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": 75781.13326370512,
-      "required_pages_per_second_upper_estimate": 0.006568516565771652
+      "required_token_positions_per_second": 75412.98556708824,
+      "required_pages_per_second_upper_estimate": 0.006536606456490707
     },
     {
       "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": 9472.64165796314,
-      "required_pages_per_second_upper_estimate": 0.003778778385975403
+      "required_token_positions_per_second": 9426.62319588603,
+      "required_pages_per_second_upper_estimate": 0.0037604209334155215
     },
     {
       "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": 18945.28331592628,
-      "required_pages_per_second_upper_estimate": 0.004382657869541083
+      "required_token_positions_per_second": 18853.24639177206,
+      "required_pages_per_second_upper_estimate": 0.004361366746932546
     },
     {
       "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": 28417.92497388942,
-      "required_pages_per_second_upper_estimate": 0.0017305025335803766
+      "required_token_positions_per_second": 28279.86958765809,
+      "required_pages_per_second_upper_estimate": 0.0017220956848795215
     },
     {
       "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": 28417.92497388942,
-      "required_pages_per_second_upper_estimate": 0.0021876905303411737
+      "required_token_positions_per_second": 28279.86958765809,
+      "required_pages_per_second_upper_estimate": 0.0021770626445473174
     },
     {
       "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": 28417.92497388942,
-      "required_pages_per_second_upper_estimate": 0.35285551949898086
+      "required_token_positions_per_second": 28279.86958765809,
+      "required_pages_per_second_upper_estimate": 0.35114133364364314
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.5560841221875279,
-    "producer_p95_seconds": 0.1928475349996006,
-    "gpu_update_seconds": 0.9729070657130373,
-    "producer_batches_per_second": 1.7982890719234934,
-    "gpu_batches_per_second": 1.0278474021227366,
-    "producer_headroom": 1.749568144268907,
+    "producer_batch_seconds": 0.08223458431250208,
+    "producer_p95_seconds": 0.18160082000031252,
+    "gpu_update_seconds": 0.977656559352245,
+    "producer_batches_per_second": 12.160333859047313,
+    "gpu_batches_per_second": 1.0228540794147167,
+    "producer_headroom": 11.888630161210804,
     "sustainable": true,
-    "buffer_seconds": 60.32023807420831,
+    "buffer_seconds": 60.614706679839195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 65,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378380.8256119398,
+    "generation_bytes_per_second": 378088.62791197986,
     "upload_bytes_per_second": 5886218.270443448,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.19171692150732,
+    "estimated_full_shards_per_hour": 81.12901809711025,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -12,3 +12,4 @@
 {"step": 281400, "loss": 0.5655051227658987, "lr": 0.0003, "grad_norm": 0.13728999638525194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41189387898, "tokens_per_second": 147921.70543950234, "data_wait_s": 1.2397998943924904e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 281500, "loss": 0.5771021479740739, "lr": 0.0003, "grad_norm": 0.14250926453553686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41203896039, "tokens_per_second": 148426.56643852254, "data_wait_s": 9.069000952877104e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 281600, "loss": 0.6916025280952454, "lr": 0.0003, "grad_norm": 0.19433525581574404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41218496383, "tokens_per_second": 149609.8044482704, "data_wait_s": 1.671999962127302e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 281700, "loss": 0.6210511066019535, "lr": 0.0003, "grad_norm": 0.1745023735196812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41233183059, "tokens_per_second": 150141.86411989835, "data_wait_s": 1.2384000001475215e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cbb9f9bca868d08a8cff4f1034d254052d745eb23ecb40b1a9ac2e310095c0ed
-size 29817
+oid sha256:60829e6d47848ff62aa5168dba31dc83b50e6081f144e076b39401ed99fdecae
+size 31935