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": 75395.9757884658,
-      "required_pages_per_second_upper_estimate": 0.006535132091990603
+      "required_token_positions_per_second": 74760.56770212285,
+      "required_pages_per_second_upper_estimate": 0.006480056529493468
     },
     {
       "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": 9424.496973558225,
-      "required_pages_per_second_upper_estimate": 0.0037595727515391036
+      "required_token_positions_per_second": 9345.070962765356,
+      "required_pages_per_second_upper_estimate": 0.0037278885283091416
     },
     {
       "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": 18848.99394711645,
-      "required_pages_per_second_upper_estimate": 0.004360383018701941
+      "required_token_positions_per_second": 18690.141925530712,
+      "required_pages_per_second_upper_estimate": 0.004323635399208175
     },
     {
       "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": 28273.490920674678,
-      "required_pages_per_second_upper_estimate": 0.001721707257526511
+      "required_token_positions_per_second": 28035.212888296068,
+      "required_pages_per_second_upper_estimate": 0.001707197375502873
     },
     {
       "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": 28273.490920674678,
-      "required_pages_per_second_upper_estimate": 0.002176571597105654
+      "required_token_positions_per_second": 28035.212888296068,
+      "required_pages_per_second_upper_estimate": 0.002158228294577339
     },
     {
       "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": 28273.490920674678,
-      "required_pages_per_second_upper_estimate": 0.35106213194773433
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28035.212888296068,
+      "required_pages_per_second_upper_estimate": 0.3481035162508669
     }
   ],
   "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": 0.36143964611724755,
-    "producer_p95_seconds": 0.3351231319975341,
-    "gpu_update_seconds": 0.977877124461582,
-    "producer_batches_per_second": 2.766713642906815,
-    "gpu_batches_per_second": 1.022623369526717,
-    "producer_headroom": 2.7055059813343445,
+    "producer_batch_seconds": 0.12429042299220328,
+    "producer_p95_seconds": 0.33666718899985426,
+    "gpu_update_seconds": 0.9861883378649955,
+    "producer_batches_per_second": 8.045672191997689,
+    "gpu_batches_per_second": 1.0140050957861715,
+    "producer_headroom": 7.934548086032815,
     "sustainable": true,
-    "buffer_seconds": 60.62838171661809,
+    "buffer_seconds": 60.15748860976473,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339232.11293893395,
+    "generation_bytes_per_second": 355953.6487992073,
     "upload_bytes_per_second": 3437263.3607144994,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.79131451726927,
+    "estimated_full_shards_per_hour": 76.37936685545124,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -279,3 +279,4 @@
 {"step": 67900, "loss": 0.5869335904717445, "lr": 0.0003, "grad_norm": 0.13526097671186324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9937397120, "tokens_per_second": 148687.36729860492, "data_wait_s": 1.2147000234108418e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 68000, "loss": 0.5799515973776579, "lr": 0.0003, "grad_norm": 0.13291845899974705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9951939582, "tokens_per_second": 148279.82142983278, "data_wait_s": 9.306000720243901e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 68100, "loss": 0.6183673404157162, "lr": 0.0003, "grad_norm": 0.14735262095656268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9966627267, "tokens_per_second": 149102.65479761144, "data_wait_s": 1.5125999198062345e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 68200, "loss": 0.694906510412693, "lr": 0.0003, "grad_norm": 0.1386208189926631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9981284720, "tokens_per_second": 149909.89668038773, "data_wait_s": 8.10399797046557e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59add0d6f50ffa52ed1c43a187a91d51f1123babcb6adeb4cedeb1a716b89d4b
-size 599238
+oid sha256:337079e41faa3bff4a2a074a0f7ea18850b45bcaf00e9557a71d2cf92dca4944
+size 601356