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": 75709.02028014527,
-      "required_pages_per_second_upper_estimate": 0.0052377256921099825
+      "required_token_positions_per_second": 76650.92221783192,
+      "required_pages_per_second_upper_estimate": 0.005302888653672746
     },
     {
       "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": 9463.627535018159,
-      "required_pages_per_second_upper_estimate": 0.003962223225896709
+      "required_token_positions_per_second": 9581.36527722899,
+      "required_pages_per_second_upper_estimate": 0.004011517559916746
     },
     {
       "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": 18927.255070036317,
-      "required_pages_per_second_upper_estimate": 0.003942048085806301
+      "required_token_positions_per_second": 19162.73055445798,
+      "required_pages_per_second_upper_estimate": 0.003991091419305212
     },
     {
       "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": 28390.882605054474,
-      "required_pages_per_second_upper_estimate": 0.0017019900618053456
+      "required_token_positions_per_second": 28744.09583168697,
+      "required_pages_per_second_upper_estimate": 0.0017231646554165944
     },
     {
       "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": 28390.882605054474,
-      "required_pages_per_second_upper_estimate": 0.0022072662442479507
+      "required_token_positions_per_second": 28744.09583168697,
+      "required_pages_per_second_upper_estimate": 0.002234727018997835
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28390.882605054474,
-      "required_pages_per_second_upper_estimate": 0.36416775830292675
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28744.09583168697,
+      "required_pages_per_second_upper_estimate": 0.3686983983233536
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06541643721093315,
-    "producer_p95_seconds": 0.14882058800003506,
-    "gpu_update_seconds": 0.9738337615146132,
-    "producer_batches_per_second": 15.286677823427357,
-    "gpu_batches_per_second": 1.02686930718513,
-    "producer_headroom": 14.886682965850284,
+    "producer_batch_seconds": 0.04386008548437914,
+    "producer_p95_seconds": 0.13777216600010433,
+    "gpu_update_seconds": 0.9618670965298323,
+    "producer_batches_per_second": 22.799773164057147,
+    "gpu_batches_per_second": 1.0396446698382151,
+    "producer_headroom": 21.930351614850437,
     "sustainable": true,
-    "buffer_seconds": 60.37769321390602,
+    "buffer_seconds": 60.597627081379436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 72,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419943.3396037274,
+    "generation_bytes_per_second": 419178.491310858,
     "upload_bytes_per_second": 3619841.3405244793,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.11006489833704,
+    "estimated_full_shards_per_hour": 89.94594625944428,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -16,3 +16,4 @@
 {"step": 844100, "loss": 0.5470116138458252, "lr": 3e-05, "grad_norm": 0.29249404624772823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123552492254, "tokens_per_second": 150459.95035574777, "data_wait_s": 1.5677999954277766e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844200, "loss": 0.5173737406730652, "lr": 3e-05, "grad_norm": 0.2874194802671103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123567140903, "tokens_per_second": 151268.59734326255, "data_wait_s": 1.1069999800383812e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844300, "loss": 0.5937459468841553, "lr": 3e-05, "grad_norm": 0.31839938163663495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123581846732, "tokens_per_second": 151976.31945156935, "data_wait_s": 8.49900015964522e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 844400, "loss": 0.5504328608512878, "lr": 3e-05, "grad_norm": 0.27842239346038167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123596652416, "tokens_per_second": 154180.72814609038, "data_wait_s": 8.850999620335642e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:73bb63f6f015506ec16e7433afabe8aaa983f2727d393e28e2cd1b9990163fc9
-size 38289
+oid sha256:2ff62a917d929134d94023a1342ec59435a14bed67af0094db4788e6f965e40c
+size 40407