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": 73200.60804078264,
-      "required_pages_per_second_upper_estimate": 0.006344843179730114
+      "required_token_positions_per_second": 75432.36722783995,
+      "required_pages_per_second_upper_estimate": 0.006538286409722294
     },
     {
       "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": 9150.07600509783,
-      "required_pages_per_second_upper_estimate": 0.0036501021242611414
+      "required_token_positions_per_second": 9429.045903479993,
+      "required_pages_per_second_upper_estimate": 0.0037613873876974602
     },
     {
       "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": 18300.15201019566,
-      "required_pages_per_second_upper_estimate": 0.0042334180958835
+      "required_token_positions_per_second": 18858.091806959987,
+      "required_pages_per_second_upper_estimate": 0.004362487648459879
     },
     {
       "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": 27450.22801529349,
-      "required_pages_per_second_upper_estimate": 0.001671574865915446
+      "required_token_positions_per_second": 28287.13771043998,
+      "required_pages_per_second_upper_estimate": 0.0017225382754240456
     },
     {
       "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": 27450.22801529349,
-      "required_pages_per_second_upper_estimate": 0.0021131945397118296
+      "required_token_positions_per_second": 28287.13771043998,
+      "required_pages_per_second_upper_estimate": 0.002177622164758519
     },
     {
       "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": 27450.22801529349,
-      "required_pages_per_second_upper_estimate": 0.34083996194660204
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28287.13771043998,
+      "required_pages_per_second_upper_estimate": 0.35123157940375205
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1227572436094988,
-    "producer_p95_seconds": 0.4096035850016051,
-    "gpu_update_seconds": 1.007204748339297,
-    "producer_batches_per_second": 8.146158797611038,
-    "gpu_batches_per_second": 0.9928467887475945,
-    "producer_headroom": 8.204849821679776,
+    "producer_batch_seconds": 0.11908409448381008,
+    "producer_p95_seconds": 0.4073327320002136,
+    "gpu_update_seconds": 0.9774053593904592,
+    "producer_batches_per_second": 8.397427081548273,
+    "gpu_batches_per_second": 1.02311696001302,
+    "producer_headroom": 8.207690234595864,
     "sustainable": true,
-    "buffer_seconds": 60.432284900357814,
+    "buffer_seconds": 60.59913228220847,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401764.79731006606,
-    "upload_bytes_per_second": 4001663.2314266916,
+    "generation_bytes_per_second": 385160.44685520086,
+    "upload_bytes_per_second": 4058092.963684238,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.2093728969239,
+    "estimated_full_shards_per_hour": 82.64646581880588,
     "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": 1.124354845000198,
-      "bytes": 904261,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.0335092330024054,
       "bytes": 5888838,
@@ -562,6 +556,12 @@
       "bytes": 11307353,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9193243289992097,
+      "bytes": 5971902,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 350100, "loss": 0.693084504455328, "lr": 0.0003, "grad_norm": 0.18748733395407843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51245680972, "tokens_per_second": 135294.73832244033, "data_wait_s": 1.1290001566521823e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350200, "loss": 0.6044878903776407, "lr": 0.0003, "grad_norm": 0.16243606073841022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51260237399, "tokens_per_second": 147832.4141225818, "data_wait_s": 6.558002496603876e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 350300, "loss": 0.6137314103543758, "lr": 0.0003, "grad_norm": 0.18405110382549328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51274718620, "tokens_per_second": 148061.49646377063, "data_wait_s": 1.1934993381146342e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 350400, "loss": 0.5734471511095762, "lr": 0.0003, "grad_norm": 0.1576187271171644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51289254835, "tokens_per_second": 147763.8878159974, "data_wait_s": 1.3365002814680338e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66231c6ed36b0e94d3c84aef778af68a4499487a45d7b496dee82a305e3e1f26
-size 753300
+oid sha256:df88914b0c56c528373b04dc32d3d3973600745318e0df92fa9da79f2716088c
+size 755418