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": 74690.47620156672,
-      "required_pages_per_second_upper_estimate": 0.006473981175870547
+      "required_token_positions_per_second": 74610.4535354208,
+      "required_pages_per_second_upper_estimate": 0.006467045013984612
     },
     {
       "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": 9336.30952519584,
-      "required_pages_per_second_upper_estimate": 0.0037243934598674964
+      "required_token_positions_per_second": 9326.3066919276,
+      "required_pages_per_second_upper_estimate": 0.003720403180121111
     },
     {
       "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": 18672.61905039168,
-      "required_pages_per_second_upper_estimate": 0.00431958179043683
+      "required_token_positions_per_second": 18652.6133838552,
+      "required_pages_per_second_upper_estimate": 0.004314953831571321
     },
     {
       "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": 28008.92857558752,
-      "required_pages_per_second_upper_estimate": 0.0017055967987620537
+      "required_token_positions_per_second": 27978.9200757828,
+      "required_pages_per_second_upper_estimate": 0.0017037694385663767
     },
     {
       "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": 28008.92857558752,
-      "required_pages_per_second_upper_estimate": 0.0021562048554254008
+      "required_token_positions_per_second": 27978.9200757828,
+      "required_pages_per_second_upper_estimate": 0.0021538947180415892
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28008.92857558752,
-      "required_pages_per_second_upper_estimate": 0.3477771530549626
+      "required_token_positions_per_second": 27978.9200757828,
+      "required_pages_per_second_upper_estimate": 0.3474045479193762
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10763798867193941,
-    "producer_p95_seconds": 0.2585400319985638,
-    "gpu_update_seconds": 0.9871138028499203,
-    "producer_batches_per_second": 9.290400279104194,
-    "gpu_batches_per_second": 1.0130544189665625,
-    "producer_headroom": 9.170682349504503,
+    "producer_batch_seconds": 0.36135673284397285,
+    "producer_p95_seconds": 0.2937687880003068,
+    "gpu_update_seconds": 0.988172521495237,
+    "producer_batches_per_second": 2.767348465129558,
+    "gpu_batches_per_second": 1.0119690420928387,
+    "producer_headroom": 2.7346177106430494,
     "sustainable": true,
-    "buffer_seconds": 60.21394197384514,
+    "buffer_seconds": 60.27852381120946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325497.9800483211,
-    "upload_bytes_per_second": 5231836.8239477705,
+    "generation_bytes_per_second": 309604.65499638574,
+    "upload_bytes_per_second": 5274027.979814421,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.84428931319451,
+    "estimated_full_shards_per_hour": 66.43395173472098,
     "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": 1.0858683609985746,
-      "bytes": 5956157,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2466707620005764,
       "bytes": 7844213,
@@ -562,6 +556,12 @@
       "bytes": 7676145,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2521359960010159,
+      "bytes": 10213095,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -241,3 +241,4 @@
 {"step": 339100, "loss": 0.6156422272324562, "lr": 0.0003, "grad_norm": 0.16153957412485814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49635297006, "tokens_per_second": 150227.11843808705, "data_wait_s": 1.4793000445934013e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339200, "loss": 0.5894141495227814, "lr": 0.0003, "grad_norm": 0.15781046016238667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49649972896, "tokens_per_second": 148500.65319454408, "data_wait_s": 1.6009998944355175e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339300, "loss": 0.5361258052289486, "lr": 0.0003, "grad_norm": 0.1372399859470047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49664499446, "tokens_per_second": 147927.6856600261, "data_wait_s": 1.0453997674630955e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 339400, "loss": 0.5926066152751446, "lr": 0.0003, "grad_norm": 0.17307040087604378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49679000476, "tokens_per_second": 148209.3000903051, "data_wait_s": 1.629799953661859e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a9e94f1fea559aad39b4171315ea93936638e4dbee3235e27cb06079e93e94d
-size 517971
+oid sha256:8bf1b056b84ab0b650d6614c9d4b741a7b1e28002f76ada1d6fb0d4bdc7c9eaf
+size 520089