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": 77186.93472175555,
-      "required_pages_per_second_upper_estimate": 0.005339971242414542
+      "required_token_positions_per_second": 76647.38330744517,
+      "required_pages_per_second_upper_estimate": 0.005302643823640787
     },
     {
       "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": 9648.366840219443,
-      "required_pages_per_second_upper_estimate": 0.004039569715189111
+      "required_token_positions_per_second": 9580.922913430646,
+      "required_pages_per_second_upper_estimate": 0.004011332351431985
     },
     {
       "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": 19296.733680438887,
-      "required_pages_per_second_upper_estimate": 0.004019000736546958
+      "required_token_positions_per_second": 19161.845826861292,
+      "required_pages_per_second_upper_estimate": 0.003990907153878657
     },
     {
       "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": 28945.10052065833,
-      "required_pages_per_second_upper_estimate": 0.0017352145796040385
+      "required_token_positions_per_second": 28742.768740291936,
+      "required_pages_per_second_upper_estimate": 0.0017230850983138135
     },
     {
       "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": 28945.10052065833,
-      "required_pages_per_second_upper_estimate": 0.002250354249439169
+      "required_token_positions_per_second": 28742.768740291936,
+      "required_pages_per_second_upper_estimate": 0.0022346238434791292
     },
     {
       "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": 28945.10052065833,
-      "required_pages_per_second_upper_estimate": 0.37127667065145814
+      "required_token_positions_per_second": 28742.768740291936,
+      "required_pages_per_second_upper_estimate": 0.3686813758198578
     }
   ],
   "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": 0.11523099663281755,
-    "producer_p95_seconds": 0.24934382699939306,
-    "gpu_update_seconds": 0.9551875620631347,
-    "producer_batches_per_second": 8.6782205241745,
-    "gpu_batches_per_second": 1.046914804711311,
-    "producer_headroom": 8.2893283055325,
+    "producer_batch_seconds": 0.6466151579999746,
+    "producer_p95_seconds": 0.2634155939995253,
+    "gpu_update_seconds": 0.9619115071973815,
+    "producer_batches_per_second": 1.546514936477934,
+    "gpu_batches_per_second": 1.039596670294124,
+    "producer_headroom": 1.4876105134507522,
     "sustainable": true,
-    "buffer_seconds": 60.176816409977484,
+    "buffer_seconds": 60.60042495343504,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367857.32091990195,
-    "upload_bytes_per_second": 3512872.2481645616,
+    "generation_bytes_per_second": 306619.2364339552,
+    "upload_bytes_per_second": 3491492.341204832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.93361778924745,
+    "estimated_full_shards_per_hour": 65.79335040821069,
     "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": 1.815276637998977,
-      "bytes": 10307802,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1417645319997973,
       "bytes": 2370286,
@@ -562,6 +556,12 @@
       "bytes": 798109,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.6809108209999977,
+      "bytes": 7259646,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -190,3 +190,4 @@
 {"step": 861500, "loss": 0.532051146030426, "lr": 3e-05, "grad_norm": 0.2918758091793663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126099099552, "tokens_per_second": 151780.9125423944, "data_wait_s": 1.42680000863038e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861600, "loss": 0.49249354004859924, "lr": 3e-05, "grad_norm": 0.2872346917661992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126113604782, "tokens_per_second": 150743.31168528905, "data_wait_s": 1.1613999959081411e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861700, "loss": 0.49686741828918457, "lr": 3e-05, "grad_norm": 0.304775112780428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126128248430, "tokens_per_second": 151406.8000760035, "data_wait_s": 1.575900023453869e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 861800, "loss": 0.4809182584285736, "lr": 3e-05, "grad_norm": 0.3109340984030623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126142892141, "tokens_per_second": 151753.0956627732, "data_wait_s": 1.397199957864359e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:440798900d97f4eeb8747317b367344c47459723a7fb6e3231052daeba0f23ed
-size 409953
+oid sha256:76aadee0790df912183f77597ec1022d1bc88bc380c99a7bec460afe5ac051ed
+size 412071