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": 76289.30146502068,
-      "required_pages_per_second_upper_estimate": 0.005277870890917515
+      "required_token_positions_per_second": 77297.28493469118,
+      "required_pages_per_second_upper_estimate": 0.0053476055261932
     },
     {
       "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": 9536.162683127584,
-      "required_pages_per_second_upper_estimate": 0.003992592177703991
+      "required_token_positions_per_second": 9662.160616836398,
+      "required_pages_per_second_upper_estimate": 0.004045344881411819
     },
     {
       "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": 19072.32536625517,
-      "required_pages_per_second_upper_estimate": 0.003972262402747706
+      "required_token_positions_per_second": 19324.321233672796,
+      "required_pages_per_second_upper_estimate": 0.004024746496352878
     },
     {
       "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": 28608.488049382755,
-      "required_pages_per_second_upper_estimate": 0.0017150351759285512
+      "required_token_positions_per_second": 28986.48185050919,
+      "required_pages_per_second_upper_estimate": 0.0017376953271429655
     },
     {
       "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": 28608.488049382755,
-      "required_pages_per_second_upper_estimate": 0.0022241841103992793
+      "required_token_positions_per_second": 28986.48185050919,
+      "required_pages_per_second_upper_estimate": 0.0022535714658178394
     },
     {
       "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": 28608.488049382755,
-      "required_pages_per_second_upper_estimate": 0.36695896729624755
+      "required_token_positions_per_second": 28986.48185050919,
+      "required_pages_per_second_upper_estimate": 0.37180746591897473
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08272313867186654,
-    "producer_p95_seconds": 0.20247803600068437,
-    "gpu_update_seconds": 0.9664264658892563,
-    "producer_batches_per_second": 12.088516176430959,
-    "gpu_batches_per_second": 1.0347398744713092,
-    "producer_headroom": 11.682661966233276,
+    "producer_batch_seconds": 0.09361182234360399,
+    "producer_p95_seconds": 0.18935493899880385,
+    "gpu_update_seconds": 0.9538239287743822,
+    "producer_batches_per_second": 10.682411419462392,
+    "gpu_batches_per_second": 1.048411525264366,
+    "producer_headroom": 10.189139628895944,
     "sustainable": true,
-    "buffer_seconds": 60.884867351023146,
+    "buffer_seconds": 60.090907512786075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437897.4821840903,
-    "upload_bytes_per_second": 3423715.5865978505,
+    "generation_bytes_per_second": 435887.6081856299,
+    "upload_bytes_per_second": 3423071.9731167015,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.96260594503433,
+    "estimated_full_shards_per_hour": 93.53133377243684,
     "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": 2.406933221000145,
-      "bytes": 7260996,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.1487480899995717,
       "bytes": 1204399,
@@ -562,6 +556,12 @@
       "bytes": 6153792,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.060665737999443,
+      "bytes": 2571654,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 938000, "loss": 0.5621144771575928, "lr": 3e-05, "grad_norm": 0.33235588155988943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137296379882, "tokens_per_second": 151453.7891966275, "data_wait_s": 1.6631000107736327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938100, "loss": 0.5423369407653809, "lr": 3e-05, "grad_norm": 0.3549244735281664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137310890076, "tokens_per_second": 151154.7651105335, "data_wait_s": 1.2222000805195421e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938200, "loss": 0.5198876857757568, "lr": 3e-05, "grad_norm": 0.34040018912698633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137325556534, "tokens_per_second": 153583.85701232433, "data_wait_s": 1.1931999324588105e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 938300, "loss": 0.5379912257194519, "lr": 3e-05, "grad_norm": 0.3346158346146954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137340265616, "tokens_per_second": 153998.72302263198, "data_wait_s": 1.6156000128830783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15b74fbeea4259f4e551fb330337e4c828b51c0193948521702eae3c576117d5
-size 281307
+oid sha256:762b410731703732b1bc4f5c573101455674c7c2ecbbeb352c740e0d16a2d01a
+size 283425