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": 75519.04554253354,
-      "required_pages_per_second_upper_estimate": 0.006545799466355755
+      "required_token_positions_per_second": 73964.52914600106,
+      "required_pages_per_second_upper_estimate": 0.006411057924990107
     },
     {
       "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": 9439.880692816692,
-      "required_pages_per_second_upper_estimate": 0.003765709547158406
+      "required_token_positions_per_second": 9245.566143250133,
+      "required_pages_per_second_upper_estimate": 0.0036881945680748893
     },
     {
       "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": 18879.761385633385,
-      "required_pages_per_second_upper_estimate": 0.00436750052411441
+      "required_token_positions_per_second": 18491.132286500266,
+      "required_pages_per_second_upper_estimate": 0.004277598021668505
     },
     {
       "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": 28319.642078450077,
-      "required_pages_per_second_upper_estimate": 0.0017245176208986212
+      "required_token_positions_per_second": 27736.698429750402,
+      "required_pages_per_second_upper_estimate": 0.001689019411161765
     },
     {
       "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": 28319.642078450077,
-      "required_pages_per_second_upper_estimate": 0.0021801244409857855
+      "required_token_positions_per_second": 27736.698429750402,
+      "required_pages_per_second_upper_estimate": 0.002135247825217584
     },
     {
       "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": 28319.642078450077,
-      "required_pages_per_second_upper_estimate": 0.35163517486931567
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27736.698429750402,
+      "required_pages_per_second_upper_estimate": 0.34439696573935463
     }
   ],
   "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": 0.09894992978877326,
-    "producer_p95_seconds": 0.3828367390015046,
-    "gpu_update_seconds": 0.9762835251734637,
-    "producer_batches_per_second": 10.106121370016968,
-    "gpu_batches_per_second": 1.0242926098976446,
-    "producer_headroom": 9.86643979695104,
+    "producer_batch_seconds": 0.2995210348516366,
+    "producer_p95_seconds": 0.39149500399798853,
+    "gpu_update_seconds": 0.9968021273341148,
+    "producer_batches_per_second": 3.338663678480329,
+    "gpu_batches_per_second": 1.0032081318630788,
+    "producer_headroom": 3.3279870571623333,
     "sustainable": true,
-    "buffer_seconds": 60.52957856075475,
+    "buffer_seconds": 60.804929767381005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449263.2120419722,
+    "generation_bytes_per_second": 424182.87732027564,
     "upload_bytes_per_second": 5609048.389935914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.4014269918859,
+    "estimated_full_shards_per_hour": 91.01977100092127,
     "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

@@ -387,3 +387,4 @@
 {"step": 393700, "loss": 0.5777887590229511, "lr": 0.0003, "grad_norm": 0.1614428000226198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57626484539, "tokens_per_second": 146853.8779125341, "data_wait_s": 1.1125004675704986e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 393800, "loss": 0.5807689875364304, "lr": 0.0003, "grad_norm": 0.16996640827632079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57641072471, "tokens_per_second": 149420.9259623452, "data_wait_s": 1.029099803417921e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 393900, "loss": 0.5727016478776932, "lr": 0.0003, "grad_norm": 0.15051099218322683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57655563623, "tokens_per_second": 146516.30754007335, "data_wait_s": 1.6882993804756552e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 394000, "loss": 0.5279850102961063, "lr": 0.0003, "grad_norm": 0.15053944217458426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57670155952, "tokens_per_second": 147895.81421272593, "data_wait_s": 1.1159005225636065e-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:ea2e20bf2befe0032e518c5484f260c296ef150e506c2ea7b227f370a4a717bf
-size 829548
+oid sha256:bba9fa272c4bd76af835c24f209354eda438c9810def52d4db46cbe968a4000a
+size 831666