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": 76943.9030721375,
-      "required_pages_per_second_upper_estimate": 0.005323157748982847
+      "required_token_positions_per_second": 77140.22324279293,
+      "required_pages_per_second_upper_estimate": 0.005336739633914345
     },
     {
       "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": 9617.987884017188,
-      "required_pages_per_second_upper_estimate": 0.00402685068061197
+      "required_token_positions_per_second": 9642.527905349116,
+      "required_pages_per_second_upper_estimate": 0.0040371250750897295
     },
     {
       "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": 19235.975768034376,
-      "required_pages_per_second_upper_estimate": 0.004006346465686998
+      "required_token_positions_per_second": 19285.05581069823,
+      "required_pages_per_second_upper_estimate": 0.004016568544246114
     },
     {
       "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": 28853.963652051563,
-      "required_pages_per_second_upper_estimate": 0.0017297510634890036
+      "required_token_positions_per_second": 28927.58371604735,
+      "required_pages_per_second_upper_estimate": 0.0017341644739142222
     },
     {
       "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": 28853.963652051563,
-      "required_pages_per_second_upper_estimate": 0.0022432687587737128
+      "required_token_positions_per_second": 28927.58371604735,
+      "required_pages_per_second_upper_estimate": 0.002248992394928946
     },
     {
       "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": 28853.963652051563,
-      "required_pages_per_second_upper_estimate": 0.37010766475611606
+      "required_token_positions_per_second": 28927.58371604735,
+      "required_pages_per_second_upper_estimate": 0.37105198388998795
     }
   ],
   "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": 0.28660525592187724,
-    "producer_p95_seconds": 0.21881538800062117,
-    "gpu_update_seconds": 0.958204575752773,
-    "producer_batches_per_second": 3.489119544522867,
-    "gpu_batches_per_second": 1.0436184769983927,
-    "producer_headroom": 3.343290312910242,
+    "producer_batch_seconds": 0.10761807482806773,
+    "producer_p95_seconds": 0.1866873819999455,
+    "gpu_update_seconds": 0.9557659661931076,
+    "producer_batches_per_second": 9.292119391632076,
+    "gpu_batches_per_second": 1.0462812397297219,
+    "producer_headroom": 8.881091468324943,
     "sustainable": true,
-    "buffer_seconds": 60.366888272424696,
+    "buffer_seconds": 60.21325587016578,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392034.75410241185,
+    "generation_bytes_per_second": 390089.33467260783,
     "upload_bytes_per_second": 5523840.133055187,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.12153212837474,
+    "estimated_full_shards_per_hour": 83.70409040578534,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 611600, "loss": 0.5925382971763611, "lr": 0.0001, "grad_norm": 0.16367374215516606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89520455201, "tokens_per_second": 154400.72257098564, "data_wait_s": 1.4102000022830907e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 611700, "loss": 0.5162065029144287, "lr": 0.0001, "grad_norm": 0.17406104491443863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89535262022, "tokens_per_second": 153441.24196737166, "data_wait_s": 1.3316000149643514e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 611800, "loss": 0.666245698928833, "lr": 0.0001, "grad_norm": 0.1913862472896558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89549970699, "tokens_per_second": 152055.72961813345, "data_wait_s": 1.149099989561364e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 611900, "loss": 0.5776656866073608, "lr": 0.0001, "grad_norm": 0.14340920056441128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89564452425, "tokens_per_second": 150589.9909592441, "data_wait_s": 9.762000445334706e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7893f50c19d4eec19c884a39bc1bcfe4c15c5fa21147ede666b3f7be13e21513
-size 158463
+oid sha256:10d5faa6a949e1ca192a07d520d215b9e5a11491105e4d452cfb1865b893be2a
+size 160581