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": 77343.99730149012,
-      "required_pages_per_second_upper_estimate": 0.005350837196115975
+      "required_token_positions_per_second": 77300.95741342573,
+      "required_pages_per_second_upper_estimate": 0.00534785959679338
     },
     {
       "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": 9667.999662686265,
-      "required_pages_per_second_upper_estimate": 0.004047789567976015
+      "required_token_positions_per_second": 9662.619676678216,
+      "required_pages_per_second_upper_estimate": 0.004045537080181328
     },
     {
       "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": 19335.99932537253,
-      "required_pages_per_second_upper_estimate": 0.0040271787348819445
+      "required_token_positions_per_second": 19325.239353356432,
+      "required_pages_per_second_upper_estimate": 0.004024937716470535
     },
     {
       "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": 29003.998988058796,
-      "required_pages_per_second_upper_estimate": 0.001738745452791944
+      "required_token_positions_per_second": 28987.859030034644,
+      "required_pages_per_second_upper_estimate": 0.0017377778869526852
     },
     {
       "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": 29003.998988058796,
-      "required_pages_per_second_upper_estimate": 0.002254933346212576
+      "required_token_positions_per_second": 28987.859030034644,
+      "required_pages_per_second_upper_estimate": 0.002253678535468371
     },
     {
       "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": 29003.998988058796,
-      "required_pages_per_second_upper_estimate": 0.3720321569510242
+      "required_token_positions_per_second": 28987.859030034644,
+      "required_pages_per_second_upper_estimate": 0.3718251308992271
     }
   ],
   "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": 0.0651110942578157,
-    "producer_p95_seconds": 0.12640841899997213,
-    "gpu_update_seconds": 0.9532478611443521,
-    "producer_batches_per_second": 15.358365750088184,
-    "gpu_batches_per_second": 1.0490451022880063,
-    "producer_headroom": 14.640329301944234,
+    "producer_batch_seconds": 0.05956265846094411,
+    "producer_p95_seconds": 0.11835774199994376,
+    "gpu_update_seconds": 0.9537786137069867,
+    "producer_batches_per_second": 16.789042427575843,
+    "gpu_batches_per_second": 1.0484613364451187,
+    "producer_headroom": 16.013029612041073,
     "sustainable": true,
-    "buffer_seconds": 60.05461525209418,
+    "buffer_seconds": 60.08805266354016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 65,
+    "upload_pages_per_commit": 66,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -11,3 +11,4 @@
 {"step": 926100, "loss": 0.5733742117881775, "lr": 3e-05, "grad_norm": 0.37525392170949484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135554837206, "tokens_per_second": 154986.27622565348, "data_wait_s": 1.0239999937766697e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926200, "loss": 0.5840272903442383, "lr": 3e-05, "grad_norm": 0.34796810006274426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135569601061, "tokens_per_second": 154807.98379555347, "data_wait_s": 8.041999990382465e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 926300, "loss": 0.5264304876327515, "lr": 3e-05, "grad_norm": 0.3554768819932935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135584278242, "tokens_per_second": 154155.39398737994, "data_wait_s": 7.819000074960059e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 926400, "loss": 0.6199615597724915, "lr": 3e-05, "grad_norm": 0.3679225866314959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135598825732, "tokens_per_second": 152803.25678106054, "data_wait_s": 9.728999884828227e-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 sha256:6569823c619d793ad799da869b442c4fdfa6c46ad53ad35c982ddf10ae2351fa
-size 27699
+oid sha256:3adba84d2e34fd6cdfac6ab9364fc008161844dd98ca0d71ba824b355d6092dc
+size 29817