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": 6180.204418947272,
-      "required_pages_per_second_upper_estimate": 0.0005356844554494531
+      "required_token_positions_per_second": 71671.4455961838,
+      "required_pages_per_second_upper_estimate": 0.006212299254659069
     },
     {
       "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": 772.525552368409,
-      "required_pages_per_second_upper_estimate": 0.0003081719931260607
+      "required_token_positions_per_second": 8958.930699522974,
+      "required_pages_per_second_upper_estimate": 0.0035738514039903363
     },
     {
       "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": 1545.051104736818,
-      "required_pages_per_second_upper_estimate": 0.0003574203811101406
+      "required_token_positions_per_second": 17917.86139904595,
+      "required_pages_per_second_upper_estimate": 0.004144981891078978
     },
     {
       "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": 2317.576657105227,
-      "required_pages_per_second_upper_estimate": 0.00014112825903271528
+      "required_token_positions_per_second": 26876.79209856892,
+      "required_pages_per_second_upper_estimate": 0.001636655627172659
     },
     {
       "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": 2317.576657105227,
-      "required_pages_per_second_upper_estimate": 0.00017841346652675512
+      "required_token_positions_per_second": 26876.79209856892,
+      "required_pages_per_second_upper_estimate": 0.0020690498554701586
     },
     {
       "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": 3,
-      "required_token_positions_per_second": 2317.576657105227,
-      "required_pages_per_second_upper_estimate": 0.02877654565113211
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 26876.79209856892,
+      "required_pages_per_second_upper_estimate": 0.3337198070274398
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 6,
+    "prefetch_batches": 59,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04482335239682499,
-    "producer_p95_seconds": 0.10684306999996807,
-    "gpu_update_seconds": 11.929702482650038,
-    "producer_batches_per_second": 22.309799390882546,
-    "gpu_batches_per_second": 0.0838243871927527,
-    "producer_headroom": 266.1492691808358,
+    "producer_batch_seconds": 0.05351816681904677,
+    "producer_p95_seconds": 0.12730015699992236,
+    "gpu_update_seconds": 1.0286941945527845,
+    "producer_batches_per_second": 18.685243898229086,
+    "gpu_batches_per_second": 0.9721061956947671,
+    "producer_headroom": 19.221401921911102,
     "sustainable": true,
-    "buffer_seconds": 71.57821489590023,
+    "buffer_seconds": 60.69295747861429,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 22,
+    "upload_pages_per_commit": 25,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -1 +1,2 @@
 {"step": 115100, "loss": 0.609047619625926, "lr": 0.0003, "grad_norm": 0.1299152272394879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16843057721, "tokens_per_second": 46429.21698519966, "data_wait_s": 1.3538999951379083e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 115200, "loss": 0.5842613279819489, "lr": 0.0003, "grad_norm": 0.13957367498710943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16857707507, "tokens_per_second": 151021.34513643512, "data_wait_s": 2.3415999976350577e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25840cbf676c147aae37e850375f5211c436c332f6eaee8840f6bf2c43007472
-size 2283
+oid sha256:4089c0dcf40e1c179d770b1824aa926a16d23bd9a0211f5cc44a3cbb56294d28
+size 4401