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": 76024.09036732094,
-      "required_pages_per_second_upper_estimate": 0.005259522971804119
+      "required_token_positions_per_second": 76319.69523552853,
+      "required_pages_per_second_upper_estimate": 0.005279973602484512
     },
     {
       "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": 9503.011295915117,
-      "required_pages_per_second_upper_estimate": 0.003978712384157817
+      "required_token_positions_per_second": 9539.961904441067,
+      "required_pages_per_second_upper_estimate": 0.003994182832331183
     },
     {
       "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": 19006.022591830235,
-      "required_pages_per_second_upper_estimate": 0.003958453283356736
+      "required_token_positions_per_second": 19079.923808882133,
+      "required_pages_per_second_upper_estimate": 0.003973844957962495
     },
     {
       "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": 28509.033887745354,
-      "required_pages_per_second_upper_estimate": 0.001709073050796101
+      "required_token_positions_per_second": 28619.885713323198,
+      "required_pages_per_second_upper_estimate": 0.0017157184484785555
     },
     {
       "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": 28509.033887745354,
-      "required_pages_per_second_upper_estimate": 0.002216451986784597
+      "required_token_positions_per_second": 28619.885713323198,
+      "required_pages_per_second_upper_estimate": 0.002225070228637623
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28509.033887745354,
-      "required_pages_per_second_upper_estimate": 0.36568327609632195
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28619.885713323198,
+      "required_pages_per_second_upper_estimate": 0.3671051642914175
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.15890016428124554,
-    "producer_p95_seconds": 0.16403411499959475,
-    "gpu_update_seconds": 0.9697978580706844,
-    "producer_batches_per_second": 6.293259698775697,
-    "gpu_batches_per_second": 1.0311427187407898,
-    "producer_headroom": 6.103189776155231,
+    "producer_batch_seconds": 0.08383759853905559,
+    "producer_p95_seconds": 0.1600887029999285,
+    "gpu_update_seconds": 0.9660415934899851,
+    "producer_batches_per_second": 11.927822569180007,
+    "gpu_batches_per_second": 1.0351521163673032,
+    "producer_headroom": 11.52277272159646,
     "sustainable": true,
-    "buffer_seconds": 60.12746720038243,
+    "buffer_seconds": 60.86062038986906,
     "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": 432076.9608395321,
-    "upload_bytes_per_second": 5152487.567426689,
+    "generation_bytes_per_second": 430568.51740247285,
+    "upload_bytes_per_second": 5235964.20776584,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.71365755929443,
+    "estimated_full_shards_per_hour": 92.38998071246756,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 1162985,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.4578549020002356,
+      "bytes": 10422176,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -32,3 +32,4 @@
 {"step": 607600, "loss": 0.5491907000541687, "lr": 0.0001, "grad_norm": 0.1370757922205129, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88934813270, "tokens_per_second": 153153.35838071772, "data_wait_s": 1.163200022347155e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 607700, "loss": 0.5888853073120117, "lr": 0.0001, "grad_norm": 0.13579830744958846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88949481422, "tokens_per_second": 152801.69091184073, "data_wait_s": 7.930999800009886e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 607800, "loss": 0.5417321920394897, "lr": 0.0001, "grad_norm": 0.14292809824885952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88963958706, "tokens_per_second": 150423.73204376292, "data_wait_s": 9.657000191509724e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 607900, "loss": 0.5578194856643677, "lr": 0.0001, "grad_norm": 0.13951009195828343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88978511604, "tokens_per_second": 151501.35394295419, "data_wait_s": 1.9536000309017254e-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 sha256:2c5aa2f59e6481dd621f7a9cbaeac8aba5c15fc825471ffd38ed49d316763d73
-size 72960
+oid sha256:b741783ea3488c123a679e8470152f99904687003aef27f498899442deb282e3
+size 75078