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": 76327.57162622061,
-      "required_pages_per_second_upper_estimate": 0.005280518509468334
+      "required_token_positions_per_second": 76466.1540811154,
+      "required_pages_per_second_upper_estimate": 0.005290105965253553
     },
     {
       "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": 9540.946453277576,
-      "required_pages_per_second_upper_estimate": 0.003994595042369312
+      "required_token_positions_per_second": 9558.269260139425,
+      "required_pages_per_second_upper_estimate": 0.004001847739861026
     },
     {
       "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": 19081.892906555153,
-      "required_pages_per_second_upper_estimate": 0.003974255069079187
+      "required_token_positions_per_second": 19116.53852027885,
+      "required_pages_per_second_upper_estimate": 0.003981470836751558
     },
     {
       "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": 28622.839359832727,
-      "required_pages_per_second_upper_estimate": 0.0017158955150768447
+      "required_token_positions_per_second": 28674.807780418276,
+      "required_pages_per_second_upper_estimate": 0.0017190109425397648
     },
     {
       "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": 28622.839359832727,
-      "required_pages_per_second_upper_estimate": 0.0022252998616620213
+      "required_token_positions_per_second": 28674.807780418276,
+      "required_pages_per_second_upper_estimate": 0.002229340177777623
     },
     {
       "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": 28622.839359832727,
-      "required_pages_per_second_upper_estimate": 0.3671430504974632
+      "required_token_positions_per_second": 28674.807780418276,
+      "required_pages_per_second_upper_estimate": 0.367809645597392
     }
   ],
   "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": 0.09984876964844602,
-    "producer_p95_seconds": 0.2026637640010449,
-    "gpu_update_seconds": 0.9659419057774978,
-    "producer_batches_per_second": 10.015145940414332,
-    "gpu_batches_per_second": 1.0352589467532092,
-    "producer_headroom": 9.67404915632359,
+    "producer_batch_seconds": 0.0859565982422481,
+    "producer_p95_seconds": 0.20224820700059354,
+    "gpu_update_seconds": 0.964191293337301,
+    "producer_batches_per_second": 11.633778214230155,
+    "gpu_batches_per_second": 1.0371385915949898,
+    "producer_headroom": 11.217187662777889,
     "sustainable": true,
-    "buffer_seconds": 60.85434006398236,
+    "buffer_seconds": 60.744051480249965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313655.62182173267,
-    "upload_bytes_per_second": 5568236.237819763,
+    "generation_bytes_per_second": 341013.1456026689,
+    "upload_bytes_per_second": 5569373.943809598,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.30319491375909,
+    "estimated_full_shards_per_hour": 73.17348266658831,
     "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": 1.059233770999981,
-      "bytes": 6001877,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2980492899996534,
       "bytes": 11801808,
@@ -562,6 +556,12 @@
       "bytes": 1081337,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3408090630000515,
+      "bytes": 7658166,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -98,3 +98,4 @@
 {"step": 614200, "loss": 0.514626681804657, "lr": 0.0001, "grad_norm": 0.15638066934972122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89900729754, "tokens_per_second": 151111.94238210047, "data_wait_s": 1.280899959965609e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614300, "loss": 0.5313458442687988, "lr": 0.0001, "grad_norm": 0.1551929440788104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89915535292, "tokens_per_second": 154554.22675837556, "data_wait_s": 1.3519000276573934e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614400, "loss": 0.5308977961540222, "lr": 0.0001, "grad_norm": 0.15364842307200952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89930116871, "tokens_per_second": 151912.3556869722, "data_wait_s": 1.3777998901787214e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 614500, "loss": 0.6621916890144348, "lr": 0.0001, "grad_norm": 0.1962394419839126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89944810027, "tokens_per_second": 152048.15024245056, "data_wait_s": 1.0489000487723388e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:34895ad7db98aeb88f749b2bc6fab35984ba8be50ac1a015dee894bec7e637b0
-size 213531
+oid sha256:c5c154a2b10b70f7a6123771feef7f7fcfe40dfd51da55aa965616b9b7e412f2
+size 215649