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": 75286.13404858578,
-      "required_pages_per_second_upper_estimate": 0.006525611288369148
+      "required_token_positions_per_second": 75320.49426918013,
+      "required_pages_per_second_upper_estimate": 0.006528589545205083
     },
     {
       "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": 9410.766756073222,
-      "required_pages_per_second_upper_estimate": 0.003754095562499291
+      "required_token_positions_per_second": 9415.061783647516,
+      "required_pages_per_second_upper_estimate": 0.0037558089132150615
     },
     {
       "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": 18821.533512146445,
-      "required_pages_per_second_upper_estimate": 0.0043540305303587795
+      "required_token_positions_per_second": 18830.12356729503,
+      "required_pages_per_second_upper_estimate": 0.004356017688437597
     },
     {
       "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": 28232.30026821967,
-      "required_pages_per_second_upper_estimate": 0.0017191989629026527
+      "required_token_positions_per_second": 28245.185350942545,
+      "required_pages_per_second_upper_estimate": 0.0017199835968376713
     },
     {
       "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": 28232.30026821967,
-      "required_pages_per_second_upper_estimate": 0.002173400626656644
+      "required_token_positions_per_second": 28245.185350942545,
+      "required_pages_per_second_upper_estimate": 0.0021743925559928427
     },
     {
       "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": 28232.30026821967,
-      "required_pages_per_second_upper_estimate": 0.3505506819004889
+      "required_token_positions_per_second": 28245.185350942545,
+      "required_pages_per_second_upper_estimate": 0.3507106715043091
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11839931055436637,
-    "producer_p95_seconds": 0.22381721999772708,
-    "gpu_update_seconds": 0.9793038377082791,
-    "producer_batches_per_second": 8.445995127149173,
-    "gpu_batches_per_second": 1.021133545580862,
-    "producer_headroom": 8.271195441282611,
+    "producer_batch_seconds": 0.44343158629678214,
+    "producer_p95_seconds": 0.25962030599839636,
+    "gpu_update_seconds": 0.9788570921549071,
+    "producer_batches_per_second": 2.255139306496572,
+    "gpu_batches_per_second": 1.021599585899253,
+    "producer_headroom": 2.207459103961468,
     "sustainable": true,
-    "buffer_seconds": 60.716837937913304,
+    "buffer_seconds": 60.68913971360424,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317785.2737379387,
+    "generation_bytes_per_second": 294107.2586854045,
     "upload_bytes_per_second": 5261200.170008222,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.1893220816004,
+    "estimated_full_shards_per_hour": 63.108571247306834,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -165,3 +165,4 @@
 {"step": 211500, "loss": 0.6214869990944862, "lr": 0.0003, "grad_norm": 0.1705022461807806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30956467534, "tokens_per_second": 147575.71681998717, "data_wait_s": 1.6814999980852008e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 211600, "loss": 0.5997838526964188, "lr": 0.0003, "grad_norm": 0.14328802200601443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30971026863, "tokens_per_second": 147683.235183447, "data_wait_s": 1.5331999748013914e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 211700, "loss": 0.6280692806467414, "lr": 0.0003, "grad_norm": 0.13826314396872577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30985629898, "tokens_per_second": 137415.2549507914, "data_wait_s": 1.3729997590417042e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 211800, "loss": 0.5685373367741704, "lr": 0.0003, "grad_norm": 0.13446272166246892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31000111798, "tokens_per_second": 147710.7019044938, "data_wait_s": 1.8356000509811565e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a70c63dea041be40f1b09e7d193b647a7c96e9a2d883ff91a88ffaa52e9e9dd5
-size 356220
+oid sha256:13dc166b5ab3cdcb67bd7213139aabf1b06b3c176de56b3c7daa569275c9630f
+size 358338