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": 77096.7375377669,
-      "required_pages_per_second_upper_estimate": 0.005333731191939906
+      "required_token_positions_per_second": 77436.3696287167,
+      "required_pages_per_second_upper_estimate": 0.005357227728046837
     },
     {
       "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": 9637.092192220862,
-      "required_pages_per_second_upper_estimate": 0.004034849255513527
+      "required_token_positions_per_second": 9679.546203589587,
+      "required_pages_per_second_upper_estimate": 0.004052623863533044
     },
     {
       "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": 19274.184384441724,
-      "required_pages_per_second_upper_estimate": 0.0040143043128556734
+      "required_token_positions_per_second": 19359.092407179174,
+      "required_pages_per_second_upper_estimate": 0.004031988414816747
     },
     {
       "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": 28911.276576662585,
-      "required_pages_per_second_upper_estimate": 0.0017331868858076658
+      "required_token_positions_per_second": 29038.63861076876,
+      "required_pages_per_second_upper_estimate": 0.001740822045281765
     },
     {
       "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": 28911.276576662585,
-      "required_pages_per_second_upper_estimate": 0.002247724586569306
+      "required_token_positions_per_second": 29038.63861076876,
+      "required_pages_per_second_upper_estimate": 0.0022576264245146777
     },
     {
       "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": 28911.276576662585,
-      "required_pages_per_second_upper_estimate": 0.3708428134152023
+      "required_token_positions_per_second": 29038.63861076876,
+      "required_pages_per_second_upper_estimate": 0.3724764768380185
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10067282678910772,
-    "producer_p95_seconds": 0.21053473099891562,
-    "gpu_update_seconds": 0.9563050571871907,
-    "producer_batches_per_second": 9.933166991474554,
-    "gpu_batches_per_second": 1.0456914271072983,
-    "producer_headroom": 9.499137827831989,
+    "producer_batch_seconds": 0.08800965217963608,
+    "producer_p95_seconds": 0.21772737200080883,
+    "gpu_update_seconds": 0.9521107504587679,
+    "producer_batches_per_second": 11.362390092837826,
+    "gpu_batches_per_second": 1.0502979821603284,
+    "producer_headroom": 10.818253758297093,
     "sustainable": true,
-    "buffer_seconds": 60.24721860279301,
+    "buffer_seconds": 60.93508802936115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484327.0970589548,
+    "generation_bytes_per_second": 493315.19661256357,
     "upload_bytes_per_second": 5372182.101540501,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.9253204710625,
+    "estimated_full_shards_per_hour": 105.85395740301782,
     "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": [

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -122,3 +122,4 @@
 {"step": 897200, "loss": 0.5151164531707764, "lr": 3e-05, "grad_norm": 0.31993874572475556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131324839262, "tokens_per_second": 153999.89372563612, "data_wait_s": 1.3525999747798778e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897300, "loss": 0.4963752031326294, "lr": 3e-05, "grad_norm": 0.3112137685009082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131339475719, "tokens_per_second": 152511.61751028922, "data_wait_s": 1.7022000974975526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897400, "loss": 0.5102488994598389, "lr": 3e-05, "grad_norm": 0.3118720297717996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131354045000, "tokens_per_second": 152348.96066376314, "data_wait_s": 2.3840999347157776e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 897500, "loss": 0.5129820704460144, "lr": 3e-05, "grad_norm": 0.32767872062354386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131368784009, "tokens_per_second": 154032.81284392608, "data_wait_s": 1.7063999621313997e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c35f034fdc715465feb9cfb5d986e6458ceb6da789213452bc62ad3b1873647
-size 264363
+oid sha256:597d11f9704da31cbfd3b8b41308b8ae4c1c6e983cd569c04cbbefc32e0476c1
+size 266481