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": 73371.98419753442,
-      "required_pages_per_second_upper_estimate": 0.00635969763064846
+      "required_token_positions_per_second": 75118.31038190557,
+      "required_pages_per_second_upper_estimate": 0.006511064758286502
     },
     {
       "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": 9171.498024691802,
-      "required_pages_per_second_upper_estimate": 0.0036586476881649123
+      "required_token_positions_per_second": 9389.788797738196,
+      "required_pages_per_second_upper_estimate": 0.0037457271412710213
     },
     {
       "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": 18342.996049383604,
-      "required_pages_per_second_upper_estimate": 0.004243329310320294
+      "required_token_positions_per_second": 18779.577595476392,
+      "required_pages_per_second_upper_estimate": 0.004344324767436077
     },
     {
       "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": 27514.494074075403,
-      "required_pages_per_second_upper_estimate": 0.0016754883317173126
+      "required_token_positions_per_second": 28169.36639321459,
+      "required_pages_per_second_upper_estimate": 0.001715366620103367
     },
     {
       "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": 27514.494074075403,
-      "required_pages_per_second_upper_estimate": 0.002118141918816698
+      "required_token_positions_per_second": 28169.36639321459,
+      "required_pages_per_second_upper_estimate": 0.0021685558027466365
     },
     {
       "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": 27514.494074075403,
-      "required_pages_per_second_upper_estimate": 0.34163793131201065
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28169.36639321459,
+      "required_pages_per_second_upper_estimate": 0.3497692537990562
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4393155994767426,
-    "producer_p95_seconds": 0.3536314320008387,
-    "gpu_update_seconds": 1.004852203553704,
-    "producer_batches_per_second": 2.276267906696402,
-    "gpu_batches_per_second": 0.9951712266375653,
-    "producer_headroom": 2.287312821922457,
+    "producer_batch_seconds": 0.11264662064058939,
+    "producer_p95_seconds": 0.33853923000060604,
+    "gpu_update_seconds": 0.9814917245231268,
+    "producer_batches_per_second": 8.877319126959012,
+    "gpu_batches_per_second": 1.0188572914212453,
+    "producer_headroom": 8.71301525906114,
     "sustainable": true,
-    "buffer_seconds": 60.29113221322224,
+    "buffer_seconds": 60.85248692043386,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 504213.01993429364,
+    "generation_bytes_per_second": 502877.6777402546,
     "upload_bytes_per_second": 5913166.47007264,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.19237659951789,
+    "estimated_full_shards_per_hour": 107.90584324985245,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 25300, "loss": 0.6347766313701868, "lr": 0.0003, "grad_norm": 0.13177125174549342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3702096044, "tokens_per_second": 148876.27846530694, "data_wait_s": 1.0828000085894018e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 25400, "loss": 0.6552773602306843, "lr": 0.0003, "grad_norm": 0.13864205696092802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3716699649, "tokens_per_second": 148996.9698914158, "data_wait_s": 1.0299001587554812e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 25500, "loss": 0.6764032393693924, "lr": 0.0003, "grad_norm": 0.1288349939310712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3731302372, "tokens_per_second": 147038.4015671184, "data_wait_s": 1.3329001376405358e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 25600, "loss": 0.6448036432266235, "lr": 0.0003, "grad_norm": 0.14769247642991706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3745867477, "tokens_per_second": 148652.01603643238, "data_wait_s": 1.3130000297678635e-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:bd48092ac9721f4a114d72bd39bac8664b4c3112d62120c7588d50329917f58e
-size 538396
+oid sha256:65498c332989c68ede56789164b56e290f2e15503d94b8ccb2b099868992e769
+size 540514