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": 74483.26741268262,
-      "required_pages_per_second_upper_estimate": 0.0064560208432159465
+      "required_token_positions_per_second": 74728.27772474983,
+      "required_pages_per_second_upper_estimate": 0.006477257716092974
     },
     {
       "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": 9310.408426585327,
-      "required_pages_per_second_upper_estimate": 0.0037140611243758285
+      "required_token_positions_per_second": 9341.034715593729,
+      "required_pages_per_second_upper_estimate": 0.0037262784089651065
     },
     {
       "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": 18620.816853170654,
-      "required_pages_per_second_upper_estimate": 0.0043075982637972784
+      "required_token_positions_per_second": 18682.069431187458,
+      "required_pages_per_second_upper_estimate": 0.004321767969712911
     },
     {
       "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": 27931.22527975598,
-      "required_pages_per_second_upper_estimate": 0.001700865075723599
+      "required_token_positions_per_second": 28023.10414678119,
+      "required_pages_per_second_upper_estimate": 0.0017064600166742736
     },
     {
       "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": 27931.22527975598,
-      "required_pages_per_second_upper_estimate": 0.002150223040615799
+      "required_token_positions_per_second": 28023.10414678119,
+      "required_pages_per_second_upper_estimate": 0.0021572961301363817
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27931.22527975598,
-      "required_pages_per_second_upper_estimate": 0.3468123381769371
+      "required_token_positions_per_second": 28023.10414678119,
+      "required_pages_per_second_upper_estimate": 0.34795316620660305
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07739173575775737,
-    "producer_p95_seconds": 0.20608820100096636,
-    "gpu_update_seconds": 0.9898599049300835,
-    "producer_batches_per_second": 12.921276286270203,
-    "gpu_batches_per_second": 1.0102439698985815,
-    "producer_headroom": 12.790253316302763,
+    "producer_batch_seconds": 0.10642547539843861,
+    "producer_p95_seconds": 0.2121063199992932,
+    "gpu_update_seconds": 0.9866144683752219,
+    "producer_batches_per_second": 9.396246493202616,
+    "gpu_batches_per_second": 1.0135671349385558,
+    "producer_headroom": 9.270472738613641,
     "sustainable": true,
-    "buffer_seconds": 60.38145420073509,
+    "buffer_seconds": 60.183482570888536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 239834.91922316788,
-    "upload_bytes_per_second": 5007847.888621609,
+    "generation_bytes_per_second": 231155.03750350277,
+    "upload_bytes_per_second": 4963393.792519046,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.46299059411313,
+    "estimated_full_shards_per_hour": 49.60049003437817,
     "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": 1.3395652879999034,
-      "bytes": 11637618,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 0.9964154829999643,
       "bytes": 2622281,
@@ -562,6 +556,12 @@
       "bytes": 2328336,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.028129626000009,
+      "bytes": 6087526,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -84,3 +84,4 @@
 {"step": 8400, "loss": 0.7150008901953697, "lr": 0.0003, "grad_norm": 0.172544872834465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1228789213, "tokens_per_second": 148404.7426836842, "data_wait_s": 1.5133000488276593e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8500, "loss": 0.7317186035215855, "lr": 0.0003, "grad_norm": 0.1523427707709444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1243424486, "tokens_per_second": 148826.6440943452, "data_wait_s": 1.1719001122401096e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8600, "loss": 0.7029031589627266, "lr": 0.0003, "grad_norm": 0.1475043923616545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1257838036, "tokens_per_second": 146864.6213369466, "data_wait_s": 9.540999599266797e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 8700, "loss": 0.7436679638922215, "lr": 0.0003, "grad_norm": 0.1510031678001198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1272326942, "tokens_per_second": 147700.68508203633, "data_wait_s": 1.547600004414562e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1df888205b8947fd63fe253da461f3464ac7fad76d84b6d28e4cbae10998cb5
-size 180037
+oid sha256:60d52439fd11d8b95aa76a37e1a0ecb3dc579a6f770ec49f3ac676b6d66f787a
+size 182120