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": 74828.03381926876,
-      "required_pages_per_second_upper_estimate": 0.006485904321536365
+      "required_token_positions_per_second": 74877.65248182749,
+      "required_pages_per_second_upper_estimate": 0.006490205141449605
     },
     {
       "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": 9353.504227408595,
-      "required_pages_per_second_upper_estimate": 0.0037312526836638725
+      "required_token_positions_per_second": 9359.706560228437,
+      "required_pages_per_second_upper_estimate": 0.0037337268869588464
     },
     {
       "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": 18707.00845481719,
-      "required_pages_per_second_upper_estimate": 0.004327537173917825
+      "required_token_positions_per_second": 18719.413120456873,
+      "required_pages_per_second_upper_estimate": 0.004330406774998906
     },
     {
       "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": 28060.512682225788,
-      "required_pages_per_second_upper_estimate": 0.0017087380002154327
+      "required_token_positions_per_second": 28079.119680685308,
+      "required_pages_per_second_upper_estimate": 0.0017098710687982414
     },
     {
       "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": 28060.512682225788,
-      "required_pages_per_second_upper_estimate": 0.002160175942748361
+      "required_token_positions_per_second": 28079.119680685308,
+      "required_pages_per_second_upper_estimate": 0.002161608360997243
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28060.512682225788,
-      "required_pages_per_second_upper_estimate": 0.3484176550185106
+      "required_token_positions_per_second": 28079.119680685308,
+      "required_pages_per_second_upper_estimate": 0.34864869166575996
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09439860396102517,
-    "producer_p95_seconds": 0.32911561600121786,
-    "gpu_update_seconds": 0.9852991751470356,
-    "producer_batches_per_second": 10.593376999652188,
-    "gpu_batches_per_second": 1.0149201635643008,
-    "producer_headroom": 10.437645619778879,
+    "producer_batch_seconds": 0.097406510632851,
+    "producer_p95_seconds": 0.31932546800089767,
+    "gpu_update_seconds": 0.9846462536721954,
+    "producer_batches_per_second": 10.266254211376538,
+    "gpu_batches_per_second": 1.0155931597470091,
+    "producer_headroom": 10.108628748478306,
     "sustainable": true,
-    "buffer_seconds": 60.10324968396917,
+    "buffer_seconds": 60.06342147400392,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488633.5744452333,
-    "upload_bytes_per_second": 4102653.020833448,
+    "generation_bytes_per_second": 467036.96441197745,
+    "upload_bytes_per_second": 4102907.7703114543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.84939026849507,
+    "estimated_full_shards_per_hour": 100.21526049870961,
     "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": 1.0569672790006734,
-      "bytes": 5933537,
-      "pages": 27,
-      "shards": 2
-    },
-    {
-      "seconds": 0.8821923329996935,
-      "bytes": 961115,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.2227748900004372,
       "bytes": 12637313,
@@ -562,6 +550,18 @@
       "bytes": 11506344,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.0356865150024532,
+      "bytes": 6075433,
+      "pages": 35,
+      "shards": 2
+    },
+    {
+      "seconds": 0.8513431389983452,
+      "bytes": 631745,
+      "pages": 18,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -198,3 +198,4 @@
 {"step": 104800, "loss": 0.610983818769455, "lr": 0.0003, "grad_norm": 0.13781168663160606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15336195977, "tokens_per_second": 148995.1361121154, "data_wait_s": 1.850099943112582e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 104900, "loss": 0.5931651797145605, "lr": 0.0003, "grad_norm": 0.12879474546570285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15350920625, "tokens_per_second": 148624.37984174673, "data_wait_s": 1.4916997315594926e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105000, "loss": 0.6216882299631834, "lr": 0.0003, "grad_norm": 0.13170376331896053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15365640815, "tokens_per_second": 148940.21746605175, "data_wait_s": 1.3888999092159793e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 105100, "loss": 0.6351448111236095, "lr": 0.0003, "grad_norm": 0.1474901384848249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15380211498, "tokens_per_second": 124904.54535929876, "data_wait_s": 1.1435997294029221e-05, "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 sha256:6bdf47be175bb2c58a44e6e28659b1eb3d3fe3c89eeb208df9b42c4453c26a33
-size 426897
+oid sha256:93d4f3733ed4f94113132dd6c7e7b55127961923fadf488b2769fe3a78d423ad
+size 429015