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": 76873.76820251234,
-      "required_pages_per_second_upper_estimate": 0.00531830565596686
+      "required_token_positions_per_second": 76733.69599285242,
+      "required_pages_per_second_upper_estimate": 0.005308615135490282
     },
     {
       "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": 9609.221025314042,
-      "required_pages_per_second_upper_estimate": 0.00402318017994579
+      "required_token_positions_per_second": 9591.711999106552,
+      "required_pages_per_second_upper_estimate": 0.0040158495163027585
     },
     {
       "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": 19218.442050628084,
-      "required_pages_per_second_upper_estimate": 0.00400269465474649
+      "required_token_positions_per_second": 19183.423998213104,
+      "required_pages_per_second_upper_estimate": 0.003995401317916594
     },
     {
       "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": 28827.663075942124,
-      "required_pages_per_second_upper_estimate": 0.0017281743841098974
+      "required_token_positions_per_second": 28775.135997319652,
+      "required_pages_per_second_upper_estimate": 0.0017250254659506852
     },
     {
       "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": 28827.663075942124,
-      "required_pages_per_second_upper_estimate": 0.0022412240046652122
+      "required_token_positions_per_second": 28775.135997319652,
+      "required_pages_per_second_upper_estimate": 0.0022371402553445167
     },
     {
       "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": 28827.663075942124,
-      "required_pages_per_second_upper_estimate": 0.36977030920514264
+      "required_token_positions_per_second": 28775.135997319652,
+      "required_pages_per_second_upper_estimate": 0.36909654823975646
     }
   ],
   "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": 0.1139808676638836,
-    "producer_p95_seconds": 0.34277320399996825,
-    "gpu_update_seconds": 0.9590787823197988,
-    "producer_batches_per_second": 8.773402242812226,
-    "gpu_batches_per_second": 1.0426672119481382,
-    "producer_headroom": 8.414383939838142,
+    "producer_batch_seconds": 0.24954020897661167,
+    "producer_p95_seconds": 0.35269529999641236,
+    "gpu_update_seconds": 0.9608295162384413,
+    "producer_batches_per_second": 4.007370211402386,
+    "gpu_batches_per_second": 1.0407673610141657,
+    "producer_headroom": 3.850399581610095,
     "sustainable": true,
-    "buffer_seconds": 60.421963286147324,
+    "buffer_seconds": 60.5322595230218,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357624.2390309097,
-    "upload_bytes_per_second": 4877290.826039317,
+    "generation_bytes_per_second": 372070.0431838927,
+    "upload_bytes_per_second": 4873933.913373116,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.73783662982434,
+    "estimated_full_shards_per_hour": 79.83756991994463,
     "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": 1.1693340630008606,
-      "bytes": 7789873,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.150511206000374,
       "bytes": 5980828,
@@ -562,6 +556,12 @@
       "bytes": 7452041,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0214342609979212,
+      "bytes": 6772893,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -394,3 +394,4 @@
 {"step": 596500, "loss": 0.5947234034538269, "lr": 0.0003, "grad_norm": 0.1938059087422956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87309746644, "tokens_per_second": 152386.78016610106, "data_wait_s": 8.946997695602477e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 596600, "loss": 0.5673349499702454, "lr": 0.0003, "grad_norm": 0.1941837076240469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87324433737, "tokens_per_second": 151868.84143900863, "data_wait_s": 1.0755000403150916e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596700, "loss": 0.5793520212173462, "lr": 0.0003, "grad_norm": 0.1720744731261329, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87339005257, "tokens_per_second": 151321.86706134118, "data_wait_s": 1.8013000953942537e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 596800, "loss": 0.5346783995628357, "lr": 0.0003, "grad_norm": 0.164191460487655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87353465319, "tokens_per_second": 149623.66422684837, "data_wait_s": 1.2577998859342188e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a37f976fa9c4fd7ff058000fa264745781dadde6004dee2c9da4f37162a7b0e
-size 845157
+oid sha256:606437a5be85ef9a391a526346b24962a32c204a79a73e932fd64e667c64eb73
+size 847275