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": 75563.37926886792,
-      "required_pages_per_second_upper_estimate": 0.006549642201391617
+      "required_token_positions_per_second": 76330.36581736231,
+      "required_pages_per_second_upper_estimate": 0.006616122651505476
     },
     {
       "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": 9445.42240860849,
-      "required_pages_per_second_upper_estimate": 0.003767920220443789
+      "required_token_positions_per_second": 9541.295727170289,
+      "required_pages_per_second_upper_estimate": 0.0038061655206519424
     },
     {
       "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": 18890.84481721698,
-      "required_pages_per_second_upper_estimate": 0.004370064480949541
+      "required_token_positions_per_second": 19082.591454340578,
+      "required_pages_per_second_upper_estimate": 0.004414421690822529
     },
     {
       "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": 28336.267225825475,
-      "required_pages_per_second_upper_estimate": 0.001725530005148376
+      "required_token_positions_per_second": 28623.88718151087,
+      "required_pages_per_second_upper_estimate": 0.0017430445514243324
     },
     {
       "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": 28336.267225825475,
-      "required_pages_per_second_upper_estimate": 0.0021814042908520972
+      "required_token_positions_per_second": 28623.88718151087,
+      "required_pages_per_second_upper_estimate": 0.002203546071223754
     },
     {
       "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": 28336.267225825475,
-      "required_pages_per_second_upper_estimate": 0.35184160355892913
+      "required_token_positions_per_second": 28623.88718151087,
+      "required_pages_per_second_upper_estimate": 0.35541288080647243
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1692556944218495,
-    "producer_p95_seconds": 0.20813430700036406,
-    "gpu_update_seconds": 0.9757107306922137,
-    "producer_batches_per_second": 5.9082207155029005,
-    "gpu_batches_per_second": 1.0248939245451922,
-    "producer_headroom": 5.7647143514142085,
+    "producer_batch_seconds": 0.0736773940702875,
+    "producer_p95_seconds": 0.19037785900036397,
+    "gpu_update_seconds": 0.9659065459794983,
+    "producer_batches_per_second": 13.572684167493898,
+    "gpu_batches_per_second": 1.0352968453960818,
+    "producer_headroom": 13.109944483894653,
     "sustainable": true,
-    "buffer_seconds": 60.49406530291725,
+    "buffer_seconds": 60.85211239670839,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286754.6043461387,
+    "generation_bytes_per_second": 314610.6813273276,
     "upload_bytes_per_second": 5592918.494714557,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.530862787133415,
+    "estimated_full_shards_per_hour": 67.50812845101234,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -49,3 +49,4 @@
 {"step": 285100, "loss": 0.5674477610737085, "lr": 0.0003, "grad_norm": 0.15156043304043362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41730571018, "tokens_per_second": 130993.04942865677, "data_wait_s": 1.2206999599584378e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285200, "loss": 0.659591244533658, "lr": 0.0003, "grad_norm": 0.1874568876131716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41745218309, "tokens_per_second": 149745.86236816508, "data_wait_s": 1.4650999219156802e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285300, "loss": 0.6301170792430639, "lr": 0.0003, "grad_norm": 0.171539052693542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41759958804, "tokens_per_second": 150894.29399413915, "data_wait_s": 1.0748999557108618e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 285400, "loss": 0.5829807613044977, "lr": 0.0003, "grad_norm": 0.1464344673972657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41774615383, "tokens_per_second": 150879.59424880205, "data_wait_s": 1.4166000255499966e-05, "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 sha256:02d01971147edd78982f99da83d2f1708f18fef92ce9f840f153cee7e9f86d0e
-size 108966
+oid sha256:efbde2646a5f238ba89bf51cfedeb87559160061e7683162f16e82d44d0bce2a
+size 111084