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": 75572.20614675905,
-      "required_pages_per_second_upper_estimate": 0.006550407292795703
+      "required_token_positions_per_second": 75501.62196555822,
+      "required_pages_per_second_upper_estimate": 0.006544289234863187
     },
     {
       "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": 9446.525768344882,
-      "required_pages_per_second_upper_estimate": 0.003768360367139334
+      "required_token_positions_per_second": 9437.702745694778,
+      "required_pages_per_second_upper_estimate": 0.003764840731488263
     },
     {
       "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": 18893.051536689763,
-      "required_pages_per_second_upper_estimate": 0.004370574966662633
+      "required_token_positions_per_second": 18875.405491389556,
+      "required_pages_per_second_upper_estimate": 0.00436649286464222
     },
     {
       "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": 28339.577305034643,
-      "required_pages_per_second_upper_estimate": 0.0017257315716055707
+      "required_token_positions_per_second": 28313.108237084336,
+      "required_pages_per_second_upper_estimate": 0.001724119744239864
     },
     {
       "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": 28339.577305034643,
-      "required_pages_per_second_upper_estimate": 0.0021816591099125047
+      "required_token_positions_per_second": 28313.108237084336,
+      "required_pages_per_second_upper_estimate": 0.0021796214477906206
     },
     {
       "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": 28339.577305034643,
-      "required_pages_per_second_upper_estimate": 0.3518827036645845
+      "required_token_positions_per_second": 28313.108237084336,
+      "required_pages_per_second_upper_estimate": 0.35155404642691357
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09052265928124825,
-    "producer_p95_seconds": 0.32191508799951407,
-    "gpu_update_seconds": 0.9755967671080337,
-    "producer_batches_per_second": 11.046957832878753,
-    "gpu_batches_per_second": 1.025013646738811,
-    "producer_headroom": 10.777376348135283,
+    "producer_batch_seconds": 0.10809764025776758,
+    "producer_p95_seconds": 0.3252332199990633,
+    "gpu_update_seconds": 0.9765088229976396,
+    "producer_batches_per_second": 9.25089574217734,
+    "gpu_batches_per_second": 1.024056287510284,
+    "producer_headroom": 9.03358131286747,
     "sustainable": true,
-    "buffer_seconds": 60.48699956069809,
+    "buffer_seconds": 60.543547025853655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325803.2771982294,
-    "upload_bytes_per_second": 5595198.95543236,
+    "generation_bytes_per_second": 402222.2851045854,
+    "upload_bytes_per_second": 5564590.260246406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.90979897461091,
+    "estimated_full_shards_per_hour": 86.30753912785694,
     "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": 1.2015854260007472,
-      "bytes": 10309364,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9169029310014594,
       "bytes": 7362656,
@@ -562,6 +556,12 @@
       "bytes": 928094,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 0.8634128480007348,
+      "bytes": 6113242,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -211,3 +211,4 @@
 {"step": 21100, "loss": 0.6893913336098194, "lr": 0.0003, "grad_norm": 0.1397953381680224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3087368786, "tokens_per_second": 147324.48810078556, "data_wait_s": 1.7758997273631394e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 21200, "loss": 0.6821452751755714, "lr": 0.0003, "grad_norm": 0.14234147294255003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3102061033, "tokens_per_second": 149341.20047718784, "data_wait_s": 1.0720003047026694e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 21300, "loss": 0.7059006355702877, "lr": 0.0003, "grad_norm": 0.13807118533059076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3116664584, "tokens_per_second": 148218.4060777925, "data_wait_s": 1.1599000572459772e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 21400, "loss": 0.6534912511706352, "lr": 0.0003, "grad_norm": 0.13234207093614064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3131302397, "tokens_per_second": 148239.98097128168, "data_wait_s": 1.206399974762462e-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 sha256:740452582f596f6f74dcbc88c8d26dee4e531d3ffdee25a98bd497f9e9677c5c
-size 448657
+oid sha256:2e27cf811d08ffbaaf04555428da1e88c4eb00162ba2d75e23001618b87fa33d
+size 450775