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": 76003.62564487285,
-      "required_pages_per_second_upper_estimate": 0.005258107174820462
+      "required_token_positions_per_second": 76970.14467561392,
+      "required_pages_per_second_upper_estimate": 0.005324973204000254
     },
     {
       "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": 9500.453205609107,
-      "required_pages_per_second_upper_estimate": 0.003977641365165691
+      "required_token_positions_per_second": 9621.26808445174,
+      "required_pages_per_second_upper_estimate": 0.004028224032035542
     },
     {
       "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": 19000.906411218213,
-      "required_pages_per_second_upper_estimate": 0.003957387717858009
+      "required_token_positions_per_second": 19242.53616890348,
+      "required_pages_per_second_upper_estimate": 0.004007712824178627
     },
     {
       "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": 28501.35961682732,
-      "required_pages_per_second_upper_estimate": 0.001708612989972498
+      "required_token_positions_per_second": 28863.80425335522,
+      "required_pages_per_second_upper_estimate": 0.00173034099251143
     },
     {
       "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": 28501.35961682732,
-      "required_pages_per_second_upper_estimate": 0.0022158553459762703
+      "required_token_positions_per_second": 28863.80425335522,
+      "required_pages_per_second_upper_estimate": 0.0022440338222408413
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28501.35961682732,
-      "required_pages_per_second_upper_estimate": 0.3655848387889755
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28863.80425335522,
+      "required_pages_per_second_upper_estimate": 0.3702338894236249
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.6221011469999453,
-    "producer_p95_seconds": 0.2026623580004525,
-    "gpu_update_seconds": 0.9700589856659507,
-    "producer_batches_per_second": 1.607455644186568,
-    "gpu_batches_per_second": 1.0308651481780715,
-    "producer_headroom": 1.5593267917026297,
+    "producer_batch_seconds": 0.05244641853124676,
+    "producer_p95_seconds": 0.17969072899995808,
+    "gpu_update_seconds": 0.9578778929248769,
+    "producer_batches_per_second": 19.067078896993042,
+    "gpu_batches_per_second": 1.0439744015247114,
+    "producer_headroom": 18.26393335808408,
     "sustainable": true,
-    "buffer_seconds": 60.143657111288945,
+    "buffer_seconds": 60.346307254267245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337903.45348609606,
-    "upload_bytes_per_second": 3980970.7197532686,
+    "generation_bytes_per_second": 337655.43159849517,
+    "upload_bytes_per_second": 4070459.5056393086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.50621512829935,
+    "estimated_full_shards_per_hour": 72.45299540487424,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 862278,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 3.149820212999657,
+      "bytes": 17435942,
+      "pages": 43,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"step": 932900, "loss": 0.5345956087112427, "lr": 3e-05, "grad_norm": 0.3351441269328984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136550277302, "tokens_per_second": 153532.8995836737, "data_wait_s": 8.703999810677487e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 933000, "loss": 0.5141292810440063, "lr": 3e-05, "grad_norm": 0.346287447739729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136564981385, "tokens_per_second": 152895.9097757887, "data_wait_s": 1.5439999515365344e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 933100, "loss": 0.5349289178848267, "lr": 3e-05, "grad_norm": 0.33806156412260024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136579669358, "tokens_per_second": 153346.9033480814, "data_wait_s": 7.098000423866324e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 933200, "loss": 0.5128114223480225, "lr": 3e-05, "grad_norm": 0.34213086251517383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136594359750, "tokens_per_second": 153478.80708891235, "data_wait_s": 1.0372999895480461e-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:af448b0b836f76feb657a808f809323a5c3d464674a02a55c9fa16ffac1e480b
-size 172506
+oid sha256:c30afd4c4ac1153dcc96ac13863c3b0846086095b27afd684aa2f8901e29da35
+size 174624