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": 76242.04132255772,
-      "required_pages_per_second_upper_estimate": 0.0066084669081522
+      "required_token_positions_per_second": 76142.23147304803,
+      "required_pages_per_second_upper_estimate": 0.00659981564336245
     },
     {
       "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": 9530.255165319715,
-      "required_pages_per_second_upper_estimate": 0.0038017612754586384
+      "required_token_positions_per_second": 9517.778934131004,
+      "required_pages_per_second_upper_estimate": 0.0037967843203011824
     },
     {
       "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": 19060.51033063943,
-      "required_pages_per_second_upper_estimate": 0.004409313611468774
+      "required_token_positions_per_second": 19035.557868262007,
+      "required_pages_per_second_upper_estimate": 0.0044035412984460955
     },
     {
       "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": 28590.765495959145,
-      "required_pages_per_second_upper_estimate": 0.0017410276145502866
+      "required_token_positions_per_second": 28553.33680239301,
+      "required_pages_per_second_upper_estimate": 0.0017387484034852873
     },
     {
       "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": 28590.765495959145,
-      "required_pages_per_second_upper_estimate": 0.0022009962721832887
+      "required_token_positions_per_second": 28553.33680239301,
+      "required_pages_per_second_upper_estimate": 0.0021981149077433125
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28590.765495959145,
-      "required_pages_per_second_upper_estimate": 0.3550016203230707
+      "required_token_positions_per_second": 28553.33680239301,
+      "required_pages_per_second_upper_estimate": 0.35453688121475857
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08035174422667524,
-    "producer_p95_seconds": 0.19821669999873848,
-    "gpu_update_seconds": 0.9670255245144663,
-    "producer_batches_per_second": 12.445280555191971,
-    "gpu_batches_per_second": 1.034098867764726,
-    "producer_headroom": 12.034903956614205,
+    "producer_batch_seconds": 0.05980398355504235,
+    "producer_p95_seconds": 0.19446895200235303,
+    "gpu_update_seconds": 0.9682931347513425,
+    "producer_batches_per_second": 16.72129414388626,
+    "gpu_batches_per_second": 1.0327451100402565,
+    "producer_headroom": 16.191114323682896,
     "sustainable": true,
-    "buffer_seconds": 60.922608044411376,
+    "buffer_seconds": 60.034174354583236,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 175,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478359.14039220393,
-    "upload_bytes_per_second": 5724954.833400937,
+    "generation_bytes_per_second": 627374.0327731093,
+    "upload_bytes_per_second": 5535911.545161966,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.6447358972987,
+    "estimated_full_shards_per_hour": 67.30993145654182,
     "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": 1.2676941070003522,
-      "bytes": 16626023,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.9997899369991501,
       "bytes": 2248595,
@@ -562,6 +556,12 @@
       "bytes": 6128146,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0307726470018679,
+      "bytes": 1163007,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -89,3 +89,4 @@
 {"step": 289100, "loss": 0.6017055995762348, "lr": 0.0003, "grad_norm": 0.15662319902066163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42316200440, "tokens_per_second": 150184.7106884939, "data_wait_s": 9.663999662734568e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 289200, "loss": 0.607087342068553, "lr": 0.0003, "grad_norm": 0.20779946782119998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42330887819, "tokens_per_second": 150426.78832285226, "data_wait_s": 8.348000847036019e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 289300, "loss": 0.5729494746774435, "lr": 0.0003, "grad_norm": 0.17114440043056495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42345554415, "tokens_per_second": 150521.54014290456, "data_wait_s": 1.6216999938478693e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 289400, "loss": 0.5982925966382027, "lr": 0.0003, "grad_norm": 0.14897812007285297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42360170725, "tokens_per_second": 150265.00037960656, "data_wait_s": 1.1796997569035739e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:efb2e645e2045de10eb65602c59755cb2da4bbb71e69081b5ba17eab064b9384
-size 193686
+oid sha256:05fe425bc08a1cdcbe26e7ee6f2f55ddb29dd897b7b8da9229e1ab964e1190a2
+size 195804