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": 74256.1431426979,
-      "required_pages_per_second_upper_estimate": 0.006436334287134871
+      "required_token_positions_per_second": 75094.83411572932,
+      "required_pages_per_second_upper_estimate": 0.006509029895034397
     },
     {
       "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": 9282.017892837237,
-      "required_pages_per_second_upper_estimate": 0.003702735715987409
+      "required_token_positions_per_second": 9386.854264466165,
+      "required_pages_per_second_upper_estimate": 0.0037445565120736257
     },
     {
       "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": 18564.035785674474,
-      "required_pages_per_second_upper_estimate": 0.004294462963144694
+      "required_token_positions_per_second": 18773.70852893233,
+      "required_pages_per_second_upper_estimate": 0.004342967062183151
     },
     {
       "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": 27846.05367851171,
-      "required_pages_per_second_upper_estimate": 0.0016956785720686795
+      "required_token_positions_per_second": 28160.56279339849,
+      "required_pages_per_second_upper_estimate": 0.0017148305270634852
     },
     {
       "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": 27846.05367851171,
-      "required_pages_per_second_upper_estimate": 0.0021436662949819328
+      "required_token_positions_per_second": 28160.56279339849,
+      "required_pages_per_second_upper_estimate": 0.002167878077262869
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27846.05367851171,
-      "required_pages_per_second_upper_estimate": 0.3457547919404958
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28160.56279339849,
+      "required_pages_per_second_upper_estimate": 0.3496599425530935
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.46104179919490207,
-    "producer_p95_seconds": 0.3826264800009085,
-    "gpu_update_seconds": 0.9928875494963028,
-    "producer_batches_per_second": 2.1690007321380795,
-    "gpu_batches_per_second": 1.0071633998304295,
-    "producer_headroom": 2.1535738217882643,
+    "producer_batch_seconds": 0.11522391932066967,
+    "producer_p95_seconds": 0.3876982180008781,
+    "gpu_update_seconds": 0.9817985600231453,
+    "producer_batches_per_second": 8.678753559987722,
+    "gpu_batches_per_second": 1.0185388741825265,
+    "producer_headroom": 8.520787747991692,
     "sustainable": true,
-    "buffer_seconds": 60.56614051927447,
+    "buffer_seconds": 60.87151072143501,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311535.51436124084,
+    "generation_bytes_per_second": 309971.9510090193,
     "upload_bytes_per_second": 4059970.7144700503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.84826920631332,
+    "estimated_full_shards_per_hour": 66.51276490881857,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 348500, "loss": 0.6066722739487886, "lr": 0.0003, "grad_norm": 0.13950159604908102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51011587370, "tokens_per_second": 150502.17881202782, "data_wait_s": 9.357994713354856e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 348600, "loss": 0.5889758747071028, "lr": 0.0003, "grad_norm": 0.14416036690160747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51026133629, "tokens_per_second": 147350.11049295284, "data_wait_s": 1.1524003639351577e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348700, "loss": 0.6345112770795822, "lr": 0.0003, "grad_norm": 0.1687719276716483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51040607783, "tokens_per_second": 147103.8436790411, "data_wait_s": 2.0815998141188174e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 348800, "loss": 0.6054599732160568, "lr": 0.0003, "grad_norm": 0.1704089607327295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51055129564, "tokens_per_second": 148341.8375666809, "data_wait_s": 1.49589977809228e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26344d87df0f0e8b97095cfd8c40d043b990187018654d78f10ddbd7e5256a38
-size 718629
+oid sha256:7835b62f254de27ed55dc76dfa387601b41669b1776b139d7d34e601f9678b7e
+size 720747