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": 76678.69646517854,
-      "required_pages_per_second_upper_estimate": 0.005304810140549317
+      "required_token_positions_per_second": 77287.5206506093,
+      "required_pages_per_second_upper_estimate": 0.005346930010364155
     },
     {
       "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": 9584.837058147317,
-      "required_pages_per_second_upper_estimate": 0.0040129711220882195
+      "required_token_positions_per_second": 9660.940081326162,
+      "required_pages_per_second_upper_estimate": 0.004044833868681362
     },
     {
       "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": 19169.674116294635,
-      "required_pages_per_second_upper_estimate": 0.0039925375801217475
+      "required_token_positions_per_second": 19321.880162652324,
+      "required_pages_per_second_upper_estimate": 0.004024238085634685
     },
     {
       "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": 28754.511174441956,
-      "required_pages_per_second_upper_estimate": 0.0017237890393114483
+      "required_token_positions_per_second": 28982.820243978484,
+      "required_pages_per_second_upper_estimate": 0.0017374758194223991
     },
     {
       "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": 28754.511174441956,
-      "required_pages_per_second_upper_estimate": 0.002235536766084784
+      "required_token_positions_per_second": 28982.820243978484,
+      "required_pages_per_second_upper_estimate": 0.002253286792015782
     },
     {
       "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": 28754.511174441956,
-      "required_pages_per_second_upper_estimate": 0.3688319951570908
+      "required_token_positions_per_second": 28982.820243978484,
+      "required_pages_per_second_upper_estimate": 0.37176049876192563
     }
   ],
   "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": 0.06254846602342212,
-    "producer_p95_seconds": 0.13980121199983842,
-    "gpu_update_seconds": 0.961518692919897,
-    "producer_batches_per_second": 15.98760231187023,
-    "gpu_batches_per_second": 1.0400213821774433,
-    "producer_headroom": 15.372378477832585,
+    "producer_batch_seconds": 0.054749620218764505,
+    "producer_p95_seconds": 0.1299775350003074,
+    "gpu_update_seconds": 0.9539444321587093,
+    "producer_batches_per_second": 18.26496688021348,
+    "gpu_batches_per_second": 1.0482790886855644,
+    "producer_headroom": 17.42376345894288,
     "sustainable": true,
-    "buffer_seconds": 60.57567765395351,
+    "buffer_seconds": 60.09849922599869,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 927600, "loss": 0.5498512983322144, "lr": 3e-05, "grad_norm": 0.3315312814857236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135774307827, "tokens_per_second": 153870.60190463637, "data_wait_s": 1.0039000244432827e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927700, "loss": 0.5500040054321289, "lr": 3e-05, "grad_norm": 0.3361282329756592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135788903980, "tokens_per_second": 152546.01956672015, "data_wait_s": 6.966000000829808e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 927800, "loss": 0.5746034383773804, "lr": 3e-05, "grad_norm": 0.3320251980846008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135803567729, "tokens_per_second": 153915.41208555488, "data_wait_s": 1.2613999842869816e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 927900, "loss": 0.555435299873352, "lr": 3e-05, "grad_norm": 0.333205786955412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135818387398, "tokens_per_second": 155677.48903948313, "data_wait_s": 9.808999948290875e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f46d92780d6c61b4536649fdf7e0c1245996a9ad251424370714b0b004c8fe37
-size 59469
+oid sha256:45756f991c03e160070bc87d0703acb84a3acaea4c32d1593412008137daf040
+size 61587