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": 76597.44192182511,
-      "required_pages_per_second_upper_estimate": 0.005299188762703611
+      "required_token_positions_per_second": 76814.30900919597,
+      "required_pages_per_second_upper_estimate": 0.005314192131008898
     },
     {
       "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": 9574.68024022814,
-      "required_pages_per_second_upper_estimate": 0.004008718674523936
+      "required_token_positions_per_second": 9601.788626149497,
+      "required_pages_per_second_upper_estimate": 0.004020068389621738
     },
     {
       "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": 19149.36048045628,
-      "required_pages_per_second_upper_estimate": 0.00398830678548322
+      "required_token_positions_per_second": 19203.577252298994,
+      "required_pages_per_second_upper_estimate": 0.003999598709265634
     },
     {
       "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": 28724.040720684417,
-      "required_pages_per_second_upper_estimate": 0.0017219623821343728
+      "required_token_positions_per_second": 28805.36587844849,
+      "required_pages_per_second_upper_estimate": 0.0017268377011660032
     },
     {
       "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": 28724.040720684417,
-      "required_pages_per_second_upper_estimate": 0.0022331678223304974
+      "required_token_positions_per_second": 28805.36587844849,
+      "required_pages_per_second_upper_estimate": 0.002239490495635089
     },
     {
       "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": 28724.040720684417,
-      "required_pages_per_second_upper_estimate": 0.36844115289291335
+      "required_token_positions_per_second": 28805.36587844849,
+      "required_pages_per_second_upper_estimate": 0.3694843046965597
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09155267565634517,
-    "producer_p95_seconds": 0.25772591100030695,
-    "gpu_update_seconds": 0.9625386716601626,
-    "producer_batches_per_second": 10.922673672079553,
-    "gpu_batches_per_second": 1.0389192968997547,
-    "producer_headroom": 10.513495807300883,
+    "producer_batch_seconds": 0.09839459959397345,
+    "producer_p95_seconds": 0.2616787309998472,
+    "gpu_update_seconds": 0.9598211707036707,
+    "producer_batches_per_second": 10.163159402309807,
+    "gpu_batches_per_second": 1.041860745024902,
+    "producer_headroom": 9.754815555573018,
     "sustainable": true,
-    "buffer_seconds": 60.639936314590244,
+    "buffer_seconds": 60.468733754331254,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427954.55291264923,
+    "generation_bytes_per_second": 426898.44844011555,
     "upload_bytes_per_second": 3879636.8138526846,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.82908478293045,
+    "estimated_full_shards_per_hour": 91.60246934797858,
     "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": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 550300, "loss": 0.5711182355880737, "lr": 0.0003, "grad_norm": 0.1465889956531403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80546357747, "tokens_per_second": 150211.515778511, "data_wait_s": 1.018900002236478e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 550400, "loss": 0.6104316711425781, "lr": 0.0003, "grad_norm": 0.16719218383811282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80561022373, "tokens_per_second": 152615.41820878926, "data_wait_s": 1.1588999768719077e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 550500, "loss": 0.5647913813591003, "lr": 0.0003, "grad_norm": 0.14572986109525263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80575750189, "tokens_per_second": 153869.68707765953, "data_wait_s": 1.0963998647639528e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 550600, "loss": 0.5639494061470032, "lr": 0.0003, "grad_norm": 0.1470442908817351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80590379800, "tokens_per_second": 153136.01655423455, "data_wait_s": 1.353900006506592e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5c458eb0292d0f59915a2ad6c7b431b6870242cd5fec73a7e641219226fcd8f
-size 544170
+oid sha256:f56c95a2b4d5b058622874138360100c8d826e8f60c13ce4ad886e8f99a054d2
+size 546288