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": 76378.33191482387,
-      "required_pages_per_second_upper_estimate": 0.006620280231264481
+      "required_token_positions_per_second": 76135.90636634141,
+      "required_pages_per_second_upper_estimate": 0.006599267399143957
     },
     {
       "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": 9547.291489352983,
-      "required_pages_per_second_upper_estimate": 0.0038085573198312524
+      "required_token_positions_per_second": 9516.988295792677,
+      "required_pages_per_second_upper_estimate": 0.003796468922846927
     },
     {
       "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": 19094.582978705967,
-      "required_pages_per_second_upper_estimate": 0.004417195718940841
+      "required_token_positions_per_second": 19033.976591585353,
+      "required_pages_per_second_upper_estimate": 0.004403175497916471
     },
     {
       "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": 28641.87446805895,
-      "required_pages_per_second_upper_estimate": 0.0017441398828031117
+      "required_token_positions_per_second": 28550.964887378028,
+      "required_pages_per_second_upper_estimate": 0.0017386039663053513
     },
     {
       "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": 28641.87446805895,
-      "required_pages_per_second_upper_estimate": 0.0022049307823342216
+      "required_token_positions_per_second": 28550.964887378028,
+      "required_pages_per_second_upper_estimate": 0.0021979323111595782
     },
     {
       "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": 28641.87446805895,
-      "required_pages_per_second_upper_estimate": 0.35563622270582407
+      "required_token_positions_per_second": 28550.964887378028,
+      "required_pages_per_second_upper_estimate": 0.3545074299685614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0629032884064884,
-    "producer_p95_seconds": 0.20650791799926083,
-    "gpu_update_seconds": 0.9652999502819793,
-    "producer_batches_per_second": 15.897420076640241,
-    "gpu_batches_per_second": 1.0359474272301414,
-    "producer_headroom": 15.345778809592565,
+    "producer_batch_seconds": 0.237465793195355,
+    "producer_p95_seconds": 0.2460462200033362,
+    "gpu_update_seconds": 0.9683735771824225,
+    "producer_batches_per_second": 4.211132839572116,
+    "gpu_batches_per_second": 1.0326593202900038,
+    "producer_headroom": 4.077949771846821,
     "sustainable": true,
-    "buffer_seconds": 60.8138968677647,
+    "buffer_seconds": 60.03916178531019,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 543062.8026689846,
-    "upload_bytes_per_second": 5443648.407667226,
+    "generation_bytes_per_second": 514803.82270870195,
+    "upload_bytes_per_second": 5422289.28426314,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.52863559772638,
+    "estimated_full_shards_per_hour": 110.46491633363527,
     "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": [
-    {
-      "seconds": 1.2733040839993919,
-      "bytes": 13322526,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 0.8405508110008668,
       "bytes": 1090490,
@@ -562,6 +556,12 @@
       "bytes": 1161800,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1932739869989746,
+      "bytes": 11243706,
+      "pages": 26,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -96,3 +96,4 @@
 {"step": 289800, "loss": 0.5929397474974394, "lr": 0.0003, "grad_norm": 0.15009253358194619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42418609542, "tokens_per_second": 150659.41350640537, "data_wait_s": 1.4784000086365268e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 289900, "loss": 0.5935659389942884, "lr": 0.0003, "grad_norm": 0.14504279381198398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42433278735, "tokens_per_second": 150763.6578952195, "data_wait_s": 1.367900040349923e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 290000, "loss": 0.6066216714680195, "lr": 0.0003, "grad_norm": 0.1527091140648715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42447903231, "tokens_per_second": 149593.93962431414, "data_wait_s": 1.0943000233964995e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 290100, "loss": 0.6231864560395479, "lr": 0.0003, "grad_norm": 0.15283337775739578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42462659362, "tokens_per_second": 136612.09095894187, "data_wait_s": 8.132999937515706e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:30f68bb0336eb7cd67c0235d65882993f79138ed8c54f8cd7e85f31d93f78a8d
-size 209295
+oid sha256:d35caa28ff2677ded33ffd5b36d370e68c6d90a06ecb09ad9b9af69492c927b6
+size 211413