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": 76784.80996920387,
-      "required_pages_per_second_upper_estimate": 0.005312151318975041
+      "required_token_positions_per_second": 76917.5736157132,
+      "required_pages_per_second_upper_estimate": 0.005321336216094658
     },
     {
       "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": 9598.101246150483,
-      "required_pages_per_second_upper_estimate": 0.004018524560617402
+      "required_token_positions_per_second": 9614.69670196415,
+      "required_pages_per_second_upper_estimate": 0.00402547273141406
     },
     {
       "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": 19196.202492300967,
-      "required_pages_per_second_upper_estimate": 0.003998062741243545
+      "required_token_positions_per_second": 19229.3934039283,
+      "required_pages_per_second_upper_estimate": 0.004004975532832317
     },
     {
       "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": 28794.30373845145,
-      "required_pages_per_second_upper_estimate": 0.0017261745427640904
+      "required_token_positions_per_second": 28844.09010589245,
+      "required_pages_per_second_upper_estimate": 0.001729159159472797
     },
     {
       "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": 28794.30373845145,
-      "required_pages_per_second_upper_estimate": 0.0022386304629075306
+      "required_token_positions_per_second": 28844.09010589245,
+      "required_pages_per_second_upper_estimate": 0.0022425011339889814
     },
     {
       "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": 28794.30373845145,
-      "required_pages_per_second_upper_estimate": 0.3693424114422782
+      "required_token_positions_per_second": 28844.09010589245,
+      "required_pages_per_second_upper_estimate": 0.36998101750737483
     }
   ],
   "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": 0.19124411546876274,
-    "producer_p95_seconds": 0.16978395300020566,
-    "gpu_update_seconds": 0.9601899129472371,
-    "producer_batches_per_second": 5.228919057451141,
-    "gpu_batches_per_second": 1.0414606386882035,
-    "producer_headroom": 5.02075533458216,
+    "producer_batch_seconds": 0.06746018778904528,
+    "producer_p95_seconds": 0.16480328700026803,
+    "gpu_update_seconds": 0.9585325762920112,
+    "producer_batches_per_second": 14.823557905398951,
+    "gpu_batches_per_second": 1.0432613608902073,
+    "producer_headroom": 14.208863148875867,
     "sustainable": true,
-    "buffer_seconds": 60.49196451567594,
+    "buffer_seconds": 60.38755230639671,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449886.7487318865,
-    "upload_bytes_per_second": 5132548.294819486,
+    "generation_bytes_per_second": 526991.889611032,
+    "upload_bytes_per_second": 5292974.006612815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.53522345035024,
+    "estimated_full_shards_per_hour": 113.08019176719876,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 2468071,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.1340931629997613,
+      "bytes": 11786170,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 607800, "loss": 0.5417321920394897, "lr": 0.0001, "grad_norm": 0.14292809824885952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88963958706, "tokens_per_second": 150423.73204376292, "data_wait_s": 9.657000191509724e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 607900, "loss": 0.5578194856643677, "lr": 0.0001, "grad_norm": 0.13951009195828343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88978511604, "tokens_per_second": 151501.35394295419, "data_wait_s": 1.9536000309017254e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608000, "loss": 0.5621930956840515, "lr": 0.0001, "grad_norm": 0.1410032505224448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88993177505, "tokens_per_second": 151813.3803332702, "data_wait_s": 1.1229999927309109e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 608100, "loss": 0.5482956767082214, "lr": 0.0001, "grad_norm": 0.13596161118764477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89007715322, "tokens_per_second": 150830.3868322628, "data_wait_s": 1.3473000308295013e-05, "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 sha256:5407a5fcb589d6f35be334f5f13f278acd0bf4bf4203767040da7c89fbee36e2
-size 77196
+oid sha256:ddec16a5690141a5ce895aedb47cb1facfeece128c1e9e546e9de44c99a8d5db
+size 79314