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": 74962.37947844232,
-      "required_pages_per_second_upper_estimate": 0.00649754906277757
+      "required_token_positions_per_second": 75055.8456507442,
+      "required_pages_per_second_upper_estimate": 0.006505650473704848
     },
     {
       "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": 9370.29743480529,
-      "required_pages_per_second_upper_estimate": 0.0037379517451752394
+      "required_token_positions_per_second": 9381.980706343025,
+      "required_pages_per_second_upper_estimate": 0.003742612376872118
     },
     {
       "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": 18740.59486961058,
-      "required_pages_per_second_upper_estimate": 0.004335306799879564
+      "required_token_positions_per_second": 18763.96141268605,
+      "required_pages_per_second_upper_estimate": 0.004340712238382961
     },
     {
       "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": 28110.89230441587,
-      "required_pages_per_second_upper_estimate": 0.001711805854885359
+      "required_token_positions_per_second": 28145.942119029074,
+      "required_pages_per_second_upper_estimate": 0.0017139402047031376
     },
     {
       "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": 28110.89230441587,
-      "required_pages_per_second_upper_estimate": 0.0021640543055242733
+      "required_token_positions_per_second": 28145.942119029074,
+      "required_pages_per_second_upper_estimate": 0.0021667525372772944
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28110.89230441587,
-      "required_pages_per_second_upper_estimate": 0.3490432013163623
+      "required_token_positions_per_second": 28145.942119029074,
+      "required_pages_per_second_upper_estimate": 0.34947840270967473
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.147942729867367,
-    "producer_p95_seconds": 0.3215311179992568,
-    "gpu_update_seconds": 0.9835333471665304,
-    "producer_batches_per_second": 6.759372365891286,
-    "gpu_batches_per_second": 1.0167423431863378,
-    "producer_headroom": 6.648068127770006,
+    "producer_batch_seconds": 0.09162682660164023,
+    "producer_p95_seconds": 0.31174834500052384,
+    "gpu_update_seconds": 0.9823085645197706,
+    "producer_batches_per_second": 10.913834267639023,
+    "gpu_batches_per_second": 1.0180100592820123,
+    "producer_headroom": 10.72075287285117,
     "sustainable": true,
-    "buffer_seconds": 60.97906752432488,
+    "buffer_seconds": 60.90313100022578,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380135.2771340188,
-    "upload_bytes_per_second": 6015171.334181287,
+    "generation_bytes_per_second": 396785.66819408553,
+    "upload_bytes_per_second": 4062636.6653248286,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.5681813766043,
+    "estimated_full_shards_per_hour": 85.14096769682813,
     "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": 1.5552335219981615,
-      "bytes": 12252293,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.9495614830011618,
       "bytes": 5780919,
@@ -562,6 +556,12 @@
       "bytes": 1224241,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 34.26547425300305,
+      "bytes": 7641108,
+      "pages": 29,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -181,3 +181,4 @@
 {"step": 103100, "loss": 0.6555536761879921, "lr": 0.0003, "grad_norm": 0.1502817850151909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15087338923, "tokens_per_second": 150393.0267746047, "data_wait_s": 1.6738998965593055e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103200, "loss": 0.5946326497942209, "lr": 0.0003, "grad_norm": 0.14209048301726152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15102059084, "tokens_per_second": 149550.96910927029, "data_wait_s": 1.8296999769518152e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103300, "loss": 0.6290736962109804, "lr": 0.0003, "grad_norm": 0.14952531139442538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15116529417, "tokens_per_second": 146797.9021008877, "data_wait_s": 9.146002412308007e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 103400, "loss": 0.6758065614849329, "lr": 0.0003, "grad_norm": 0.14666754645000032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15131179122, "tokens_per_second": 148702.89277946352, "data_wait_s": 1.7318001482635736e-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 sha256:9944b080b8b9a1fe07d8cb543372239cf1b79d3aee7d329230fe4ae38c183cc4
-size 390108
+oid sha256:4bb05d57755c964d7db55e4b2af9c8770b81466e4ca5f16a23f0a05977ab0112
+size 392226