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": 76814.30900919597,
-      "required_pages_per_second_upper_estimate": 0.005314192131008898
+      "required_token_positions_per_second": 77617.99700492236,
+      "required_pages_per_second_upper_estimate": 0.005369793131366321
     },
     {
       "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": 9601.788626149497,
-      "required_pages_per_second_upper_estimate": 0.004020068389621738
+      "required_token_positions_per_second": 9702.249625615295,
+      "required_pages_per_second_upper_estimate": 0.004062129312233844
     },
     {
       "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": 19203.577252298994,
-      "required_pages_per_second_upper_estimate": 0.003999598709265634
+      "required_token_positions_per_second": 19404.49925123059,
+      "required_pages_per_second_upper_estimate": 0.004041445462973654
     },
     {
       "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": 28805.36587844849,
-      "required_pages_per_second_upper_estimate": 0.0017268377011660032
+      "required_token_positions_per_second": 29106.748876845886,
+      "required_pages_per_second_upper_estimate": 0.001744905151734734
     },
     {
       "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": 28805.36587844849,
-      "required_pages_per_second_upper_estimate": 0.002239490495635089
+      "required_token_positions_per_second": 29106.748876845886,
+      "required_pages_per_second_upper_estimate": 0.002262921698116775
     },
     {
       "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": 28805.36587844849,
-      "required_pages_per_second_upper_estimate": 0.3694843046965597
+      "required_token_positions_per_second": 29106.748876845886,
+      "required_pages_per_second_upper_estimate": 0.3733501222001499
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "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,
+    "producer_batch_seconds": 0.11524620646108019,
+    "producer_p95_seconds": 0.27684578000116744,
+    "gpu_update_seconds": 0.949882795807322,
+    "producer_batches_per_second": 8.677075200195072,
+    "gpu_batches_per_second": 1.0527614611127707,
+    "producer_headroom": 8.242204450591673,
     "sustainable": true,
-    "buffer_seconds": 60.468733754331254,
+    "buffer_seconds": 60.79249893166861,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426898.44844011555,
-    "upload_bytes_per_second": 3879636.8138526846,
+    "generation_bytes_per_second": 444871.7372911825,
+    "upload_bytes_per_second": 3817765.1137371226,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.60246934797858,
+    "estimated_full_shards_per_hour": 95.45911873866659,
     "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": 1.275124282001343,
-      "bytes": 10806597,
-      "pages": 46,
-      "shards": 3
-    },
     {
       "seconds": 0.989914507001231,
       "bytes": 672610,
@@ -562,6 +556,12 @@
       "bytes": 11040131,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9862021689987159,
+      "bytes": 2654656,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -254,3 +254,4 @@
 {"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}
+{"step": 550700, "loss": 0.6051796674728394, "lr": 0.0003, "grad_norm": 0.19117282867623211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80605073568, "tokens_per_second": 152610.71400702806, "data_wait_s": 1.5298999642254785e-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:f56c95a2b4d5b058622874138360100c8d826e8f60c13ce4ad886e8f99a054d2
-size 546288
+oid sha256:420e048fc53f4c60b530cf3cd5502b94fd75199540d79d77d69ef801b6ad3df6
+size 548406