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": 76484.13551232943,
-      "required_pages_per_second_upper_estimate": 0.005291349962387609
+      "required_token_positions_per_second": 76307.43237288605,
+      "required_pages_per_second_upper_estimate": 0.005279125229193144
     },
     {
       "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": 9560.51693904118,
-      "required_pages_per_second_upper_estimate": 0.004002788796080317
+      "required_token_positions_per_second": 9538.429046610756,
+      "required_pages_per_second_upper_estimate": 0.003993541056767343
     },
     {
       "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": 19121.03387808236,
-      "required_pages_per_second_upper_estimate": 0.003982407101231471
+      "required_token_positions_per_second": 19076.85809322151,
+      "required_pages_per_second_upper_estimate": 0.003973206450238746
     },
     {
       "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": 28681.550817123538,
-      "required_pages_per_second_upper_estimate": 0.0017194151772942244
+      "required_token_positions_per_second": 28615.287139832268,
+      "required_pages_per_second_upper_estimate": 0.0017154427710193896
     },
     {
       "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": 28681.550817123538,
-      "required_pages_per_second_upper_estimate": 0.0022298644192219733
+      "required_token_positions_per_second": 28615.287139832268,
+      "required_pages_per_second_upper_estimate": 0.0022247127097756907
     },
     {
       "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": 28681.550817123538,
-      "required_pages_per_second_upper_estimate": 0.36789613803213833
+      "required_token_positions_per_second": 28615.287139832268,
+      "required_pages_per_second_upper_estimate": 0.36704617872823936
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2629896840078345,
-    "producer_p95_seconds": 0.2785621760012873,
-    "gpu_update_seconds": 0.963964611826133,
-    "producer_batches_per_second": 3.802430516514898,
-    "gpu_batches_per_second": 1.0373824803647114,
-    "producer_headroom": 3.665408456848126,
+    "producer_batch_seconds": 0.10983881710154719,
+    "producer_p95_seconds": 0.28670929899817565,
+    "gpu_update_seconds": 0.9661968396435969,
+    "producer_batches_per_second": 9.104249539354461,
+    "gpu_batches_per_second": 1.0349857906478686,
+    "producer_headroom": 8.796497132250954,
     "sustainable": true,
-    "buffer_seconds": 60.72977054504638,
+    "buffer_seconds": 60.87040089754661,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305487.83355709194,
-    "upload_bytes_per_second": 4809986.688386439,
+    "generation_bytes_per_second": 307012.07180832996,
+    "upload_bytes_per_second": 4840114.741581453,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.55057768854684,
+    "estimated_full_shards_per_hour": 65.87764373481201,
     "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": 1.6235010039999906,
-      "bytes": 6476306,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 3.9122833619985613,
       "bytes": 6289054,
@@ -562,6 +556,12 @@
       "bytes": 10959545,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.2728104929992696,
+      "bytes": 7484326,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -253,3 +253,4 @@
 {"step": 950300, "loss": 0.5320939421653748, "lr": 3e-05, "grad_norm": 0.34095561353472936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139098139897, "tokens_per_second": 143550.26608170656, "data_wait_s": 1.6551999578950927e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 950400, "loss": 0.4987389147281647, "lr": 3e-05, "grad_norm": 0.33573880354791236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139112727748, "tokens_per_second": 151723.88885000825, "data_wait_s": 1.0514999303268269e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 950500, "loss": 0.5391099452972412, "lr": 3e-05, "grad_norm": 0.34557957165445835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139127313949, "tokens_per_second": 151654.7177385169, "data_wait_s": 1.4551002095686272e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 950600, "loss": 0.5312020182609558, "lr": 3e-05, "grad_norm": 0.3469100994073712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139141801350, "tokens_per_second": 150682.1748941606, "data_wait_s": 1.0732001101132482e-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 sha256:4a80bc885ee6c2aab964e66c47b95f1d014e8509a30d9d7923bbdc5af498d4cd
-size 544170
+oid sha256:c7d5d9c9fdda9be6b005e31f265998ca1a5c92e9b199ab2f39a9fbc6702b914d
+size 546288