Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76052.66524384508,
-      "required_pages_per_second_upper_estimate": 0.005261499848064907
+      "required_token_positions_per_second": 76104.59618803913,
+      "required_pages_per_second_upper_estimate": 0.0052650925512806425
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9506.583155480635,
-      "required_pages_per_second_upper_estimate": 0.003980207847168991
+      "required_token_positions_per_second": 9513.07452350489,
+      "required_pages_per_second_upper_estimate": 0.0039829256474055675
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19013.16631096127,
-      "required_pages_per_second_upper_estimate": 0.0039599411316591894
+      "required_token_positions_per_second": 19026.14904700978,
+      "required_pages_per_second_upper_estimate": 0.003962645093200321
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28519.749466441906,
-      "required_pages_per_second_upper_estimate": 0.001709715433377218
+      "required_token_positions_per_second": 28539.22357051467,
+      "required_pages_per_second_upper_estimate": 0.001710882876181145
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28519.749466441906,
-      "required_pages_per_second_upper_estimate": 0.0022172850758954044
+      "required_token_positions_per_second": 28539.22357051467,
+      "required_pages_per_second_upper_estimate": 0.0022187991018295305
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28519.749466441906,
-      "required_pages_per_second_upper_estimate": 0.3658207240343493
+      "required_token_positions_per_second": 28539.22357051467,
+      "required_pages_per_second_upper_estimate": 0.36607051693173087
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12213434905484633,
-    "producer_p95_seconds": 0.2939315510011511,
-    "gpu_update_seconds": 0.969433480912318,
-    "producer_batches_per_second": 8.18770483274066,
-    "gpu_batches_per_second": 1.0315302903082286,
-    "producer_headroom": 7.937435196686387,
+    "producer_batch_seconds": 0.07221397799227702,
+    "producer_p95_seconds": 0.2651190430005954,
+    "gpu_update_seconds": 0.9687719755825649,
+    "producer_batches_per_second": 13.847734577188723,
+    "gpu_batches_per_second": 1.032234648817805,
+    "producer_headroom": 13.415297183686112,
     "sustainable": true,
-    "buffer_seconds": 60.10487581656371,
+    "buffer_seconds": 60.06386248611902,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456606.1302953728,
+    "generation_bytes_per_second": 456189.33657039143,
     "upload_bytes_per_second": 3192815.50960094,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.97704631467712,
+    "estimated_full_shards_per_hour": 97.88761208375747,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 826800, "loss": 0.5056031942367554, "lr": 3e-05, "grad_norm": 0.2900156346911759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121020137440, "tokens_per_second": 151174.91341448275, "data_wait_s": 8.86499765329063e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 826900, "loss": 0.5027215480804443, "lr": 3e-05, "grad_norm": 0.27999472463592445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121034757006, "tokens_per_second": 151124.35402478158, "data_wait_s": 9.398998372489586e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 827000, "loss": 0.5485939979553223, "lr": 3e-05, "grad_norm": 0.29933487786131624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121049563757, "tokens_per_second": 153425.40087427286, "data_wait_s": 9.273000614484772e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 827100, "loss": 0.5386779308319092, "lr": 3e-05, "grad_norm": 0.2909685545523433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121064265463, "tokens_per_second": 152396.25845027762, "data_wait_s": 9.840998245636001e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1e6ed56211a5314e3c4c0cc8518696000b61980e39fe8c75f641769d4257449
-size 575940
+oid sha256:444b0a63dc4b542265e83e7f66902fe158628a27350a7ca93b1d359365685b25
+size 578058