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": 75902.24289695892,
-      "required_pages_per_second_upper_estimate": 0.005251093281079391
+      "required_token_positions_per_second": 76109.27578882316,
+      "required_pages_per_second_upper_estimate": 0.005265416296921051
     },
     {
       "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": 9487.780362119865,
-      "required_pages_per_second_upper_estimate": 0.003972335510235811
+      "required_token_positions_per_second": 9513.659473602895,
+      "required_pages_per_second_upper_estimate": 0.003983170553796454
     },
     {
       "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": 18975.56072423973,
-      "required_pages_per_second_upper_estimate": 0.003952108879671105
+      "required_token_positions_per_second": 19027.31894720579,
+      "required_pages_per_second_upper_estimate": 0.0039628887525588056
     },
     {
       "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": 28463.341086359593,
-      "required_pages_per_second_upper_estimate": 0.0017063338371218926
+      "required_token_positions_per_second": 28540.978420808686,
+      "required_pages_per_second_upper_estimate": 0.00171098807677677
     },
     {
       "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": 28463.341086359593,
-      "required_pages_per_second_upper_estimate": 0.002212899572458247
+      "required_token_positions_per_second": 28540.978420808686,
+      "required_pages_per_second_upper_estimate": 0.0022189355337211194
     },
     {
       "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": 28463.341086359593,
-      "required_pages_per_second_upper_estimate": 0.3650971779012531
+      "required_token_positions_per_second": 28540.978420808686,
+      "required_pages_per_second_upper_estimate": 0.36609302626709106
     }
   ],
   "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": 0.08773826146098429,
-    "producer_p95_seconds": 0.2399070160008705,
-    "gpu_update_seconds": 0.9713546950132875,
-    "producer_batches_per_second": 11.397536073183796,
-    "gpu_batches_per_second": 1.0294900566536311,
-    "producer_headroom": 11.071050176270388,
+    "producer_batch_seconds": 0.1872666168046777,
+    "producer_p95_seconds": 0.2463551250002638,
+    "gpu_update_seconds": 0.9687124103580965,
+    "producer_batches_per_second": 5.3399800619189755,
+    "gpu_batches_per_second": 1.0322981199655918,
+    "producer_headroom": 5.172904957045708,
     "sustainable": true,
-    "buffer_seconds": 60.22399109082383,
+    "buffer_seconds": 60.06016944220198,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481064.46792713675,
-    "upload_bytes_per_second": 4688953.7933579,
+    "generation_bytes_per_second": 464063.7288726379,
+    "upload_bytes_per_second": 4696463.237457393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.22523620949342,
+    "estimated_full_shards_per_hour": 99.57727336534836,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 2621559,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.171350075999726,
+      "bytes": 6139992,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -115,3 +115,4 @@
 {"step": 731500, "loss": 0.5645458102226257, "lr": 5e-05, "grad_norm": 0.2401260459543286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107070779789, "tokens_per_second": 152551.42215007058, "data_wait_s": 1.853399953688495e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731600, "loss": 0.5479867458343506, "lr": 5e-05, "grad_norm": 0.2170199591214406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107085434730, "tokens_per_second": 151005.64977266974, "data_wait_s": 1.911899926199112e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731700, "loss": 0.5008614659309387, "lr": 5e-05, "grad_norm": 0.21254040492607887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107100156922, "tokens_per_second": 151196.52390834645, "data_wait_s": 2.17869983316632e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 731800, "loss": 0.5290740132331848, "lr": 5e-05, "grad_norm": 0.22209223910479947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107114684526, "tokens_per_second": 149706.74671501736, "data_wait_s": 2.137299998139497e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:487c667fe22d7a6321a6da5dd9432f5840a683d0f24fbfa5abf2283574c82898
-size 249537
+oid sha256:1aae7d16de9563f9de802dd5b5e455066aa7031c3e801ef08179bf6a2613002b
+size 251655