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": 76771.88890547687,
-      "required_pages_per_second_upper_estimate": 0.005311257409805408
+      "required_token_positions_per_second": 77331.3246144179,
+      "required_pages_per_second_upper_estimate": 0.005349960470219633
     },
     {
       "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": 9596.486113184608,
-      "required_pages_per_second_upper_estimate": 0.0040178483381724025
+      "required_token_positions_per_second": 9666.415576802237,
+      "required_pages_per_second_upper_estimate": 0.004047126344295848
     },
     {
       "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": 19192.972226369217,
-      "required_pages_per_second_upper_estimate": 0.003997389962037802
+      "required_token_positions_per_second": 19332.831153604475,
+      "required_pages_per_second_upper_estimate": 0.004026518888252967
     },
     {
       "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": 28789.458339553825,
-      "required_pages_per_second_upper_estimate": 0.0017258840684986735
+      "required_token_positions_per_second": 28999.24673040671,
+      "required_pages_per_second_upper_estimate": 0.0017384605622019788
     },
     {
       "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": 28789.458339553825,
-      "required_pages_per_second_upper_estimate": 0.002238253754455898
+      "required_token_positions_per_second": 28999.24673040671,
+      "required_pages_per_second_upper_estimate": 0.0022545638790830986
     },
     {
       "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": 28789.458339553825,
-      "required_pages_per_second_upper_estimate": 0.36928025986780344
+      "required_token_positions_per_second": 28999.24673040671,
+      "required_pages_per_second_upper_estimate": 0.37197120009243995
     }
   ],
   "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": 0.06271172823438498,
-    "producer_p95_seconds": 0.13018736599997283,
-    "gpu_update_seconds": 0.9603515173473904,
-    "producer_batches_per_second": 15.945980571010605,
-    "gpu_batches_per_second": 1.041285385545205,
-    "producer_headroom": 15.31374663696204,
+    "producer_batch_seconds": 0.042553780859377,
+    "producer_p95_seconds": 0.13049337699999342,
+    "gpu_update_seconds": 0.9534040748379204,
+    "producer_batches_per_second": 23.499674524916944,
+    "gpu_batches_per_second": 1.0488732179689928,
+    "producer_headroom": 22.40468544942069,
     "sustainable": true,
-    "buffer_seconds": 60.5021455928856,
+    "buffer_seconds": 60.06445671478899,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 60,
+    "upload_pages_per_commit": 65,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312305.2293024011,
-    "upload_bytes_per_second": 4168168.17086391,
+    "generation_bytes_per_second": 312285.0804309098,
+    "upload_bytes_per_second": 3904223.0874083876,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.01343211463951,
+    "estimated_full_shards_per_hour": 67.0091086358592,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 6063645,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8495643409999047,
+      "bytes": 825792,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -6,3 +6,4 @@
 {"step": 605000, "loss": 0.5390992760658264, "lr": 0.0001, "grad_norm": 0.14674679412242164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88553910942, "tokens_per_second": 129367.78419255553, "data_wait_s": 1.1035000170522835e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605100, "loss": 0.5622690320014954, "lr": 0.0001, "grad_norm": 0.14328663181039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88568593606, "tokens_per_second": 139609.19308701303, "data_wait_s": 1.6875000028448994e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605200, "loss": 0.5281063318252563, "lr": 0.0001, "grad_norm": 0.1335078791685974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88583296602, "tokens_per_second": 152306.21731186987, "data_wait_s": 1.2056000059601502e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 605300, "loss": 0.5635167956352234, "lr": 0.0001, "grad_norm": 0.13613936133076468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88597999350, "tokens_per_second": 150328.2228645332, "data_wait_s": 7.193999863375211e-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 sha256:420e113ee1553564d6ea223e2d3e148c4c04153c268845d3ebe3bacf5d201474
-size 17892
+oid sha256:e82e1645129f94656570f3e92b8b158b48371ae8e40058e24e29deeaed6d5682
+size 20010