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": 77198.54924100851,
-      "required_pages_per_second_upper_estimate": 0.00534077476180586
+      "required_token_positions_per_second": 77162.28224273756,
+      "required_pages_per_second_upper_estimate": 0.0053382657267144465
     },
     {
       "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": 9649.818655126064,
-      "required_pages_per_second_upper_estimate": 0.0040401775597731695
+      "required_token_positions_per_second": 9645.285280342196,
+      "required_pages_per_second_upper_estimate": 0.0040382795304187945
     },
     {
       "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": 19299.63731025213,
-      "required_pages_per_second_upper_estimate": 0.004019605486063189
+      "required_token_positions_per_second": 19290.57056068439,
+      "required_pages_per_second_upper_estimate": 0.004017717121234494
     },
     {
       "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": 28949.45596537819,
-      "required_pages_per_second_upper_estimate": 0.0017354756818646151
+      "required_token_positions_per_second": 28935.855841026587,
+      "required_pages_per_second_upper_estimate": 0.001734660375175924
     },
     {
       "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": 28949.45596537819,
-      "required_pages_per_second_upper_estimate": 0.0022506928661085615
+      "required_token_positions_per_second": 28935.855841026587,
+      "required_pages_per_second_upper_estimate": 0.002249635516260849
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28949.45596537819,
-      "required_pages_per_second_upper_estimate": 0.37133253761981233
+      "required_token_positions_per_second": 28935.855841026587,
+      "required_pages_per_second_upper_estimate": 0.3711580898273058
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10302674103118648,
-    "producer_p95_seconds": 0.27400233499793103,
-    "gpu_update_seconds": 0.955043854125112,
-    "producer_batches_per_second": 9.70621791964959,
-    "gpu_batches_per_second": 1.0470723367107273,
-    "producer_headroom": 9.269863770960372,
+    "producer_batch_seconds": 0.10228135367964342,
+    "producer_p95_seconds": 0.30837233099737205,
+    "gpu_update_seconds": 0.9554927337175697,
+    "producer_batches_per_second": 9.776953120234518,
+    "gpu_batches_per_second": 1.0465804340649083,
+    "producer_headroom": 9.341807664281403,
     "sustainable": true,
-    "buffer_seconds": 60.16776280988205,
+    "buffer_seconds": 60.19604222420689,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491712.5320499528,
+    "generation_bytes_per_second": 491469.6143755537,
     "upload_bytes_per_second": 5398907.772270573,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.51006289600312,
+    "estimated_full_shards_per_hour": 105.45793841791112,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 671800, "loss": 0.517085611820221, "lr": 0.0001, "grad_norm": 0.18825984839203022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98334522407, "tokens_per_second": 152695.5954018993, "data_wait_s": 7.808001100784168e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 671900, "loss": 0.49347272515296936, "lr": 0.0001, "grad_norm": 0.24404606406302695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98349092549, "tokens_per_second": 151066.72561517902, "data_wait_s": 1.3559998478740454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 672000, "loss": 0.5388064980506897, "lr": 0.0001, "grad_norm": 0.18466606216347412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98363623643, "tokens_per_second": 150813.16569094366, "data_wait_s": 1.1814998288173229e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 672100, "loss": 0.5250399112701416, "lr": 0.0001, "grad_norm": 0.21554353769799287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98378141448, "tokens_per_second": 151300.95974596552, "data_wait_s": 8.882001566234976e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c209c17895d241426a5e1d059d1f57151b5dc7d6e607b26f4c5fa731cc41767a
-size 575940
+oid sha256:4ca686751c97687fde46cd2e3650fd6fb3e80f6e22f5f5fe346412d38104f3d4
+size 578058