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": 77017.31014897076,
-      "required_pages_per_second_upper_estimate": 0.005328236220886057
+      "required_token_positions_per_second": 76876.0115875668,
+      "required_pages_per_second_upper_estimate": 0.005318460858550298
     },
     {
       "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": 9627.163768621345,
-      "required_pages_per_second_upper_estimate": 0.0040306924318814705
+      "required_token_positions_per_second": 9609.50144844585,
+      "required_pages_per_second_upper_estimate": 0.004023297587255177
     },
     {
       "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": 19254.32753724269,
-      "required_pages_per_second_upper_estimate": 0.004010168655244391
+      "required_token_positions_per_second": 19219.0028968917,
+      "required_pages_per_second_upper_estimate": 0.004002811464232691
     },
     {
       "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": 28881.491305864034,
-      "required_pages_per_second_upper_estimate": 0.0017314013043027677
+      "required_token_positions_per_second": 28828.50434533755,
+      "required_pages_per_second_upper_estimate": 0.0017282248169256092
     },
     {
       "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": 28881.491305864034,
-      "required_pages_per_second_upper_estimate": 0.0022454089127762794
+      "required_token_positions_per_second": 28828.50434533755,
+      "required_pages_per_second_upper_estimate": 0.002241289409660354
     },
     {
       "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": 28881.491305864034,
-      "required_pages_per_second_upper_estimate": 0.3704607599423306
+      "required_token_positions_per_second": 28828.50434533755,
+      "required_pages_per_second_upper_estimate": 0.36978110010566245
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.5249778470000166,
-    "producer_p95_seconds": 0.2083826729995053,
-    "gpu_update_seconds": 0.957291287600042,
-    "producer_batches_per_second": 1.9048422818495965,
-    "gpu_batches_per_second": 1.0446141241993647,
-    "producer_headroom": 1.8234889206668023,
+    "producer_batch_seconds": 0.056321398796825406,
+    "producer_p95_seconds": 0.1929896709989407,
+    "gpu_update_seconds": 0.9590507946164584,
+    "producer_batches_per_second": 17.75524083852061,
+    "gpu_batches_per_second": 1.0426976398053223,
+    "producer_headroom": 17.028177834789783,
     "sustainable": true,
-    "buffer_seconds": 60.309351118802645,
+    "buffer_seconds": 60.420200060836876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 229959.44610931008,
-    "upload_bytes_per_second": 4684113.223649963,
+    "generation_bytes_per_second": 229934.4085740447,
+    "upload_bytes_per_second": 4787164.619330915,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.34394395312764,
+    "estimated_full_shards_per_hour": 49.33857148090368,
     "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": 0.9099277899999834,
-      "bytes": 1221459,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 6.556346244999986,
       "bytes": 10834593,
@@ -562,6 +556,12 @@
       "bytes": 6763666,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.232765078999364,
+      "bytes": 12254761,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -101,3 +101,4 @@
 {"step": 450300, "loss": 0.5975652933120728, "lr": 0.0003, "grad_norm": 0.16137010197765644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65909904151, "tokens_per_second": 151171.20417849618, "data_wait_s": 7.103000825736672e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 450400, "loss": 0.6156168580055237, "lr": 0.0003, "grad_norm": 0.19648303706350098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65924449840, "tokens_per_second": 121494.6019841358, "data_wait_s": 1.4375998944160528e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450500, "loss": 0.6055036187171936, "lr": 0.0003, "grad_norm": 0.1672120317372175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65938946967, "tokens_per_second": 150269.87771515214, "data_wait_s": 1.5751000319141895e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 450600, "loss": 0.5948117971420288, "lr": 0.0003, "grad_norm": 0.14550870833572543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65953663832, "tokens_per_second": 153184.70147082658, "data_wait_s": 1.693900048849173e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:639f3cb62971a60b9a126e8d7e4e8481a449a71dfed8e54c78ba5cf2d0820f52
-size 219885
+oid sha256:fb551d2885f93a074e2dd043b18f2c088615e674a7d33378e506af47a13558b3
+size 222003