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": 75631.06268419111,
-      "required_pages_per_second_upper_estimate": 0.006555508828289784
+      "required_token_positions_per_second": 75415.18761914835,
+      "required_pages_per_second_upper_estimate": 0.006536797324782227
     },
     {
       "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": 9453.88283552389,
-      "required_pages_per_second_upper_estimate": 0.0037712952112349965
+      "required_token_positions_per_second": 9426.898452393543,
+      "required_pages_per_second_upper_estimate": 0.0037605307373518204
     },
     {
       "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": 18907.76567104778,
-      "required_pages_per_second_upper_estimate": 0.0043739788226864395
+      "required_token_positions_per_second": 18853.796904787087,
+      "required_pages_per_second_upper_estimate": 0.004361494098429874
     },
     {
       "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": 28361.64850657167,
-      "required_pages_per_second_upper_estimate": 0.0017270755919805306
+      "required_token_positions_per_second": 28280.695357180633,
+      "required_pages_per_second_upper_estimate": 0.0017221459699109667
     },
     {
       "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": 28361.64850657167,
-      "required_pages_per_second_upper_estimate": 0.0021833582121038213
+      "required_token_positions_per_second": 28280.695357180633,
+      "required_pages_per_second_upper_estimate": 0.002177126214571039
     },
     {
       "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": 28361.64850657167,
-      "required_pages_per_second_upper_estimate": 0.3521567541201146
+      "required_token_positions_per_second": 28280.695357180633,
+      "required_pages_per_second_upper_estimate": 0.35115158693744036
     }
   ],
   "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": 0.10986396203907134,
-    "producer_p95_seconds": 0.3852085789985722,
-    "gpu_update_seconds": 0.9748375519707075,
-    "producer_batches_per_second": 9.10216581889124,
-    "gpu_batches_per_second": 1.0258119396184777,
-    "producer_headroom": 8.873133044519388,
+    "producer_batch_seconds": 0.49201269036689155,
+    "producer_p95_seconds": 0.3861126480042003,
+    "gpu_update_seconds": 0.9776280127065551,
+    "producer_batches_per_second": 2.0324679008874846,
+    "gpu_batches_per_second": 1.0228839466572857,
+    "producer_headroom": 1.9869975548344951,
     "sustainable": true,
-    "buffer_seconds": 60.439928222183866,
+    "buffer_seconds": 60.61293678780642,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353916.43879189505,
+    "generation_bytes_per_second": 314134.54637670214,
     "upload_bytes_per_second": 3167593.213277648,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.94222901170386,
+    "estimated_full_shards_per_hour": 67.40596097446249,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 347900, "loss": 0.5391279589384794, "lr": 0.0003, "grad_norm": 0.13975359883808136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50923617558, "tokens_per_second": 148858.53260996024, "data_wait_s": 1.3570002920459956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348000, "loss": 0.5806275233626366, "lr": 0.0003, "grad_norm": 0.1628066548025042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50938233409, "tokens_per_second": 149651.37517678246, "data_wait_s": 1.3744000170845538e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348100, "loss": 0.5532508082687855, "lr": 0.0003, "grad_norm": 0.1465484121571448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50952768903, "tokens_per_second": 145933.23827819858, "data_wait_s": 1.0547999409027398e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 348200, "loss": 0.5723830554634333, "lr": 0.0003, "grad_norm": 0.14796907512204213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50967406095, "tokens_per_second": 148289.79851081828, "data_wait_s": 1.0226998711004853e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:196c13d80b840b6c0fd8867f03049eea9c43ba5a5a7de5fc3b796ed678b464d4
-size 705921
+oid sha256:4079836850b94765608ab8ca6fc1820e7e9f37cbf4bba2b42f3241e4e43b0250
+size 708039