Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75541.45076049298,
-      "required_pages_per_second_upper_estimate": 0.006547741493862974
+      "required_token_positions_per_second": 75432.39668341892,
+      "required_pages_per_second_upper_estimate": 0.006538288962857225
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9442.681345061623,
-      "required_pages_per_second_upper_estimate": 0.0037668267692123913
+      "required_token_positions_per_second": 9429.049585427365,
+      "required_pages_per_second_upper_estimate": 0.003761388856481317
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18885.362690123246,
-      "required_pages_per_second_upper_estimate": 0.004368796287328547
+      "required_token_positions_per_second": 18858.09917085473,
+      "required_pages_per_second_upper_estimate": 0.0043624893519673275
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28328.044035184867,
-      "required_pages_per_second_upper_estimate": 0.0017250292559821097
+      "required_token_positions_per_second": 28287.148756282095,
+      "required_pages_per_second_upper_estimate": 0.0017225389480578772
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28328.044035184867,
-      "required_pages_per_second_upper_estimate": 0.0021807712468733345
+      "required_token_positions_per_second": 28287.148756282095,
+      "required_pages_per_second_upper_estimate": 0.002177623015098022
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28328.044035184867,
-      "required_pages_per_second_upper_estimate": 0.35173949905242147
+      "required_token_positions_per_second": 28287.148756282095,
+      "required_pages_per_second_upper_estimate": 0.35123171655614305
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06498833267976067,
-    "producer_p95_seconds": 0.21245356899999024,
-    "gpu_update_seconds": 0.9759939643435948,
-    "producer_batches_per_second": 15.387377376299886,
-    "gpu_batches_per_second": 1.0245965001151933,
-    "producer_headroom": 15.017987446345868,
+    "producer_batch_seconds": 0.12325831225786033,
+    "producer_p95_seconds": 0.3108334569988074,
+    "gpu_update_seconds": 0.9774049777236685,
+    "producer_batches_per_second": 8.113043101774492,
+    "gpu_batches_per_second": 1.023117359529879,
+    "producer_headroom": 7.92972871216106,
     "sustainable": true,
-    "buffer_seconds": 60.511625789302876,
+    "buffer_seconds": 60.59910861886745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282455.2507749909,
+    "generation_bytes_per_second": 283821.52381626284,
     "upload_bytes_per_second": 5839741.981799227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.60832159459395,
+    "estimated_full_shards_per_hour": 60.9014919840423,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -159,3 +159,4 @@
 {"step": 130900, "loss": 0.627563763409853, "lr": 0.0003, "grad_norm": 0.13571884190445932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19156667235, "tokens_per_second": 148638.82289956714, "data_wait_s": 1.6165999113582075e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 131000, "loss": 0.6175642311573029, "lr": 0.0003, "grad_norm": 0.1527774523782554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19171133531, "tokens_per_second": 148219.90765500045, "data_wait_s": 1.1596999684115872e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 131100, "loss": 0.622124794870615, "lr": 0.0003, "grad_norm": 0.1508883228916356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19185718227, "tokens_per_second": 148411.04620728875, "data_wait_s": 1.0424999345559627e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 131200, "loss": 0.6396949496120214, "lr": 0.0003, "grad_norm": 0.17329341245236954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19200236062, "tokens_per_second": 133299.3226844823, "data_wait_s": 1.3447002856992185e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f383c73792659cd2b5726125493dd8ee2ba9be5d1c3f3b7d17faa6de95b6b62d
-size 343512
+oid sha256:9ca41d7c5e31e705b02d0f3f8ce3dc141f1e2ed09db5b6cce3dc1ad6642df867
+size 345630