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": 75708.99618841062,
-      "required_pages_per_second_upper_estimate": 0.006562263906914869
+      "required_token_positions_per_second": 74505.36266274125,
+      "required_pages_per_second_upper_estimate": 0.006457936003491129
     },
     {
       "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": 9463.624523551327,
-      "required_pages_per_second_upper_estimate": 0.003775181316240357
+      "required_token_positions_per_second": 9313.170332842656,
+      "required_pages_per_second_upper_estimate": 0.0037151628900760555
     },
     {
       "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": 18927.249047102654,
-      "required_pages_per_second_upper_estimate": 0.004378485958841028
+      "required_token_positions_per_second": 18626.34066568531,
+      "required_pages_per_second_upper_estimate": 0.004308876100606765
     },
     {
       "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": 28390.87357065398,
-      "required_pages_per_second_upper_estimate": 0.0017288552450510821
+      "required_token_positions_per_second": 27939.510998527967,
+      "required_pages_per_second_upper_estimate": 0.0017013696325250062
     },
     {
       "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": 28390.87357065398,
-      "required_pages_per_second_upper_estimate": 0.002185608038437033
+      "required_token_positions_per_second": 27939.510998527967,
+      "required_pages_per_second_upper_estimate": 0.002150860898183204
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28390.87357065398,
-      "required_pages_per_second_upper_estimate": 0.35251963160601935
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27939.510998527967,
+      "required_pages_per_second_upper_estimate": 0.34691521907356826
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08237946571080101,
-    "producer_p95_seconds": 0.21225158300148905,
-    "gpu_update_seconds": 0.9738340714030777,
-    "producer_batches_per_second": 12.138947386604464,
-    "gpu_batches_per_second": 1.026868980420066,
-    "producer_headroom": 11.821320556044775,
+    "producer_batch_seconds": 0.3090939151872476,
+    "producer_p95_seconds": 0.2605821119977918,
+    "gpu_update_seconds": 0.9895663528777106,
+    "producer_batches_per_second": 3.235262652757835,
+    "gpu_batches_per_second": 1.0105436559074061,
+    "producer_headroom": 3.2015070638910377,
     "sustainable": true,
-    "buffer_seconds": 60.37771242699081,
+    "buffer_seconds": 60.36354752554034,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473665.3828290354,
+    "generation_bytes_per_second": 424995.6776533209,
     "upload_bytes_per_second": 5755775.831788403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.63756478932663,
+    "estimated_full_shards_per_hour": 91.1941790313694,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 373600, "loss": 0.5689271334558725, "lr": 0.0003, "grad_norm": 0.15346934870243273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54684110789, "tokens_per_second": 149867.72203256062, "data_wait_s": 1.2165000953245908e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 373700, "loss": 0.5965260341763496, "lr": 0.0003, "grad_norm": 0.1407083360208807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54698785941, "tokens_per_second": 150111.3880085052, "data_wait_s": 1.3315999240148813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 373800, "loss": 0.5403846129775047, "lr": 0.0003, "grad_norm": 0.2319163103574092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54713423612, "tokens_per_second": 148937.27212668382, "data_wait_s": 9.621999197406694e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 373900, "loss": 0.5923236533999443, "lr": 0.0003, "grad_norm": 0.16723194428624796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54727941165, "tokens_per_second": 148546.19985122856, "data_wait_s": 1.1596999684115872e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9136bb093e5c47224d5bc061e1136376ebe2a0d337e1f155c13b79744e577132
-size 400698
+oid sha256:a590c25ae883621b72b07ed4d801bc146b767ce3757d33a802940fa656cbdf78
+size 402816