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": 75227.67341060555,
-      "required_pages_per_second_upper_estimate": 0.006520544068436154
+      "required_token_positions_per_second": 73724.98782096013,
+      "required_pages_per_second_upper_estimate": 0.006390295090047497
     },
     {
       "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": 9403.459176325694,
-      "required_pages_per_second_upper_estimate": 0.003751180459679948
+      "required_token_positions_per_second": 9215.623477620016,
+      "required_pages_per_second_upper_estimate": 0.0036762499910722897
     },
     {
       "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": 18806.918352651388,
-      "required_pages_per_second_upper_estimate": 0.00435064957042762
+      "required_token_positions_per_second": 18431.246955240033,
+      "required_pages_per_second_upper_estimate": 0.0042637446042272835
     },
     {
       "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": 28210.377528977082,
-      "required_pages_per_second_upper_estimate": 0.0017178639831025026
+      "required_token_positions_per_second": 27646.87043286005,
+      "required_pages_per_second_upper_estimate": 0.0016835493574422187
     },
     {
       "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": 28210.377528977082,
-      "required_pages_per_second_upper_estimate": 0.002171712953503723
+      "required_token_positions_per_second": 27646.87043286005,
+      "required_pages_per_second_upper_estimate": 0.0021283326173571547
     },
     {
       "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": 28210.377528977082,
-      "required_pages_per_second_upper_estimate": 0.35027847484978436
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27646.87043286005,
+      "required_pages_per_second_upper_estimate": 0.3432816026529428
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11117438153885928,
-    "producer_p95_seconds": 0.33460860999912256,
-    "gpu_update_seconds": 0.9800648705108813,
-    "producer_batches_per_second": 8.994878011985751,
-    "gpu_batches_per_second": 1.0203406224311733,
-    "producer_headroom": 8.815563954077989,
+    "producer_batch_seconds": 0.14045185108597025,
+    "producer_p95_seconds": 0.3432903819993953,
+    "gpu_update_seconds": 1.0000408569621901,
+    "producer_batches_per_second": 7.119877682408774,
+    "gpu_batches_per_second": 0.9999591447070331,
+    "producer_headroom": 7.120168578982042,
     "sustainable": true,
-    "buffer_seconds": 60.764021971674644,
+    "buffer_seconds": 60.00245141773141,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 261913.8135438077,
+    "generation_bytes_per_second": 262955.49392596853,
     "upload_bytes_per_second": 4487133.915199309,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.20060734496759,
+    "estimated_full_shards_per_hour": 56.42412770590107,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 266400, "loss": 0.6018778197467327, "lr": 0.0003, "grad_norm": 0.15254020031180637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38993517013, "tokens_per_second": 147141.5573219985, "data_wait_s": 9.244002285413444e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 266500, "loss": 0.5274950340390205, "lr": 0.0003, "grad_norm": 0.1670249046064422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39008052295, "tokens_per_second": 116731.62916602202, "data_wait_s": 1.4427998394239694e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 266600, "loss": 0.582127632573247, "lr": 0.0003, "grad_norm": 0.1489782496607366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39022636228, "tokens_per_second": 148984.46680246398, "data_wait_s": 1.3501001376425847e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 266700, "loss": 0.671088632196188, "lr": 0.0003, "grad_norm": 0.19726455650691796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39037307794, "tokens_per_second": 149247.32758563315, "data_wait_s": 1.2584001524373889e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e8ec5c2f9034d29be8f91374a660cec6bd44ac874615171fcb9520419935ce0
-size 629673
+oid sha256:f2f665746d0210c0fb96f5e47fa56f0ff09a7224e79f6b463c2556d9b697b346
+size 631791