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": 75945.39485631048,
-      "required_pages_per_second_upper_estimate": 0.006582754344301484
+      "required_token_positions_per_second": 75945.22228342804,
+      "required_pages_per_second_upper_estimate": 0.006582739386121416
     },
     {
       "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": 9493.17435703881,
-      "required_pages_per_second_upper_estimate": 0.003786969186627896
+      "required_token_positions_per_second": 9493.152785428505,
+      "required_pages_per_second_upper_estimate": 0.003786960581390021
     },
     {
       "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": 18986.34871407762,
-      "required_pages_per_second_upper_estimate": 0.004392157626677072
+      "required_token_positions_per_second": 18986.30557085701,
+      "required_pages_per_second_upper_estimate": 0.004392147646252279
     },
     {
       "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": 28479.52307111643,
-      "required_pages_per_second_upper_estimate": 0.001734253534521262
+      "required_token_positions_per_second": 28479.45835628552,
+      "required_pages_per_second_upper_estimate": 0.0017342495937275918
     },
     {
       "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": 28479.52307111643,
-      "required_pages_per_second_upper_estimate": 0.0021924325223801554
+      "required_token_positions_per_second": 28479.45835628552,
+      "required_pages_per_second_upper_estimate": 0.00219242754045333
     },
     {
       "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": 28479.52307111643,
-      "required_pages_per_second_upper_estimate": 0.3536203617109705
+      "required_token_positions_per_second": 28479.45835628552,
+      "required_pages_per_second_upper_estimate": 0.3536195581693572
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14992873260933948,
-    "producer_p95_seconds": 0.35971107899968047,
-    "gpu_update_seconds": 0.970802774012752,
-    "producer_batches_per_second": 6.66983561186795,
-    "gpu_batches_per_second": 1.0300753425606348,
-    "producer_headroom": 6.475094914210446,
+    "producer_batch_seconds": 0.29812809410918817,
+    "producer_p95_seconds": 0.3895399949979037,
+    "gpu_update_seconds": 0.9708049800005409,
+    "producer_batches_per_second": 3.354262881490646,
+    "gpu_batches_per_second": 1.030073001891114,
+    "producer_headroom": 3.2563351095820834,
     "sustainable": true,
-    "buffer_seconds": 60.18977198879062,
+    "buffer_seconds": 60.18990876003354,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349756.1616309234,
-    "upload_bytes_per_second": 5063000.7891543815,
+    "generation_bytes_per_second": 300136.3070807675,
+    "upload_bytes_per_second": 5053720.341839957,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.04953037925507,
+    "estimated_full_shards_per_hour": 64.40226468388815,
     "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": 1.1448558080010116,
-      "bytes": 6260573,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1358172180007386,
       "bytes": 2620953,
@@ -562,6 +556,12 @@
       "bytes": 7556763,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9536960379991797,
+      "bytes": 4538069,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 276000, "loss": 0.6485577970743179, "lr": 0.0003, "grad_norm": 0.1465950070699948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40397759181, "tokens_per_second": 150855.9904269503, "data_wait_s": 1.0570998711045831e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276100, "loss": 0.6443565767258406, "lr": 0.0003, "grad_norm": 0.1798039695456819, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40412682300, "tokens_per_second": 152806.60512173743, "data_wait_s": 1.4905999705661088e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276200, "loss": 0.6205026488751173, "lr": 0.0003, "grad_norm": 0.1742606156034186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40427451945, "tokens_per_second": 127623.37690401972, "data_wait_s": 1.6697995306458324e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 276300, "loss": 0.572891453281045, "lr": 0.0003, "grad_norm": 0.20148212619272768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40442115845, "tokens_per_second": 149892.38877773884, "data_wait_s": 1.2851000064983964e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5405f9555674770fa8a81bfa0c0dc0886c3d0071be7a837f0408df1160aba272
-size 834567
+oid sha256:6cfb47d3c93ef2b7114e950e436e78d9ecd6abb5349b1b5595fea0e0870e0b96
+size 836685