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": 77032.31139389469,
-      "required_pages_per_second_upper_estimate": 0.005329274041817578
+      "required_token_positions_per_second": 77106.85078526253,
+      "required_pages_per_second_upper_estimate": 0.005334430849867622
     },
     {
       "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": 9629.038924236836,
-      "required_pages_per_second_upper_estimate": 0.004031477520379975
+      "required_token_positions_per_second": 9638.356348157817,
+      "required_pages_per_second_upper_estimate": 0.004035378531205752
     },
     {
       "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": 19258.077848473673,
-      "required_pages_per_second_upper_estimate": 0.0040109497461714464
+      "required_token_positions_per_second": 19276.712696315633,
+      "required_pages_per_second_upper_estimate": 0.0040148308935429415
     },
     {
       "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": 28887.116772710506,
-      "required_pages_per_second_upper_estimate": 0.0017317385424506233
+      "required_token_positions_per_second": 28915.069044473454,
+      "required_pages_per_second_upper_estimate": 0.0017334142384621643
     },
     {
       "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": 28887.116772710506,
-      "required_pages_per_second_upper_estimate": 0.002245846268079779
+      "required_token_positions_per_second": 28915.069044473454,
+      "required_pages_per_second_upper_estimate": 0.0022480194342602983
     },
     {
       "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": 28887.116772710506,
-      "required_pages_per_second_upper_estimate": 0.3705329173908814
+      "required_token_positions_per_second": 28915.069044473454,
+      "required_pages_per_second_upper_estimate": 0.3708914591202454
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17726682745313838,
-    "producer_p95_seconds": 0.1937381080006162,
-    "gpu_update_seconds": 0.9571048650351601,
-    "producer_batches_per_second": 5.641213386437778,
-    "gpu_batches_per_second": 1.0448175916055593,
-    "producer_headroom": 5.399232776861068,
+    "producer_batch_seconds": 0.057279369343788744,
+    "producer_p95_seconds": 0.17993910600125673,
+    "gpu_update_seconds": 0.9561796292955549,
+    "producer_batches_per_second": 17.45829277550239,
+    "gpu_batches_per_second": 1.0458285968053187,
+    "producer_headroom": 16.693263914213137,
     "sustainable": true,
-    "buffer_seconds": 60.297606497215085,
+    "buffer_seconds": 60.23931664561996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412916.8299053652,
-    "upload_bytes_per_second": 3205704.4568790128,
+    "generation_bytes_per_second": 400151.9081010628,
+    "upload_bytes_per_second": 3255452.07842997,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.60233948584286,
+    "estimated_full_shards_per_hour": 85.86328441881098,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 7238361,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.2277038499996706,
+      "bytes": 10100946,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -113,3 +113,4 @@
 {"step": 936300, "loss": 0.5013501048088074, "lr": 3e-05, "grad_norm": 0.330866853622343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137047836782, "tokens_per_second": 153380.1884403988, "data_wait_s": 1.5082001482369378e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 936400, "loss": 0.5900183320045471, "lr": 3e-05, "grad_norm": 0.3628460645698595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137062309962, "tokens_per_second": 150364.27521855503, "data_wait_s": 1.2307000361033715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 936500, "loss": 0.5569081902503967, "lr": 3e-05, "grad_norm": 0.34107267559015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137076852437, "tokens_per_second": 152175.62380885164, "data_wait_s": 8.95000084710773e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 936600, "loss": 0.5320873260498047, "lr": 3e-05, "grad_norm": 0.3335653994005593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137091502988, "tokens_per_second": 153362.91034865612, "data_wait_s": 1.6969999705906957e-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 sha256:78b18b633c23e226a216c8ebfbf0a65849683756ffbaac34cf52c26ef03c070c
-size 245301
+oid sha256:516f8d99a2f2bb47b48257cd7c5459a128164f55e1b56a213f8a2960ccb924a5
+size 247419