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": 77016.84597947262,
-      "required_pages_per_second_upper_estimate": 0.005328204108563154
+      "required_token_positions_per_second": 76159.37984808919,
+      "required_pages_per_second_upper_estimate": 0.00526888261199851
     },
     {
       "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": 9627.105747434078,
-      "required_pages_per_second_upper_estimate": 0.004030668139621982
+      "required_token_positions_per_second": 9519.922481011148,
+      "required_pages_per_second_upper_estimate": 0.003985792744211823
     },
     {
       "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": 19254.211494868156,
-      "required_pages_per_second_upper_estimate": 0.004010144486678018
+      "required_token_positions_per_second": 19039.844962022296,
+      "required_pages_per_second_upper_estimate": 0.00396549759111186
     },
     {
       "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": 28881.317242302233,
-      "required_pages_per_second_upper_estimate": 0.0017313908694580175
+      "required_token_positions_per_second": 28559.767443033445,
+      "required_pages_per_second_upper_estimate": 0.0017121144499699684
     },
     {
       "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": 28881.317242302233,
-      "required_pages_per_second_upper_estimate": 0.002245395380099974
+      "required_token_positions_per_second": 28559.767443033445,
+      "required_pages_per_second_upper_estimate": 0.00222039629755492
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28881.317242302233,
-      "required_pages_per_second_upper_estimate": 0.3704585272418547
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28559.767443033445,
+      "required_pages_per_second_upper_estimate": 0.36633403167010997
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09208427330463564,
-    "producer_p95_seconds": 0.3300673630001256,
-    "gpu_update_seconds": 0.9572970570575013,
-    "producer_batches_per_second": 10.859617653622278,
-    "gpu_batches_per_second": 1.0446078284976212,
-    "producer_headroom": 10.395880020582295,
+    "producer_batch_seconds": 0.3242035134143748,
+    "producer_p95_seconds": 0.32502524200390326,
+    "gpu_update_seconds": 0.9680751096852558,
+    "producer_batches_per_second": 3.084482303934406,
+    "gpu_batches_per_second": 1.0329776997624944,
+    "producer_headroom": 2.986010544703531,
     "sustainable": true,
-    "buffer_seconds": 60.309714594622584,
+    "buffer_seconds": 60.02065680048586,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419520.58599200175,
+    "generation_bytes_per_second": 408303.6400887465,
     "upload_bytes_per_second": 4786256.147254406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.01935181446113,
+    "estimated_full_shards_per_hour": 87.61245634075924,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 960400, "loss": 0.5473693013191223, "lr": 3e-05, "grad_norm": 0.3466411055769713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140576307698, "tokens_per_second": 150334.93396943517, "data_wait_s": 1.3853998098056763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960500, "loss": 0.5665472149848938, "lr": 3e-05, "grad_norm": 0.3649950479541535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140590879064, "tokens_per_second": 151317.5549284369, "data_wait_s": 1.82330040843226e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960600, "loss": 0.516963005065918, "lr": 3e-05, "grad_norm": 0.33967234324128964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140605578391, "tokens_per_second": 152385.32862985204, "data_wait_s": 2.1840998670086265e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 960700, "loss": 0.53617262840271, "lr": 3e-05, "grad_norm": 0.35134323500284453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140620181073, "tokens_per_second": 152239.77556202843, "data_wait_s": 1.535100454930216e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ddaab3c6eb15408135c4a4636707f002dd4e67942eb0d86d50de3f607447ec99
-size 759654
+oid sha256:70161748211f6aa5007594e3d78c62d273061065c4112f599997aee07ae232ec
+size 761772