,
+      "required_pages_per_second_upper_estimate": 0.0022479682227431715
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28726.155417609505,
-      "required_pages_per_second_upper_estimate": 0.3684682779544837
+      "required_token_positions_per_second": 28914.410338178026,
+      "required_pages_per_second_upper_estimate": 0.3708830099431514
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08303914088287456,
-    "producer_p95_seconds": 0.30308098300156416,
-    "gpu_update_seconds": 0.9624678136723934,
-    "producer_batches_per_second": 12.042513799733124,
-    "gpu_batches_per_second": 1.0389957833336771,
-    "producer_headroom": 11.590531927948767,
+    "producer_batch_seconds": 0.16915779603141345,
+    "producer_p95_seconds": 0.30514907900214894,
+    "gpu_update_seconds": 0.9562014122589289,
+    "producer_batches_per_second": 5.91164003942387,
+    "gpu_batches_per_second": 1.0458047720695176,
+    "producer_headroom": 5.652718554463535,
     "sustainable": true,
-    "buffer_seconds": 60.63547226136079,
+    "buffer_seconds": 60.24068897231252,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300795.9906249822,
+    "generation_bytes_per_second": 311065.315654647,
     "upload_bytes_per_second": 5689494.321895249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.54381741582966,
+    "estimated_full_shards_per_hour": 66.74737550954397,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -313,3 +313,4 @@
 {"step": 428400, "loss": 0.5906609296798706, "lr": 0.0003, "grad_norm": 0.16108718480436882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62706002550, "tokens_per_second": 153828.52408089308, "data_wait_s": 9.537998266750947e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 428500, "loss": 0.6180595755577087, "lr": 0.0003, "grad_norm": 0.16318784764526642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62720771923, "tokens_per_second": 152615.36355919624, "data_wait_s": 1.1893000191776082e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 428600, "loss": 0.6451922655105591, "lr": 0.0003, "grad_norm": 0.1757598669205419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62735335312, "tokens_per_second": 150982.79336658292, "data_wait_s": 1.2710999726550654e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 428700, "loss": 0.6041020154953003, "lr": 0.0003, "grad_norm": 0.17475329121748306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62749992062, "tokens_per_second": 151182.9673968852, "data_wait_s": 1.6968002455541864e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4933a64b1c6e22ec6b250c913f9f44393fa4abf4fcc881f25f588d1741de2637
-size 672033
+oid sha256:f660dbe722625fe4860215682b8921031097b87134d64f55f52a7eb6677f33b3
+size 674151