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": 75372.49826663696,
-      "required_pages_per_second_upper_estimate": 0.006533097119901715
+      "required_token_positions_per_second": 74265.49277086511,
+      "required_pages_per_second_upper_estimate": 0.006437144689207989
     },
     {
       "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": 9421.56228332962,
-      "required_pages_per_second_upper_estimate": 0.0037584020597293843
+      "required_token_positions_per_second": 9283.186596358139,
+      "required_pages_per_second_upper_estimate": 0.0037032019292955716
     },
     {
       "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": 18843.12456665924,
-      "required_pages_per_second_upper_estimate": 0.004359025240830733
+      "required_token_positions_per_second": 18566.373192716277,
+      "required_pages_per_second_upper_estimate": 0.004295003681126857
     },
     {
       "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": 28264.686849988862,
-      "required_pages_per_second_upper_estimate": 0.0017211711358131338
+      "required_token_positions_per_second": 27849.559789074417,
+      "required_pages_per_second_upper_estimate": 0.001695892075806808
     },
     {
       "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": 28264.686849988862,
-      "required_pages_per_second_upper_estimate": 0.0021758938353730324
+      "required_token_positions_per_second": 27849.559789074417,
+      "required_pages_per_second_upper_estimate": 0.0021439362050785856
     },
     {
       "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": 28264.686849988862,
-      "required_pages_per_second_upper_estimate": 0.3509528148551456
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27849.559789074417,
+      "required_pages_per_second_upper_estimate": 0.3457983260994874
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.07302337479687537,
-    "producer_p95_seconds": 0.18822209500103781,
-    "gpu_update_seconds": 0.978181720064268,
-    "producer_batches_per_second": 13.69424520274006,
-    "gpu_batches_per_second": 1.0223049352571203,
-    "producer_headroom": 13.395460327398123,
+    "producer_batch_seconds": 0.3996763170702593,
+    "producer_p95_seconds": 0.20647266800006037,
+    "gpu_update_seconds": 0.9927625502664681,
+    "producer_batches_per_second": 2.5020246566778925,
+    "gpu_batches_per_second": 1.007290212278444,
+    "producer_headroom": 2.4839163789931287,
     "sustainable": true,
-    "buffer_seconds": 60.647266643984615,
+    "buffer_seconds": 60.558515566254556,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330898.9716154215,
-    "upload_bytes_per_second": 5455770.564442175,
+    "generation_bytes_per_second": 301794.5100952643,
+    "upload_bytes_per_second": 5535286.687176615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.00321637484535,
+    "estimated_full_shards_per_hour": 64.75807644980857,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 1242347,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.07777969900053,
+      "bytes": 7140650,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 91300, "loss": 0.661818029358983, "lr": 0.0003, "grad_norm": 0.13152827585548027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13361373514, "tokens_per_second": 150239.82422898128, "data_wait_s": 7.86600139690563e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 91400, "loss": 0.6057727262377739, "lr": 0.0003, "grad_norm": 0.12462032566448279, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13375999485, "tokens_per_second": 149018.34903248824, "data_wait_s": 1.842299934651237e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91500, "loss": 0.6084352731704712, "lr": 0.0003, "grad_norm": 0.12859608848031054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13390574656, "tokens_per_second": 148032.37163812228, "data_wait_s": 1.0539999493630603e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 91600, "loss": 0.6036572568118572, "lr": 0.0003, "grad_norm": 0.1329230079673119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13405010480, "tokens_per_second": 146787.359406769, "data_wait_s": 1.424999936716631e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a93840f5d4b39318b65547fecee9de49476d4101082fdf7204309f1286514460
-size 138618
+oid sha256:b3709f687a478ca90eeb43146219c05b7f247e1951306f4e3ae091458d974d79
+size 140736