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": 76350.98367734747,
-      "required_pages_per_second_upper_estimate": 0.005282138209488733
+      "required_token_positions_per_second": 76468.50326882956,
+      "required_pages_per_second_upper_estimate": 0.005290268487510485
     },
     {
       "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": 9543.872959668433,
-      "required_pages_per_second_upper_estimate": 0.003995820309482761
+      "required_token_positions_per_second": 9558.562908603695,
+      "required_pages_per_second_upper_estimate": 0.004001970684340939
     },
     {
       "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": 19087.745919336867,
-      "required_pages_per_second_upper_estimate": 0.003975474097287292
+      "required_token_positions_per_second": 19117.12581720739,
+      "required_pages_per_second_upper_estimate": 0.003981593155213712
     },
     {
       "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": 28631.618879005302,
-      "required_pages_per_second_upper_estimate": 0.0017164218338456912
+      "required_token_positions_per_second": 28675.688725811087,
+      "required_pages_per_second_upper_estimate": 0.0017190637538709382
     },
     {
       "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": 28631.618879005302,
-      "required_pages_per_second_upper_estimate": 0.002225982430660665
+      "required_token_positions_per_second": 28675.688725811087,
+      "required_pages_per_second_upper_estimate": 0.002229408667407045
     },
     {
       "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": 28631.618879005302,
-      "required_pages_per_second_upper_estimate": 0.3672556647426957
+      "required_token_positions_per_second": 28675.688725811087,
+      "required_pages_per_second_upper_estimate": 0.3678209454190055
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11978487112514813,
-    "producer_p95_seconds": 0.32558981500187656,
-    "gpu_update_seconds": 0.9656457120653223,
-    "producer_batches_per_second": 8.348299669289839,
-    "gpu_batches_per_second": 1.0355764930195783,
-    "producer_headroom": 8.06149977868608,
+    "producer_batch_seconds": 0.233934948093804,
+    "producer_p95_seconds": 0.34167955099837855,
+    "gpu_update_seconds": 0.964161672431391,
+    "producer_batches_per_second": 4.274692636343573,
+    "gpu_batches_per_second": 1.0371704544925886,
+    "producer_headroom": 4.121494801387171,
     "sustainable": true,
-    "buffer_seconds": 60.835679860115306,
+    "buffer_seconds": 60.74218536317763,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282218.74633497396,
-    "upload_bytes_per_second": 3609971.504761287,
+    "generation_bytes_per_second": 288256.0733182201,
+    "upload_bytes_per_second": 3600010.0695605013,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.557573247307914,
+    "estimated_full_shards_per_hour": 61.85304307613327,
     "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": 1.406910452999,
-      "bytes": 5890041,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.0961665050017473,
       "bytes": 2428850,
@@ -562,6 +556,12 @@
       "bytes": 10753564,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.9835506360032014,
+      "bytes": 6796671,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 881800, "loss": 0.5664958953857422, "lr": 3e-05, "grad_norm": 0.32144307865781946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129070716406, "tokens_per_second": 149217.93866730624, "data_wait_s": 1.19440010166727e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881900, "loss": 0.4846755862236023, "lr": 3e-05, "grad_norm": 0.3075932364423472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129085222977, "tokens_per_second": 150371.0277726748, "data_wait_s": 1.9138999050483108e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882000, "loss": 0.5371716022491455, "lr": 3e-05, "grad_norm": 0.31736149239073724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129099909604, "tokens_per_second": 151683.04700585108, "data_wait_s": 1.6454003343824297e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 882100, "loss": 0.5230187773704529, "lr": 3e-05, "grad_norm": 0.31251775305386836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129114604171, "tokens_per_second": 152962.4350397474, "data_wait_s": 1.761299790814519e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4853e415d8e279e71f77d1bc01be72608f117cc95992991cb853ffb03af32f0c
-size 843039
+oid sha256:c095b688d30ae82fb6d0134a323a8e555587ab45b291f5958901442d49d3115d
+size 845157