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": 75707.96999687672,
-      "required_pages_per_second_upper_estimate": 0.006562174959233571
+      "required_token_positions_per_second": 76058.77202635186,
+      "required_pages_per_second_upper_estimate": 0.00659258159004886
     },
     {
       "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": 9463.49624960959,
-      "required_pages_per_second_upper_estimate": 0.003775130145847132
+      "required_token_positions_per_second": 9507.346503293982,
+      "required_pages_per_second_upper_estimate": 0.0037926226676615535
     },
     {
       "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": 18926.99249921918,
-      "required_pages_per_second_upper_estimate": 0.004378426611003275
+      "required_token_positions_per_second": 19014.693006587964,
+      "required_pages_per_second_upper_estimate": 0.00439871457990683
     },
     {
       "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": 28390.48874882877,
-      "required_pages_per_second_upper_estimate": 0.001728831811420931
+      "required_token_positions_per_second": 28522.039509881946,
+      "required_pages_per_second_upper_estimate": 0.001736842562575567
     },
     {
       "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": 28390.48874882877,
-      "required_pages_per_second_upper_estimate": 0.002185578413787673
+      "required_token_positions_per_second": 28522.039509881946,
+      "required_pages_per_second_upper_estimate": 0.00219570555552936
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28390.48874882877,
-      "required_pages_per_second_upper_estimate": 0.35251485340686606
+      "required_token_positions_per_second": 28522.039509881946,
+      "required_pages_per_second_upper_estimate": 0.3541482735870711
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08124563793753481,
-    "producer_p95_seconds": 0.163327212000695,
-    "gpu_update_seconds": 0.9738472713380321,
-    "producer_batches_per_second": 12.308353105293403,
-    "gpu_batches_per_second": 1.0268550618065961,
-    "producer_headroom": 11.986456086254973,
+    "producer_batch_seconds": 0.06951040813281395,
+    "producer_p95_seconds": 0.17688230900057533,
+    "gpu_update_seconds": 0.9693556447960491,
+    "producer_batches_per_second": 14.386334749888018,
+    "gpu_batches_per_second": 1.0316131188470032,
+    "producer_headroom": 13.945474797729506,
     "sustainable": true,
-    "buffer_seconds": 60.37853082295799,
+    "buffer_seconds": 60.10004997735504,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346461.6766393653,
-    "upload_bytes_per_second": 5428834.2915613465,
+    "generation_bytes_per_second": 362095.1116716933,
+    "upload_bytes_per_second": 5500886.077133686,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.34261059175225,
+    "estimated_full_shards_per_hour": 77.69718182194804,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 2621511,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.8686625299997104,
+      "bytes": 5945555,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 359200, "loss": 0.6014014109969139, "lr": 0.0003, "grad_norm": 0.156046113746063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52577275485, "tokens_per_second": 151141.48779721805, "data_wait_s": 1.648499983275542e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359300, "loss": 0.5911583099514246, "lr": 0.0003, "grad_norm": 0.15084249697825547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52591963907, "tokens_per_second": 149539.5361958171, "data_wait_s": 1.127700033975998e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359400, "loss": 0.6074926014989614, "lr": 0.0003, "grad_norm": 0.15124208134927822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52606498900, "tokens_per_second": 148466.48859939937, "data_wait_s": 1.2035000509058591e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 359500, "loss": 0.5957230180501938, "lr": 0.0003, "grad_norm": 0.16041565634478688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52621131080, "tokens_per_second": 148511.89742624585, "data_wait_s": 1.3359999684325885e-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:f16a6002d1967660e4a0455209eb28e4b75e6cc0a2b0904abc5bf4d1e9792d21
-size 93357
+oid sha256:25cc806f390a7442a1620a14433332d20f69ef75bce86d9c6dbb991b2c59e263
+size 95475