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": 75992.12644967414,
-      "required_pages_per_second_upper_estimate": 0.006586804920374105
+      "required_token_positions_per_second": 75610.21414028478,
+      "required_pages_per_second_upper_estimate": 0.006553701729344134
     },
     {
       "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": 9499.015806209267,
-      "required_pages_per_second_upper_estimate": 0.0037892994280394393
+      "required_token_positions_per_second": 9451.276767535597,
+      "required_pages_per_second_upper_estimate": 0.003770255611750278
     },
     {
       "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": 18998.031612418534,
-      "required_pages_per_second_upper_estimate": 0.004394860259596254
+      "required_token_positions_per_second": 18902.553535071194,
+      "required_pages_per_second_upper_estimate": 0.004372773086758717
     },
     {
       "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": 28497.0474186278,
-      "required_pages_per_second_upper_estimate": 0.0017353206753415588
+      "required_token_positions_per_second": 28353.83030260679,
+      "required_pages_per_second_upper_estimate": 0.0017265995043780156
     },
     {
       "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": 28497.0474186278,
-      "required_pages_per_second_upper_estimate": 0.0021937815951620795
+      "required_token_positions_per_second": 28353.83030260679,
+      "required_pages_per_second_upper_estimate": 0.002182756345120431
     },
     {
       "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": 28497.0474186278,
-      "required_pages_per_second_upper_estimate": 0.35383795545684343
+      "required_token_positions_per_second": 28353.83030260679,
+      "required_pages_per_second_upper_estimate": 0.3520596781927163
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11259993203105978,
-    "producer_p95_seconds": 0.3923639889981132,
-    "gpu_update_seconds": 0.9702057758421386,
-    "producer_batches_per_second": 8.881000032257196,
-    "gpu_batches_per_second": 1.0307091803612487,
-    "producer_headroom": 8.616397526550152,
+    "producer_batch_seconds": 0.33562050373421926,
+    "producer_p95_seconds": 0.37606325300293975,
+    "gpu_update_seconds": 0.9751063508854428,
+    "producer_batches_per_second": 2.979555744877579,
+    "gpu_batches_per_second": 1.0255291631440535,
+    "producer_headroom": 2.9053837296473337,
     "sustainable": true,
-    "buffer_seconds": 60.152758102212594,
+    "buffer_seconds": 60.456593754897455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 508421.10911946173,
-    "upload_bytes_per_second": 5627562.392582926,
+    "generation_bytes_per_second": 446895.26767355576,
+    "upload_bytes_per_second": 5679518.923861766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.09533457935227,
+    "estimated_full_shards_per_hour": 95.89332125334744,
     "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": [
-    {
-      "seconds": 1.1451003039983334,
-      "bytes": 6976542,
-      "pages": 21,
-      "shards": 3
-    },
     {
       "seconds": 0.8673626480012899,
       "bytes": 1124477,
@@ -562,6 +556,12 @@
       "bytes": 1164563,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.0845683700026711,
+      "bytes": 10426904,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 312500, "loss": 0.5605333168059587, "lr": 0.0003, "grad_norm": 0.15028700770938597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45741536649, "tokens_per_second": 150805.988811795, "data_wait_s": 9.02800093172118e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 312600, "loss": 0.5956666208803654, "lr": 0.0003, "grad_norm": 0.18772054048968298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45756026804, "tokens_per_second": 148709.25539628958, "data_wait_s": 1.050400169333443e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 312700, "loss": 0.5665262192487717, "lr": 0.0003, "grad_norm": 0.15577467373689494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45770504431, "tokens_per_second": 148665.95959429743, "data_wait_s": 1.5102996258065104e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 312800, "loss": 0.5574280023574829, "lr": 0.0003, "grad_norm": 0.14596527930463818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45785057663, "tokens_per_second": 149486.37292087192, "data_wait_s": 1.041700306814164e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70026526950300ab233665c609d2bea3eec472c3d5996e2c90d703d82ddd79f6
-size 693213
+oid sha256:6af9c60b8b3f0964c568cba99902a2edb84ef6f4fdfadbe1d8213543a1068aae
+size 695331