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": 76068.09922458114,
-      "required_pages_per_second_upper_estimate": 0.005262567606663996
+      "required_token_positions_per_second": 75733.07042625574,
+      "required_pages_per_second_upper_estimate": 0.005239389537021933
     },
     {
       "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": 9508.512403072642,
-      "required_pages_per_second_upper_estimate": 0.003981015582848493
+      "required_token_positions_per_second": 9466.633803281968,
+      "required_pages_per_second_upper_estimate": 0.003963481887640746
     },
     {
       "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": 19017.024806145284,
-      "required_pages_per_second_upper_estimate": 0.003960744754450659
+      "required_token_positions_per_second": 18933.267606563935,
+      "required_pages_per_second_upper_estimate": 0.003943300338603754
     },
     {
       "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": 28525.537209217924,
-      "required_pages_per_second_upper_estimate": 0.0017100623997192688
+      "required_token_positions_per_second": 28399.9014098459,
+      "required_pages_per_second_upper_estimate": 0.001702530725381674
     },
     {
       "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": 28525.537209217924,
-      "required_pages_per_second_upper_estimate": 0.0022177350474386496
+      "required_token_positions_per_second": 28399.9014098459,
+      "required_pages_per_second_upper_estimate": 0.0022079674166509545
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28525.537209217924,
-      "required_pages_per_second_upper_estimate": 0.36589496298428603
+      "required_token_positions_per_second": 28399.9014098459,
+      "required_pages_per_second_upper_estimate": 0.36428344184715306
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11714792945292629,
-    "producer_p95_seconds": 0.29625767200195696,
-    "gpu_update_seconds": 0.9692367858742954,
-    "producer_batches_per_second": 8.536215745937117,
-    "gpu_batches_per_second": 1.031739627069514,
-    "producer_headroom": 8.273614313121643,
+    "producer_batch_seconds": 0.09275898617963207,
+    "producer_p95_seconds": 0.27953707499909797,
+    "gpu_update_seconds": 0.9735245063356021,
+    "producer_batches_per_second": 10.780626666870353,
+    "gpu_batches_per_second": 1.0271955081686162,
+    "producer_headroom": 10.495204253853387,
     "sustainable": true,
-    "buffer_seconds": 60.09268072420632,
+    "buffer_seconds": 60.35851939280733,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314175.1220160655,
+    "generation_bytes_per_second": 313045.1883312096,
     "upload_bytes_per_second": 5184345.82673646,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.41466756211733,
+    "estimated_full_shards_per_hour": 67.17221009685723,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 626700, "loss": 0.5207452178001404, "lr": 0.0001, "grad_norm": 0.15020049541035962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91732187394, "tokens_per_second": 150719.6849077421, "data_wait_s": 9.079998562810943e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 626800, "loss": 0.5304588079452515, "lr": 0.0001, "grad_norm": 0.14932174681587979, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91746832525, "tokens_per_second": 150063.39197607973, "data_wait_s": 1.4050998288439587e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626900, "loss": 0.5559431910514832, "lr": 0.0001, "grad_norm": 0.16029947237022812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91761452094, "tokens_per_second": 151988.67797562748, "data_wait_s": 1.8910999642685056e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 627000, "loss": 0.47953253984451294, "lr": 0.0001, "grad_norm": 0.175877610725448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91776072408, "tokens_per_second": 151197.93908179356, "data_wait_s": 1.3818000297760591e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:819f23231a7625dc9f822de6800a433805275506379fd2bd079e250f7429d8d7
-size 480630
+oid sha256:7fca5047860c8b07e8f62c4910faa40039ac006b273b0d61d648e41e87b88a0d
+size 482748