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": 77307.72276015721,
-      "required_pages_per_second_upper_estimate": 0.005348327639178032
+      "required_token_positions_per_second": 77326.46562729405,
+      "required_pages_per_second_upper_estimate": 0.005349624314216002
     },
     {
       "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": 9663.465345019651,
-      "required_pages_per_second_upper_estimate": 0.004045891143856324
+      "required_token_positions_per_second": 9665.808203411756,
+      "required_pages_per_second_upper_estimate": 0.004046872049740652
     },
     {
       "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": 19326.930690039302,
-      "required_pages_per_second_upper_estimate": 0.004025289977298029
+      "required_token_positions_per_second": 19331.616406823512,
+      "required_pages_per_second_upper_estimate": 0.004026265888533519
     },
     {
       "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": 28990.396035058955,
-      "required_pages_per_second_upper_estimate": 0.00173792997653011
+      "required_token_positions_per_second": 28997.424610235266,
+      "required_pages_per_second_upper_estimate": 0.0017383513288799172
     },
     {
       "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": 28990.396035058955,
-      "required_pages_per_second_upper_estimate": 0.0022538757764499055
+      "required_token_positions_per_second": 28997.424610235266,
+      "required_pages_per_second_upper_estimate": 0.0022544222172544287
     },
     {
       "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": 28990.396035058955,
-      "required_pages_per_second_upper_estimate": 0.3718576728756552
+      "required_token_positions_per_second": 28997.424610235266,
+      "required_pages_per_second_upper_estimate": 0.37194782789132086
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.047912293875015166,
-    "producer_p95_seconds": 0.14184625100006087,
-    "gpu_update_seconds": 0.953695146715638,
-    "producer_batches_per_second": 20.87146991143061,
-    "gpu_batches_per_second": 1.0485530973328614,
-    "producer_headroom": 19.905019559352837,
+    "producer_batch_seconds": 0.05272832177342224,
+    "producer_p95_seconds": 0.14888240300024336,
+    "gpu_update_seconds": 0.9534639841857212,
+    "producer_batches_per_second": 18.965139916591294,
+    "gpu_batches_per_second": 1.0488073137382548,
+    "producer_headroom": 18.082577865512793,
     "sustainable": true,
-    "buffer_seconds": 60.082794243085196,
+    "buffer_seconds": 60.06823100370043,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 687800, "loss": 0.5615301728248596, "lr": 8e-05, "grad_norm": 0.20430051949621292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100675435495, "tokens_per_second": 152622.106763418, "data_wait_s": 1.5920999885565834e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687900, "loss": 0.5349336266517639, "lr": 8e-05, "grad_norm": 0.2251875163657823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100690033392, "tokens_per_second": 152880.1175923334, "data_wait_s": 1.6185999811568763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688000, "loss": 0.47442516684532166, "lr": 8e-05, "grad_norm": 0.2050190755562884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100704686333, "tokens_per_second": 153563.69876033106, "data_wait_s": 1.732400005494128e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688100, "loss": 0.5432969927787781, "lr": 8e-05, "grad_norm": 0.19581578127697455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100719252292, "tokens_per_second": 152816.92337104827, "data_wait_s": 1.467199990656809e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c8ce6ae3e2538fb1737a0605822c995c307b02097720cfa80bea81fa455af95
-size 63705
+oid sha256:67409c631317fd5f801c99a62b347606efae4a530ca4c97a78f0e2fc2ccfa254
+size 65823