Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75648.29900775228,
-      "required_pages_per_second_upper_estimate": 0.006557002829131004
+      "required_token_positions_per_second": 75734.17276168206,
+      "required_pages_per_second_upper_estimate": 0.006564446148476596
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9456.037375969036,
-      "required_pages_per_second_upper_estimate": 0.003772154689631816
+      "required_token_positions_per_second": 9466.771595210257,
+      "required_pages_per_second_upper_estimate": 0.0037764367301780185
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18912.07475193807,
-      "required_pages_per_second_upper_estimate": 0.004374975652713176
+      "required_token_positions_per_second": 18933.543190420514,
+      "required_pages_per_second_upper_estimate": 0.0043799419981244755
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28368.112127907105,
-      "required_pages_per_second_upper_estimate": 0.0017274691926078584
+      "required_token_positions_per_second": 28400.31478563077,
+      "required_pages_per_second_upper_estimate": 0.001729430165508942
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28368.112127907105,
-      "required_pages_per_second_upper_estimate": 0.002183855799566672
+      "required_token_positions_per_second": 28400.31478563077,
+      "required_pages_per_second_upper_estimate": 0.0021863348493009016
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28368.112127907105,
-      "required_pages_per_second_upper_estimate": 0.35223701066475166
+      "required_token_positions_per_second": 28400.31478563077,
+      "required_pages_per_second_upper_estimate": 0.35263685989831717
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.041225877039096304,
-    "producer_p95_seconds": 0.14148862000001827,
-    "gpu_update_seconds": 0.9746154370562187,
-    "producer_batches_per_second": 24.25660948466072,
-    "gpu_batches_per_second": 1.0260457222188624,
-    "producer_headroom": 23.64086605439463,
+    "producer_batch_seconds": 0.04027699835937426,
+    "producer_p95_seconds": 0.1380104609997943,
+    "gpu_update_seconds": 0.9735103363709403,
+    "producer_batches_per_second": 24.828066656741196,
+    "gpu_batches_per_second": 1.027210459549724,
+    "producer_headroom": 24.17037952244425,
     "sustainable": true,
-    "buffer_seconds": 60.42615709748556,
+    "buffer_seconds": 60.3576408549983,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 89100, "loss": 0.5991051606833935, "lr": 0.0003, "grad_norm": 0.13633313821056686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13039297606, "tokens_per_second": 149385.12140472018, "data_wait_s": 1.4216999261407182e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 89200, "loss": 0.6012036129832268, "lr": 0.0003, "grad_norm": 0.18671982782268762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13053837626, "tokens_per_second": 148796.3517379957, "data_wait_s": 1.532500027678907e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 89300, "loss": 0.607091560959816, "lr": 0.0003, "grad_norm": 0.12861040424271264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13068424989, "tokens_per_second": 149216.16988226844, "data_wait_s": 1.3524000678444281e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 89400, "loss": 0.5994367077946663, "lr": 0.0003, "grad_norm": 0.14024797043852974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13083191395, "tokens_per_second": 150982.8180724597, "data_wait_s": 8.67000017024111e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb4ed316ed352cce49bdfd6b61747efd4fa846af9afae78e11b956380ca5a320
-size 91239
+oid sha256:f6157a39e82674efa322c7e73c29c3accb361df4be44464f77709ad708fffd70
+size 93357