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": 77035.63491816306,
-      "required_pages_per_second_upper_estimate": 0.005329503970938103
+      "required_token_positions_per_second": 77253.0906250334,
+      "required_pages_per_second_upper_estimate": 0.005344548061306148
     },
     {
       "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": 9629.454364770383,
-      "required_pages_per_second_upper_estimate": 0.004031651456655986
+      "required_token_positions_per_second": 9656.636328129174,
+      "required_pages_per_second_upper_estimate": 0.00404303197709037
     },
     {
       "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": 19258.908729540766,
-      "required_pages_per_second_upper_estimate": 0.004011122796785914
+      "required_token_positions_per_second": 19313.27265625835,
+      "required_pages_per_second_upper_estimate": 0.004022445369048029
     },
     {
       "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": 28888.363094311153,
-      "required_pages_per_second_upper_estimate": 0.0017318132575275592
+      "required_token_positions_per_second": 28969.908984387523,
+      "required_pages_per_second_upper_estimate": 0.0017367018091242728
     },
     {
       "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": 28888.363094311153,
-      "required_pages_per_second_upper_estimate": 0.00224594316410224
+      "required_token_positions_per_second": 28969.908984387523,
+      "required_pages_per_second_upper_estimate": 0.002252282998373214
     },
     {
       "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": 28888.363094311153,
-      "required_pages_per_second_upper_estimate": 0.3705489038661786
+      "required_token_positions_per_second": 28969.908984387523,
+      "required_pages_per_second_upper_estimate": 0.3715948869869232
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11572239733573042,
-    "producer_p95_seconds": 0.31440124600339914,
-    "gpu_update_seconds": 0.9570635729597496,
-    "producer_batches_per_second": 8.641369544901748,
-    "gpu_batches_per_second": 1.044862669788453,
-    "producer_headroom": 8.270340011909232,
+    "producer_batch_seconds": 0.5347739565156076,
+    "producer_p95_seconds": 0.3436727450025501,
+    "gpu_update_seconds": 0.9543695844850625,
+    "producer_batches_per_second": 1.8699489528540918,
+    "gpu_batches_per_second": 1.0478121015765163,
+    "producer_headroom": 1.7846224051436375,
     "sustainable": true,
-    "buffer_seconds": 60.29500509646423,
+    "buffer_seconds": 60.125283822558934,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366337.2425047756,
-    "upload_bytes_per_second": 3666978.9556036056,
+    "generation_bytes_per_second": 316728.63991648133,
+    "upload_bytes_per_second": 3643615.055064747,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.60744434697581,
+    "estimated_full_shards_per_hour": 67.96259306069211,
     "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": 1.5676271939992148,
-      "bytes": 12360812,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 1.7888847590002115,
       "bytes": 1223080,
@@ -562,6 +556,12 @@
       "bytes": 1034486,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.9976537119946443,
+      "bytes": 11257129,
+      "pages": 36,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -397,3 +397,4 @@
 {"step": 882200, "loss": 0.5362783074378967, "lr": 3e-05, "grad_norm": 0.29593270333345456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129129319154, "tokens_per_second": 152992.74771197283, "data_wait_s": 1.4457000361289829e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882300, "loss": 0.5350361466407776, "lr": 3e-05, "grad_norm": 0.3114949750423096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129144052720, "tokens_per_second": 152699.12172993395, "data_wait_s": 1.0168005246669054e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882400, "loss": 0.5504578948020935, "lr": 3e-05, "grad_norm": 0.3056236144632442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129158723361, "tokens_per_second": 151690.85855227493, "data_wait_s": 1.5284000255633146e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 882500, "loss": 0.5453665256500244, "lr": 3e-05, "grad_norm": 0.32312718729762313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129173267711, "tokens_per_second": 150456.0791782488, "data_wait_s": 1.6260994016192853e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b52c308ddeb1b67e3a7fdb32eea205ef121cdcde2e7148e6560558eaa609302f
-size 851511
+oid sha256:9e5b4d0c255c7ad987dae7f1fa38b1d5d27145adf36dd6a69ee358f006a56f76
+size 853629