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": 75970.7839611355,
-      "required_pages_per_second_upper_estimate": 0.006584955007559619
+      "required_token_positions_per_second": 74825.75348463499,
+      "required_pages_per_second_upper_estimate": 0.006485706667909765
     },
     {
       "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": 9496.347995141938,
-      "required_pages_per_second_upper_estimate": 0.0037882351983173517
+      "required_token_positions_per_second": 9353.219185579374,
+      "required_pages_per_second_upper_estimate": 0.0037311389762164407
     },
     {
       "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": 18992.695990283875,
-      "required_pages_per_second_upper_estimate": 0.004393625957319143
+      "required_token_positions_per_second": 18706.438371158747,
+      "required_pages_per_second_upper_estimate": 0.004327405295096573
     },
     {
       "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": 28489.043985425815,
-      "required_pages_per_second_upper_estimate": 0.0017348333082503266
+      "required_token_positions_per_second": 28059.657556738122,
+      "required_pages_per_second_upper_estimate": 0.0017086859275597306
     },
     {
       "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": 28489.043985425815,
-      "required_pages_per_second_upper_estimate": 0.0021931654687192695
+      "required_token_positions_per_second": 28059.657556738122,
+      "required_pages_per_second_upper_estimate": 0.0021601101128211815
     },
     {
       "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": 28489.043985425815,
-      "required_pages_per_second_upper_estimate": 0.35373857960224264
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28059.657556738122,
+      "required_pages_per_second_upper_estimate": 0.3484070372218747
     }
   ],
   "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": 0.13594559418746144,
-    "producer_p95_seconds": 0.3557829350029351,
-    "gpu_update_seconds": 0.970478335957638,
-    "producer_batches_per_second": 7.355883844393334,
-    "gpu_batches_per_second": 1.0304197043339778,
-    "producer_headroom": 7.138725912804516,
+    "producer_batch_seconds": 0.20454988270324748,
+    "producer_p95_seconds": 0.348550978997082,
+    "gpu_update_seconds": 0.9853292024000746,
+    "producer_batches_per_second": 4.888783052741999,
+    "gpu_batches_per_second": 1.0148892345463731,
+    "producer_headroom": 4.817060706065275,
     "sustainable": true,
-    "buffer_seconds": 60.16965682937356,
+    "buffer_seconds": 60.10508134640455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425698.9549265678,
-    "upload_bytes_per_second": 5454432.645918093,
+    "generation_bytes_per_second": 448366.07144974644,
+    "upload_bytes_per_second": 5517080.547400296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.34508596275116,
+    "estimated_full_shards_per_hour": 96.20892150515837,
     "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": 0.9564802560016687,
-      "bytes": 6117579,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0467319720009982,
       "bytes": 5732511,
@@ -562,6 +556,12 @@
       "bytes": 2675452,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1101803599995037,
+      "bytes": 11781038,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 180700, "loss": 0.5505474973469973, "lr": 0.0003, "grad_norm": 0.1358629476055678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26446880042, "tokens_per_second": 147843.4271292146, "data_wait_s": 7.974998879944906e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 180800, "loss": 0.5887566804885864, "lr": 0.0003, "grad_norm": 0.1595766333864267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26461442479, "tokens_per_second": 149108.69555331505, "data_wait_s": 1.1955999070778489e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 180900, "loss": 0.6022106818854809, "lr": 0.0003, "grad_norm": 0.1460082143141704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26476208702, "tokens_per_second": 149031.7166816439, "data_wait_s": 1.2014999811071903e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 181000, "loss": 0.5989765413105488, "lr": 0.0003, "grad_norm": 0.17027742300177381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26490912366, "tokens_per_second": 149997.79116370252, "data_wait_s": 1.3525001122616231e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c4f4b38ce675f95162c7e5b3ce7848d46552eadbd49842cf204a88e5e1d9c75
-size 552642
+oid sha256:9f583c226e83f86ab8e3826e9b44ad7ae186d91103c3c84e1fa0d600be8165ac
+size 554760