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": 77500.08118848542,
-      "required_pages_per_second_upper_estimate": 0.005361635441582823
+      "required_token_positions_per_second": 77625.1623749524,
+      "required_pages_per_second_upper_estimate": 0.005370288848290948
     },
     {
       "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": 9687.510148560677,
-      "required_pages_per_second_upper_estimate": 0.004055958200986357
+      "required_token_positions_per_second": 9703.14529686905,
+      "required_pages_per_second_upper_estimate": 0.004062504311084048
     },
     {
       "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": 19375.020297121355,
-      "required_pages_per_second_upper_estimate": 0.004035305774244998
+      "required_token_positions_per_second": 19406.2905937381,
+      "required_pages_per_second_upper_estimate": 0.004041818552377089
     },
     {
       "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": 29062.53044568203,
-      "required_pages_per_second_upper_estimate": 0.0017422543243040959
+      "required_token_positions_per_second": 29109.435890607154,
+      "required_pages_per_second_upper_estimate": 0.0017450662341068897
     },
     {
       "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": 29062.53044568203,
-      "required_pages_per_second_upper_estimate": 0.002259483909590109
+      "required_token_positions_per_second": 29109.435890607154,
+      "required_pages_per_second_upper_estimate": 0.002263130601617789
     },
     {
       "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": 29062.53044568203,
-      "required_pages_per_second_upper_estimate": 0.3727829356432323
+      "required_token_positions_per_second": 29109.435890607154,
+      "required_pages_per_second_upper_estimate": 0.37338458832758886
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12746190218766174,
-    "producer_p95_seconds": 0.2733534459985094,
-    "gpu_update_seconds": 0.951328035653131,
-    "producer_batches_per_second": 7.845481534770314,
-    "gpu_batches_per_second": 1.051162125494865,
-    "producer_headroom": 7.463626537225953,
+    "producer_batch_seconds": 0.07049209470346796,
+    "producer_p95_seconds": 0.26394312900083605,
+    "gpu_update_seconds": 0.9497951146803666,
+    "producer_batches_per_second": 14.18598786440664,
+    "gpu_batches_per_second": 1.0528586476637425,
+    "producer_headroom": 13.473781970528393,
     "sustainable": true,
-    "buffer_seconds": 60.88499428180038,
+    "buffer_seconds": 60.78688733954346,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358224.3771892655,
-    "upload_bytes_per_second": 3801139.738771449,
+    "generation_bytes_per_second": 357956.8866297491,
+    "upload_bytes_per_second": 3811008.0018162453,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.86661230810617,
+    "estimated_full_shards_per_hour": 76.80921506089548,
     "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": 0.8898182180000731,
-      "bytes": 896140,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.8745819040013885,
       "bytes": 13367716,
@@ -562,6 +556,12 @@
       "bytes": 5848497,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0607272609995562,
+      "bytes": 2667610,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 667500, "loss": 0.530362069606781, "lr": 0.0001, "grad_norm": 0.2233472783097298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97704981479, "tokens_per_second": 151789.40905616796, "data_wait_s": 1.690299905021675e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 667600, "loss": 0.5643426775932312, "lr": 0.0001, "grad_norm": 0.19327152648849397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97719647119, "tokens_per_second": 153327.2008079842, "data_wait_s": 1.5107998478924856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 667700, "loss": 0.5745958685874939, "lr": 0.0001, "grad_norm": 0.21178127155839305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97734395267, "tokens_per_second": 153873.54285004552, "data_wait_s": 8.281000191345811e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 667800, "loss": 0.5833641886711121, "lr": 0.0001, "grad_norm": 0.19772748002870502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97749062969, "tokens_per_second": 152365.44524410993, "data_wait_s": 1.1034000635845587e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d298b2f440335d8b0cbbcda3555e888a764faed533914c6b74d2d80b7abc61c4
-size 484083
+oid sha256:f00e0e8600f076a19353645a9ccac8c5d83c15a66ebce43c53fdbbc282c50ca5
+size 486201