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": 75813.1127586283,
-      "required_pages_per_second_upper_estimate": 0.006571288467340303
+      "required_token_positions_per_second": 74282.69745693586,
+      "required_pages_per_second_upper_estimate": 0.006438635947791723
     },
     {
       "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": 9476.639094828537,
-      "required_pages_per_second_upper_estimate": 0.003780373023308017
+      "required_token_positions_per_second": 9285.337182116982,
+      "required_pages_per_second_upper_estimate": 0.0037040598301088965
     },
     {
       "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": 18953.278189657074,
-      "required_pages_per_second_upper_estimate": 0.0043845073428737304
+      "required_token_positions_per_second": 18570.674364233964,
+      "required_pages_per_second_upper_estimate": 0.00429599868145944
     },
     {
       "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": 28429.91728448561,
-      "required_pages_per_second_upper_estimate": 0.0017312328023768933
+      "required_token_positions_per_second": 27856.011546350943,
+      "required_pages_per_second_upper_estimate": 0.001696284953975194
     },
     {
       "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": 28429.91728448561,
-      "required_pages_per_second_upper_estimate": 0.002188613731607709
+      "required_token_positions_per_second": 27856.011546350943,
+      "required_pages_per_second_upper_estimate": 0.0021444328792133397
     },
     {
       "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": 28429.91728448561,
-      "required_pages_per_second_upper_estimate": 0.3530044238609038
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27856.011546350943,
+      "required_pages_per_second_upper_estimate": 0.34587843533221924
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07421587351562664,
-    "producer_p95_seconds": 0.20151812700078153,
-    "gpu_update_seconds": 0.972496673955773,
-    "producer_batches_per_second": 13.47420642821705,
-    "gpu_batches_per_second": 1.0282811517826105,
-    "producer_headroom": 13.103620935634577,
+    "producer_batch_seconds": 0.8249203895000008,
+    "producer_p95_seconds": 0.22316516099999717,
+    "gpu_update_seconds": 0.9925326155898225,
+    "producer_batches_per_second": 1.212238190167803,
+    "gpu_batches_per_second": 1.0075235657679016,
+    "producer_headroom": 1.203185941605122,
     "sustainable": true,
-    "buffer_seconds": 60.29479378525792,
+    "buffer_seconds": 60.54448955097917,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 251046.91612777958,
-    "upload_bytes_per_second": 4577422.298684732,
+    "generation_bytes_per_second": 253417.46506770217,
+    "upload_bytes_per_second": 4615836.474392189,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.86882412791291,
+    "estimated_full_shards_per_hour": 54.37748874686526,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 800987,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 0.8824849620004898,
+      "bytes": 7297737,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 6500, "loss": 0.783741295337677, "lr": 0.0003, "grad_norm": 0.16574513455776693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 950845409, "tokens_per_second": 149441.65644197306, "data_wait_s": 1.2858999980380759e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 6600, "loss": 0.745708305388689, "lr": 0.0003, "grad_norm": 0.17421476492957289, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 965517196, "tokens_per_second": 149686.27890422332, "data_wait_s": 1.0661000487743877e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 6700, "loss": 0.7228674329817295, "lr": 0.0003, "grad_norm": 0.16560404275845136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 980035792, "tokens_per_second": 148625.6122175926, "data_wait_s": 1.119799981097458e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 6800, "loss": 0.7337647527456284, "lr": 0.0003, "grad_norm": 0.16146701328371382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 994691340, "tokens_per_second": 136329.71091196107, "data_wait_s": 1.1813999662990682e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8fd0168dab2a2bde066b24b3b588fd90bced9146b4b76cdd6df6d8436a8a1c2
-size 140460
+oid sha256:dfa531e41d903b92b0390068f69598a836325afc2b86cae1aa21fb950c6f8793
+size 142543