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": 75131.57427665639,
-      "required_pages_per_second_upper_estimate": 0.0065122144390664695
+      "required_token_positions_per_second": 74102.7755245799,
+      "required_pages_per_second_upper_estimate": 0.006423040770703077
     },
     {
       "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": 9391.446784582049,
-      "required_pages_per_second_upper_estimate": 0.003746388537012147
+      "required_token_positions_per_second": 9262.846940572488,
+      "required_pages_per_second_upper_estimate": 0.0036950881364977216
     },
     {
       "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": 18782.893569164098,
-      "required_pages_per_second_upper_estimate": 0.004345091859589583
+      "required_token_positions_per_second": 18525.693881144976,
+      "required_pages_per_second_upper_estimate": 0.0042855932383262674
     },
     {
       "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": 28174.34035374615,
-      "required_pages_per_second_upper_estimate": 0.001715669508203385
+      "required_token_positions_per_second": 27788.540821717466,
+      "required_pages_per_second_upper_estimate": 0.0016921763408365494
     },
     {
       "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": 28174.34035374615,
-      "required_pages_per_second_upper_estimate": 0.00216893871199716
+      "required_token_positions_per_second": 27788.540821717466,
+      "required_pages_per_second_upper_estimate": 0.002139238796060135
     },
     {
       "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": 28174.34035374615,
-      "required_pages_per_second_upper_estimate": 0.34983101374208314
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27788.540821717466,
+      "required_pages_per_second_upper_estimate": 0.3450406747422609
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3799541037342351,
-    "producer_p95_seconds": 0.3748887929978082,
-    "gpu_update_seconds": 0.9813184497973112,
-    "producer_batches_per_second": 2.63189682693746,
-    "gpu_batches_per_second": 1.0190371945076009,
-    "producer_headroom": 2.5827289142367307,
+    "producer_batch_seconds": 0.11452659066407023,
+    "producer_p95_seconds": 0.3818352460002643,
+    "gpu_update_seconds": 0.9949424900494369,
+    "producer_batches_per_second": 8.731596690354673,
+    "gpu_batches_per_second": 1.0050832183780911,
+    "producer_headroom": 8.6874365532089,
     "sustainable": true,
-    "buffer_seconds": 60.841743887433296,
+    "buffer_seconds": 60.69149189301565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403600.0020644393,
-    "upload_bytes_per_second": 6001683.48335616,
+    "generation_bytes_per_second": 404880.64709084755,
+    "upload_bytes_per_second": 6008795.479559451,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.60316511583218,
+    "estimated_full_shards_per_hour": 86.87796172660894,
     "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": 1.0306341840005189,
-      "bytes": 6113692,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1684751539978606,
       "bytes": 6202267,
@@ -562,6 +556,12 @@
       "bytes": 12314988,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.0287573200039333,
+      "bytes": 6604059,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -305,3 +305,4 @@
 {"step": 115500, "loss": 0.5834883004426956, "lr": 0.0003, "grad_norm": 0.1572690547827181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16901693141, "tokens_per_second": 149983.04633133605, "data_wait_s": 1.4511999324895442e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115600, "loss": 0.7221038471907377, "lr": 0.0003, "grad_norm": 0.17080569046466865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16916387672, "tokens_per_second": 147272.63591024862, "data_wait_s": 7.412003469653428e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 115700, "loss": 0.6948383757844567, "lr": 0.0003, "grad_norm": 0.1684999602518139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16931012208, "tokens_per_second": 148826.94533303508, "data_wait_s": 1.2807002349290997e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 115800, "loss": 0.603748444467783, "lr": 0.0003, "grad_norm": 0.1584251483595853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16945791532, "tokens_per_second": 150606.8792175921, "data_wait_s": 1.4125005691312253e-05, "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:d46c77f2f1b623a3a7c7aa9e0255e47af1b75679b7d270280fcc87005f8c57fa
-size 655089
+oid sha256:b5fca066a42ba1f50057c2e7685b4bdcec2cde9e91f596560d6b6b9c20591032
+size 657207