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": 77635.36988100778,
-      "required_pages_per_second_upper_estimate": 0.0053709950272445356
+      "required_token_positions_per_second": 77487.75283785345,
+      "required_pages_per_second_upper_estimate": 0.0053607825376803895
     },
     {
       "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": 9704.421235125972,
-      "required_pages_per_second_upper_estimate": 0.004063038519787601
+      "required_token_positions_per_second": 9685.969104731681,
+      "required_pages_per_second_upper_estimate": 0.004055312998115811
     },
     {
       "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": 19408.842470251944,
-      "required_pages_per_second_upper_estimate": 0.004042350040957414
+      "required_token_positions_per_second": 19371.938209463362,
+      "required_pages_per_second_upper_estimate": 0.004034663856665956
     },
     {
       "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": 29113.263705377914,
-      "required_pages_per_second_upper_estimate": 0.001745295705757662
+      "required_token_positions_per_second": 29057.907314195043,
+      "required_pages_per_second_upper_estimate": 0.0017419771746305642
     },
     {
       "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": 29113.263705377914,
-      "required_pages_per_second_upper_estimate": 0.002263428197379438
+      "required_token_positions_per_second": 29057.907314195043,
+      "required_pages_per_second_upper_estimate": 0.002259124481452002
     },
     {
       "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": 29113.263705377914,
-      "required_pages_per_second_upper_estimate": 0.3734336874254809
+      "required_token_positions_per_second": 29057.907314195043,
+      "required_pages_per_second_upper_estimate": 0.372723635076449
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05284081613281799,
-    "producer_p95_seconds": 0.1376730189999762,
-    "gpu_update_seconds": 0.9496702355254232,
-    "producer_batches_per_second": 18.924764475371667,
-    "gpu_batches_per_second": 1.0529970958253008,
-    "producer_headroom": 17.97228553658937,
+    "producer_batch_seconds": 0.05700349705466934,
+    "producer_p95_seconds": 0.1404629449998538,
+    "gpu_update_seconds": 0.951479392547092,
+    "producer_batches_per_second": 17.542783367149347,
+    "gpu_batches_per_second": 1.0509949115377257,
+    "producer_headroom": 16.691596861760488,
     "sustainable": true,
-    "buffer_seconds": 60.77889507362708,
+    "buffer_seconds": 60.89468112301389,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 648000, "loss": 0.5908306241035461, "lr": 0.0001, "grad_norm": 0.16779232858622106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94850495306, "tokens_per_second": 154319.66274522644, "data_wait_s": 1.3393000244832365e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648100, "loss": 0.5829259157180786, "lr": 0.0001, "grad_norm": 0.21165426190517386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94865161079, "tokens_per_second": 153921.6830137373, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648200, "loss": 0.5496170520782471, "lr": 0.0001, "grad_norm": 0.17056241102373523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94879908023, "tokens_per_second": 154547.97673710226, "data_wait_s": 1.6008999864425277e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 648300, "loss": 0.5251896977424622, "lr": 0.0001, "grad_norm": 0.17190561853882502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94894560854, "tokens_per_second": 152289.052323069, "data_wait_s": 1.3718999980483204e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4e4326cd5b5d5cbdd2bed095d09b76acebb2421bf01d5cca37dbfa2c2acf1c0
-size 67941
+oid sha256:cc117d4c44141cd6db64cec44a31ff93d652fb68fa317c07dd6bfec20c282a33
+size 70059