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": 75646.86491128222,
-      "required_pages_per_second_upper_estimate": 0.006556878525283662
+      "required_token_positions_per_second": 74732.41838315224,
+      "required_pages_per_second_upper_estimate": 0.006477616617868893
     },
     {
       "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": 9455.858113910277,
-      "required_pages_per_second_upper_estimate": 0.00377208317931637
+      "required_token_positions_per_second": 9341.55229789403,
+      "required_pages_per_second_upper_estimate": 0.0037264848802832417
     },
     {
       "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": 18911.716227820554,
-      "required_pages_per_second_upper_estimate": 0.004374892714468397
+      "required_token_positions_per_second": 18683.10459578806,
+      "required_pages_per_second_upper_estimate": 0.004322007436824986
     },
     {
       "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": 28367.574341730833,
-      "required_pages_per_second_upper_estimate": 0.0017274364442512694
+      "required_token_positions_per_second": 28024.656893682088,
+      "required_pages_per_second_upper_estimate": 0.0017065545708139051
     },
     {
       "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": 28367.574341730833,
-      "required_pages_per_second_upper_estimate": 0.0021838143992981354
+      "required_token_positions_per_second": 28024.656893682088,
+      "required_pages_per_second_upper_estimate": 0.0021574156648910916
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28367.574341730833,
-      "required_pages_per_second_upper_estimate": 0.3522303331602969
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28024.656893682088,
+      "required_pages_per_second_upper_estimate": 0.3479724461264026
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11914833331232444,
-    "producer_p95_seconds": 0.2559921789979853,
-    "gpu_update_seconds": 0.9746339135992926,
-    "producer_batches_per_second": 8.392899608412417,
-    "gpu_batches_per_second": 1.0260262710406118,
-    "producer_headroom": 8.180004591792965,
+    "producer_batch_seconds": 0.23974966496081151,
+    "producer_p95_seconds": 0.2606211339989386,
+    "gpu_update_seconds": 0.9865598035647315,
+    "producer_batches_per_second": 4.171017299079254,
+    "gpu_batches_per_second": 1.0136232962124598,
+    "producer_headroom": 4.114958007244725,
     "sustainable": true,
-    "buffer_seconds": 60.427302643156146,
+    "buffer_seconds": 60.18014801744862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458629.9039014619,
-    "upload_bytes_per_second": 5859882.753710557,
+    "generation_bytes_per_second": 457685.121456541,
+    "upload_bytes_per_second": 5920682.190912063,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.41130101950543,
+    "estimated_full_shards_per_hour": 98.20857270023511,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 6115749,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.5552335219981615,
+      "bytes": 12252293,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -102,3 +102,4 @@
 {"step": 95200, "loss": 0.5910352412611246, "lr": 0.0003, "grad_norm": 0.12485652502635318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13931717642, "tokens_per_second": 148402.46917393446, "data_wait_s": 9.585000952938572e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 95300, "loss": 0.625568326562643, "lr": 0.0003, "grad_norm": 0.13915032065916308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13946370067, "tokens_per_second": 148887.99585646426, "data_wait_s": 1.2813001376343891e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 95400, "loss": 0.5794907696545124, "lr": 0.0003, "grad_norm": 0.13514553789882475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13961003735, "tokens_per_second": 148911.6350202893, "data_wait_s": 8.459999662591144e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 95500, "loss": 0.6024086624383926, "lr": 0.0003, "grad_norm": 0.13020203483025394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13975402149, "tokens_per_second": 147120.69842619725, "data_wait_s": 8.13799852039665e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0871a57311211a69051af25f8a405ba7468b9597144af7d6631f0d6f911a9dc8
-size 222003
+oid sha256:c96ae2ac269f0bddca95f7a87d885fe3b5ff105a6d431a08d86c8775ba4862db
+size 224121