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": 74982.73123579101,
-      "required_pages_per_second_upper_estimate": 0.006499313101523496
+      "required_token_positions_per_second": 73051.97955148797,
+      "required_pages_per_second_upper_estimate": 0.006331960438973514
     },
     {
       "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": 9372.841404473877,
-      "required_pages_per_second_upper_estimate": 0.0037389665727117745
+      "required_token_positions_per_second": 9131.497443935996,
+      "required_pages_per_second_upper_estimate": 0.003642690858439443
     },
     {
       "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": 18745.682808947753,
-      "required_pages_per_second_upper_estimate": 0.004336483805100545
+      "required_token_positions_per_second": 18262.99488787199,
+      "required_pages_per_second_upper_estimate": 0.004224822449576937
     },
     {
       "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": 28118.52421342163,
-      "required_pages_per_second_upper_estimate": 0.0017122705981022779
+      "required_token_positions_per_second": 27394.492331807985,
+      "required_pages_per_second_upper_estimate": 0.0016681808552137104
     },
     {
       "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": 28118.52421342163,
-      "required_pages_per_second_upper_estimate": 0.0021646418310060133
+      "required_token_positions_per_second": 27394.492331807985,
+      "required_pages_per_second_upper_estimate": 0.002108903852510869
     },
     {
       "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": 28118.52421342163,
-      "required_pages_per_second_upper_estimate": 0.3491379640838575
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27394.492331807985,
+      "required_pages_per_second_upper_estimate": 0.3401479112930453
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06103274574998352,
-    "producer_p95_seconds": 0.1992335989998537,
-    "gpu_update_seconds": 0.9832663972742552,
-    "producer_batches_per_second": 16.384647089227016,
-    "gpu_batches_per_second": 1.0170183815618357,
-    "producer_headroom": 16.11047291403436,
+    "producer_batch_seconds": 0.2381576133905554,
+    "producer_p95_seconds": 0.22262769200096955,
+    "gpu_update_seconds": 1.0092539648160468,
+    "producer_batches_per_second": 4.198899987967619,
+    "gpu_batches_per_second": 0.9908308858437495,
+    "producer_headroom": 4.2377564607223706,
     "sustainable": true,
-    "buffer_seconds": 60.96251663100382,
+    "buffer_seconds": 60.55523788896281,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345319.28604515665,
+    "generation_bytes_per_second": 345326.6161975366,
     "upload_bytes_per_second": 5167348.234019328,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.09748016372704,
+    "estimated_full_shards_per_hour": 74.09905304379056,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 203200, "loss": 0.5928876809775829, "lr": 0.0003, "grad_norm": 0.14121212756443854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29740710600, "tokens_per_second": 148024.2565727021, "data_wait_s": 1.2005999451503158e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203300, "loss": 0.5808873251080513, "lr": 0.0003, "grad_norm": 0.15666509603809722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29755398714, "tokens_per_second": 147805.4060011676, "data_wait_s": 1.6381000023102388e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203400, "loss": 0.6004175543785095, "lr": 0.0003, "grad_norm": 0.15220060382221018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29770145221, "tokens_per_second": 148313.08790736206, "data_wait_s": 1.804100065783132e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 203500, "loss": 0.5775925777852535, "lr": 0.0003, "grad_norm": 0.1425142528755673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29784752155, "tokens_per_second": 147981.58452242424, "data_wait_s": 1.9011000404134393e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5cc723813290fd17b39e49b31025781b815b72b520da1e7b86fd06315c16fd8
-size 178860
+oid sha256:e87ad8f3e633b999847a3649f11d2bc91f5dbbebc2b71c728b6f648dc7d7ec8b
+size 180978