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": 75171.67439676764,
-      "required_pages_per_second_upper_estimate": 0.006515690215844887
+      "required_token_positions_per_second": 74605.63399931267,
+      "required_pages_per_second_upper_estimate": 0.006466627268809762
     },
     {
       "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": 9396.459299595956,
-      "required_pages_per_second_upper_estimate": 0.0037483881041949717
+      "required_token_positions_per_second": 9325.704249914084,
+      "required_pages_per_second_upper_estimate": 0.0037201628569946085
     },
     {
       "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": 18792.91859919191,
-      "required_pages_per_second_upper_estimate": 0.004347410973852016
+      "required_token_positions_per_second": 18651.40849982817,
+      "required_pages_per_second_upper_estimate": 0.004314675102856901
     },
     {
       "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": 28189.377898787865,
-      "required_pages_per_second_upper_estimate": 0.001716585215800524
+      "required_token_positions_per_second": 27977.112749742257,
+      "required_pages_per_second_upper_estimate": 0.0017036593818928146
     },
     {
       "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": 28189.377898787865,
-      "required_pages_per_second_upper_estimate": 0.00217009634384106
+      "required_token_positions_per_second": 27977.112749742257,
+      "required_pages_per_second_upper_estimate": 0.0021537555850799895
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28189.377898787865,
-      "required_pages_per_second_upper_estimate": 0.35001772972407547
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27977.112749742257,
+      "required_pages_per_second_upper_estimate": 0.34738210697868377
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.1496975644300278,
-    "producer_p95_seconds": 0.37976628899923526,
-    "gpu_update_seconds": 0.9807949682064057,
-    "producer_batches_per_second": 6.680135403722108,
-    "gpu_batches_per_second": 1.0195810871957418,
-    "producer_headroom": 6.55184319090811,
+    "producer_batch_seconds": 0.19244506474160517,
+    "producer_p95_seconds": 0.3721370540006319,
+    "gpu_update_seconds": 0.9882363576010792,
+    "producer_batches_per_second": 5.1962881009325645,
+    "gpu_batches_per_second": 1.0119036729507471,
+    "producer_headroom": 5.135160825911427,
     "sustainable": true,
-    "buffer_seconds": 60.80928802879715,
+    "buffer_seconds": 60.28241781366583,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386629.51134227915,
-    "upload_bytes_per_second": 5177168.466736914,
+    "generation_bytes_per_second": 370937.5151567315,
+    "upload_bytes_per_second": 5270220.275108785,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.96169285966187,
+    "estimated_full_shards_per_hour": 79.594555769219,
     "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": 0.9079114560008748,
-      "bytes": 1141999,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1828810689985403,
       "bytes": 7222759,
@@ -562,6 +556,12 @@
       "bytes": 871659,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.484286307997536,
+      "bytes": 11527011,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 391300, "loss": 0.58982796035707, "lr": 0.0003, "grad_norm": 0.14220308563823453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57275549122, "tokens_per_second": 150404.7049975993, "data_wait_s": 1.0481002391315997e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 391400, "loss": 0.6011860184371471, "lr": 0.0003, "grad_norm": 0.1481824306291748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57290258808, "tokens_per_second": 149502.60229895407, "data_wait_s": 2.0356004824861884e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 391500, "loss": 0.5959842074662447, "lr": 0.0003, "grad_norm": 0.15582991121489526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57304947576, "tokens_per_second": 150066.05898748786, "data_wait_s": 1.0757001291494817e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 391600, "loss": 0.5928702764213085, "lr": 0.0003, "grad_norm": 0.1683905331201567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57319512166, "tokens_per_second": 148596.08297080142, "data_wait_s": 9.602998034097254e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8569975be90f70284389c8b6d4472e2299699399c105d915b552a82080cf025d
-size 778716
+oid sha256:2847400836cafaba1cde6a2e1d026b79a92bbd8b3b3e30784c310c3fbf680c3d
+size 780834