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": 76853.63444472375,
-      "required_pages_per_second_upper_estimate": 0.005316912755886291
+      "required_token_positions_per_second": 77256.62007174829,
+      "required_pages_per_second_upper_estimate": 0.005344792236619329
     },
     {
       "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": 9606.704305590469,
-      "required_pages_per_second_upper_estimate": 0.004022126481952614
+      "required_token_positions_per_second": 9657.077508968536,
+      "required_pages_per_second_upper_estimate": 0.004043216690294908
     },
     {
       "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": 19213.408611180937,
-      "required_pages_per_second_upper_estimate": 0.004001646322050374
+      "required_token_positions_per_second": 19314.15501793707,
+      "required_pages_per_second_upper_estimate": 0.004022629141716264
     },
     {
       "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": 28820.112916771406,
-      "required_pages_per_second_upper_estimate": 0.0017277217635960384
+      "required_token_positions_per_second": 28971.232526905605,
+      "required_pages_per_second_upper_estimate": 0.0017367811534772996
     },
     {
       "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": 28820.112916771406,
-      "required_pages_per_second_upper_estimate": 0.0022406370129993292
+      "required_token_positions_per_second": 28971.232526905605,
+      "required_pages_per_second_upper_estimate": 0.0022523858979823464
     },
     {
       "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": 28820.112916771406,
-      "required_pages_per_second_upper_estimate": 0.36967346387003
+      "required_token_positions_per_second": 28971.232526905605,
+      "required_pages_per_second_upper_estimate": 0.37161186396923596
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13042824582839785,
-    "producer_p95_seconds": 0.3381571759964572,
-    "gpu_update_seconds": 0.9593300373195515,
-    "producer_batches_per_second": 7.667050903342536,
-    "gpu_batches_per_second": 1.0423941303809103,
-    "producer_headroom": 7.355232229234495,
+    "producer_batch_seconds": 0.42941833892194836,
+    "producer_p95_seconds": 0.347555418004049,
+    "gpu_update_seconds": 0.9543259843820342,
+    "producer_batches_per_second": 2.3287314708321327,
+    "gpu_batches_per_second": 1.0478599727613427,
+    "producer_headroom": 2.222368953263297,
     "sustainable": true,
-    "buffer_seconds": 60.43779235113174,
+    "buffer_seconds": 60.12253701606816,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450137.808768403,
-    "upload_bytes_per_second": 4746473.526917738,
+    "generation_bytes_per_second": 414164.6061822194,
+    "upload_bytes_per_second": 4747188.705197843,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.58909508980815,
+    "estimated_full_shards_per_hour": 88.87008322811066,
     "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": 1.0184783190015878,
-      "bytes": 1341723,
-      "pages": 35,
-      "shards": 1
-    },
-    {
-      "seconds": 1.3343794770007662,
-      "bytes": 7400925,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.8426530299984734,
       "bytes": 5834868,
@@ -562,6 +550,18 @@
       "bytes": 6077960,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0119541929961997,
+      "bytes": 2485722,
+      "pages": 30,
+      "shards": 2
+    },
+    {
+      "seconds": 1.1930818900000304,
+      "bytes": 5619417,
+      "pages": 14,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -348,3 +348,4 @@
 {"step": 959800, "loss": 0.4464470148086548, "lr": 3e-05, "grad_norm": 0.4452609082534058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140488578085, "tokens_per_second": 153033.0980581865, "data_wait_s": 1.3746001059189439e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959900, "loss": 0.6030485033988953, "lr": 3e-05, "grad_norm": 0.3866498325193008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140503235806, "tokens_per_second": 152363.01700591584, "data_wait_s": 1.022600190481171e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960000, "loss": 0.5418819785118103, "lr": 3e-05, "grad_norm": 0.350432378987098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140517911346, "tokens_per_second": 152964.01751645192, "data_wait_s": 1.2600998161360621e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 960100, "loss": 0.5036099553108215, "lr": 3e-05, "grad_norm": 0.3559221631435305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140532670170, "tokens_per_second": 140340.27813467168, "data_wait_s": 1.2641001376323402e-05, "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 sha256:88f40c127c8c773143bfa24532b5697aaf4c725a54ba5b4b1f987ff9f3766878
-size 746946
+oid sha256:cd63f06a3df833161c41cf3faec40407fb152b2f254c91136f8ed007703d6bb3
+size 749064