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": 77609.00028242862,
-      "required_pages_per_second_upper_estimate": 0.005369170717228931
+      "required_token_positions_per_second": 76741.9490690328,
+      "required_pages_per_second_upper_estimate": 0.005309186102450216
     },
     {
       "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": 9701.125035303578,
-      "required_pages_per_second_upper_estimate": 0.004061658469754444
+      "required_token_positions_per_second": 9592.7436336291,
+      "required_pages_per_second_upper_estimate": 0.004016281440134371
     },
     {
       "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": 19402.250070607155,
-      "required_pages_per_second_upper_estimate": 0.004040977017964674
+      "required_token_positions_per_second": 19185.4872672582,
+      "required_pages_per_second_upper_estimate": 0.0039958310424466235
     },
     {
       "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": 29103.375105910734,
-      "required_pages_per_second_upper_estimate": 0.0017447028993186216
+      "required_token_positions_per_second": 28778.230900887298,
+      "required_pages_per_second_upper_estimate": 0.0017252110006939214
     },
     {
       "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": 29103.375105910734,
-      "required_pages_per_second_upper_estimate": 0.002262659402266217
+      "required_token_positions_per_second": 28778.230900887298,
+      "required_pages_per_second_upper_estimate": 0.002237380870483859
     },
     {
       "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": 29103.375105910734,
-      "required_pages_per_second_upper_estimate": 0.37330684708906675
+      "required_token_positions_per_second": 28778.230900887298,
+      "required_pages_per_second_upper_estimate": 0.3691362463396737
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12046738655459421,
-    "producer_p95_seconds": 0.29918241500126896,
-    "gpu_update_seconds": 0.9499929097359173,
-    "producer_batches_per_second": 8.30100186117023,
-    "gpu_batches_per_second": 1.0526394352542945,
-    "producer_headroom": 7.885892911816371,
+    "producer_batch_seconds": 0.10669844648447224,
+    "producer_p95_seconds": 0.29845412699796725,
+    "gpu_update_seconds": 0.9607261855400413,
+    "producer_batches_per_second": 9.372207683881596,
+    "gpu_batches_per_second": 1.040879300523991,
+    "producer_headroom": 9.00412533822463,
     "sustainable": true,
-    "buffer_seconds": 60.79954622309871,
+    "buffer_seconds": 60.525749689022604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280704.17095901835,
-    "upload_bytes_per_second": 5474776.982346963,
+    "generation_bytes_per_second": 280539.1769618955,
+    "upload_bytes_per_second": 5361523.345347216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.23258062913811,
+    "estimated_full_shards_per_hour": 60.19717675821923,
     "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": 1.2983564580026723,
-      "bytes": 12027088,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2329098789996351,
       "bytes": 6852339,
@@ -562,6 +556,12 @@
       "bytes": 9700555,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1995469100002083,
+      "bytes": 2400407,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 522800, "loss": 0.6225700974464417, "lr": 0.0003, "grad_norm": 0.16724649868037628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76520210382, "tokens_per_second": 153180.08142781453, "data_wait_s": 1.105399860534817e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522900, "loss": 0.6080971956253052, "lr": 0.0003, "grad_norm": 0.1603530909927749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76534888937, "tokens_per_second": 153614.17240374276, "data_wait_s": 1.4880999515298754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523000, "loss": 0.5470998883247375, "lr": 0.0003, "grad_norm": 0.1361541966712855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76549333904, "tokens_per_second": 150580.81061198952, "data_wait_s": 1.6494996089022607e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 523100, "loss": 0.5512901544570923, "lr": 0.0003, "grad_norm": 0.18323662469231008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76563821788, "tokens_per_second": 151782.80051866503, "data_wait_s": 9.128001693170518e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a5ed7043431a890250bf3e5cc7185148578a59b177ba58d0b688763f598d231
-size 810486
+oid sha256:e7bf9d0e9d6f5b9c2407b33077c540dfb8ae32eff024c1e69bf740704091fef6
+size 812604