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": 75600.67446784506,
-      "required_pages_per_second_upper_estimate": 0.006552874854715138
+      "required_token_positions_per_second": 73773.05961907508,
+      "required_pages_per_second_upper_estimate": 0.006394461831671242
     },
     {
       "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": 9450.084308480633,
-      "required_pages_per_second_upper_estimate": 0.0037697799220044012
+      "required_token_positions_per_second": 9221.632452384385,
+      "required_pages_per_second_upper_estimate": 0.0036786470609479756
     },
     {
       "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": 18900.168616961266,
-      "required_pages_per_second_upper_estimate": 0.0043722213779271105
+      "required_token_positions_per_second": 18443.26490476877,
+      "required_pages_per_second_upper_estimate": 0.004266524745342069
     },
     {
       "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": 28350.2529254419,
-      "required_pages_per_second_upper_estimate": 0.001726381660877721
+      "required_token_positions_per_second": 27664.897357153157,
+      "required_pages_per_second_upper_estimate": 0.0016846471025517068
     },
     {
       "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": 28350.2529254419,
-      "required_pages_per_second_upper_estimate": 0.002182480948723464
+      "required_token_positions_per_second": 27664.897357153157,
+      "required_pages_per_second_upper_estimate": 0.0021297203798909584
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28350.2529254419,
-      "required_pages_per_second_upper_estimate": 0.3520152591410395
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27664.897357153157,
+      "required_pages_per_second_upper_estimate": 0.3435054367204286
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10639942882026787,
-    "producer_p95_seconds": 0.22485303800021939,
-    "gpu_update_seconds": 0.9752293946975095,
-    "producer_batches_per_second": 9.398546694167136,
-    "gpu_batches_per_second": 1.0253997730556241,
-    "producer_headroom": 9.165739003588893,
+    "producer_batch_seconds": 0.45107277337498886,
+    "producer_p95_seconds": 0.20726455800104304,
+    "gpu_update_seconds": 0.99938921309069,
+    "producer_batches_per_second": 2.2169371751654654,
+    "gpu_batches_per_second": 1.0006111601979584,
+    "producer_headroom": 2.2155830989601117,
     "sustainable": true,
-    "buffer_seconds": 60.464222471245584,
+    "buffer_seconds": 60.96274199853209,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 208411.00682648493,
-    "upload_bytes_per_second": 6477284.460919276,
+    "generation_bytes_per_second": 200840.21677173884,
+    "upload_bytes_per_second": 6470989.221340737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.72015050502692,
+    "estimated_full_shards_per_hour": 43.095635198250996,
     "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": 0.9253262060001362,
-      "bytes": 5975019,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.8811729529998047,
       "bytes": 3757694,
@@ -562,6 +556,12 @@
       "bytes": 8735177,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.964613582000311,
+      "bytes": 5800981,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -102,3 +102,4 @@
 {"step": 10200, "loss": 0.7503373622894287, "lr": 0.0003, "grad_norm": 0.18641110026414284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1491652255, "tokens_per_second": 149026.39579448832, "data_wait_s": 1.327900099568069e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 10300, "loss": 0.6576240304857492, "lr": 0.0003, "grad_norm": 0.1747125886445304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1506259651, "tokens_per_second": 148875.70960588314, "data_wait_s": 1.2914999388158321e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 10400, "loss": 0.7129305824637413, "lr": 0.0003, "grad_norm": 0.14615158784072166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1520980370, "tokens_per_second": 149812.06111519784, "data_wait_s": 1.2311000318732113e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 10500, "loss": 0.7457594126462936, "lr": 0.0003, "grad_norm": 0.15446676887494146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1535736075, "tokens_per_second": 149147.70182806658, "data_wait_s": 2.084299921989441e-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 sha256:8e0f802763dd66097648838fa46fb4e3b3ce51e7c03d234735239ebfcc32a61e
-size 218304
+oid sha256:15c1603c0ecb11f24b37fa52222685b5602a12e6d4e26555ccaafeca8c70cedd
+size 220387