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": 76997.41244358459,
-      "required_pages_per_second_upper_estimate": 0.005326859651458404
+      "required_token_positions_per_second": 77196.25219599612,
+      "required_pages_per_second_upper_estimate": 0.005340615846902023
     },
     {
       "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": 9624.676555448073,
-      "required_pages_per_second_upper_estimate": 0.004029651087664739
+      "required_token_positions_per_second": 9649.531524499514,
+      "required_pages_per_second_upper_estimate": 0.004040057344175803
     },
     {
       "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": 19249.353110896147,
-      "required_pages_per_second_upper_estimate": 0.00400913261342085
+      "required_token_positions_per_second": 19299.06304899903,
+      "required_pages_per_second_upper_estimate": 0.004019485882588456
     },
     {
       "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": 28874.02966634422,
-      "required_pages_per_second_upper_estimate": 0.0017309539904068188
+      "required_token_positions_per_second": 28948.59457349854,
+      "required_pages_per_second_upper_estimate": 0.0017354240427366993
     },
     {
       "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": 28874.02966634422,
-      "required_pages_per_second_upper_estimate": 0.0022448288031238976
+      "required_token_positions_per_second": 28948.59457349854,
+      "required_pages_per_second_upper_estimate": 0.002250625896678781
     },
     {
       "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": 28874.02966634422,
-      "required_pages_per_second_upper_estimate": 0.37036505004225473
+      "required_token_positions_per_second": 28948.59457349854,
+      "required_pages_per_second_upper_estimate": 0.37132148860967074
     }
   ],
   "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": 0.34258388731251443,
-    "producer_p95_seconds": 0.23198293300083606,
-    "gpu_update_seconds": 0.9575386712380749,
-    "producer_batches_per_second": 2.91899309055295,
-    "gpu_batches_per_second": 1.044344244297751,
-    "producer_headroom": 2.795048765281193,
+    "producer_batch_seconds": 0.1234200835078525,
+    "producer_p95_seconds": 0.2275281330003054,
+    "gpu_update_seconds": 0.9550722723275419,
+    "producer_batches_per_second": 8.10240903731341,
+    "gpu_batches_per_second": 1.047041181043784,
+    "producer_headroom": 7.738386210594131,
     "sustainable": true,
-    "buffer_seconds": 60.32493628799872,
+    "buffer_seconds": 60.16955315663514,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381057.9463217083,
-    "upload_bytes_per_second": 4709599.254748465,
+    "generation_bytes_per_second": 395171.39919200586,
+    "upload_bytes_per_second": 4647958.401071017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.76616470564305,
+    "estimated_full_shards_per_hour": 84.79458314724094,
     "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": 1.0178890370007139,
-      "bytes": 11702648,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0858605960002023,
       "bytes": 2526198,
@@ -562,6 +556,12 @@
       "bytes": 6221614,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0521548180004174,
+      "bytes": 6267595,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -171,3 +171,4 @@
 {"step": 457300, "loss": 0.5756521224975586, "lr": 0.0003, "grad_norm": 0.17757718417413582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66934877632, "tokens_per_second": 152934.34616986127, "data_wait_s": 1.2229000276420265e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457400, "loss": 0.6224752068519592, "lr": 0.0003, "grad_norm": 0.1910308007215812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66949595786, "tokens_per_second": 152821.2799298036, "data_wait_s": 1.3413999113254249e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457500, "loss": 0.5915499925613403, "lr": 0.0003, "grad_norm": 0.17881144806741683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66964305240, "tokens_per_second": 152904.36875887544, "data_wait_s": 9.677998605184257e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 457600, "loss": 0.5711995363235474, "lr": 0.0003, "grad_norm": 0.13963287596159765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66978781488, "tokens_per_second": 150979.49881917232, "data_wait_s": 1.1741998605430126e-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 sha256:16028bbadc7bea836ba179a55464820fd65291d02fdc7cf80b0886e9e1f4f7f8
-size 368928
+oid sha256:cb917665f88a8ca9b933db45e6e151185196e983da221c7bd31f2ff34311a6fd
+size 371046