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": 74627.63820974228,
-      "required_pages_per_second_upper_estimate": 0.006468534538000639
+      "required_token_positions_per_second": 75601.37298276144,
+      "required_pages_per_second_upper_estimate": 0.00655293540021772
     },
     {
       "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": 9328.454776217784,
-      "required_pages_per_second_upper_estimate": 0.0037212600830611874
+      "required_token_positions_per_second": 9450.17162284518,
+      "required_pages_per_second_upper_estimate": 0.003769814753009885
     },
     {
       "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": 18656.90955243557,
-      "required_pages_per_second_upper_estimate": 0.004315947674562405
+      "required_token_positions_per_second": 18900.34324569036,
+      "required_pages_per_second_upper_estimate": 0.0043722617752102256
     },
     {
       "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": 27985.36432865335,
-      "required_pages_per_second_upper_estimate": 0.0017041618597558115
+      "required_token_positions_per_second": 28350.51486853554,
+      "required_pages_per_second_upper_estimate": 0.0017263976118378147
     },
     {
       "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": 27985.36432865335,
-      "required_pages_per_second_upper_estimate": 0.002154390814466398
+      "required_token_positions_per_second": 28350.51486853554,
+      "required_pages_per_second_upper_estimate": 0.0021825011138279175
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27985.36432865335,
-      "required_pages_per_second_upper_estimate": 0.3474845639725015
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28350.51486853554,
+      "required_pages_per_second_upper_estimate": 0.3520185115975954
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5709764443436427,
-    "producer_p95_seconds": 0.3146800370013807,
-    "gpu_update_seconds": 0.9879449727832225,
-    "producer_batches_per_second": 1.7513857356226574,
-    "gpu_batches_per_second": 1.0122021241555754,
-    "producer_headroom": 1.7302727329126504,
+    "producer_batch_seconds": 0.09264454103140451,
+    "producer_p95_seconds": 0.2275148660010018,
+    "gpu_update_seconds": 0.9752203841167195,
+    "producer_batches_per_second": 10.79394413170034,
+    "gpu_batches_per_second": 1.0254092472705274,
+    "producer_headroom": 10.526474342251214,
     "sustainable": true,
-    "buffer_seconds": 60.26464333977657,
+    "buffer_seconds": 60.46366381523661,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 180407.0831318308,
-    "upload_bytes_per_second": 6093389.06911037,
+    "generation_bytes_per_second": 180806.24797445786,
+    "upload_bytes_per_second": 6128555.531456836,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 38.711160378133705,
+    "estimated_full_shards_per_hour": 38.79681186127951,
     "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": 1.1199327470003482,
-      "bytes": 6072182,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 0.9855451060011546,
       "bytes": 8735177,
@@ -562,6 +556,12 @@
       "bytes": 4735418,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0588101130015275,
+      "bytes": 8209294,
+      "pages": 47,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -176,3 +176,4 @@
 {"step": 17600, "loss": 0.6619683243334293, "lr": 0.0003, "grad_norm": 0.13800178226358997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2575613586, "tokens_per_second": 150652.68747997703, "data_wait_s": 1.6114001482492313e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17700, "loss": 0.6716101486235857, "lr": 0.0003, "grad_norm": 0.13604749911766006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2590120230, "tokens_per_second": 147127.25941452244, "data_wait_s": 1.0526000551180914e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17800, "loss": 0.6375476289540529, "lr": 0.0003, "grad_norm": 0.1435686926483544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2604557371, "tokens_per_second": 142427.46112601698, "data_wait_s": 1.2294000043766573e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 17900, "loss": 0.7036605700850487, "lr": 0.0003, "grad_norm": 0.1436700177685992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2619072920, "tokens_per_second": 148300.02777097083, "data_wait_s": 1.3864999345969409e-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 sha256:29a3546b9826a47fd676646c9c499a055455b5b2db3b946789b06416cb24398a
-size 373744
+oid sha256:2b711d3d803d8215a5a71a2948bafbd4edca160151db8a2864d58f1e13b8ce83
+size 375862