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": 74854.88167634227,
-      "required_pages_per_second_upper_estimate": 0.006488231425742225
+      "required_token_positions_per_second": 75289.23599358511,
+      "required_pages_per_second_upper_estimate": 0.006525880157099882
     },
     {
       "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": 9356.860209542783,
-      "required_pages_per_second_upper_estimate": 0.0037325914351136044
+      "required_token_positions_per_second": 9411.15449919814,
+      "required_pages_per_second_upper_estimate": 0.0037542502390290966
     },
     {
       "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": 18713.720419085566,
-      "required_pages_per_second_upper_estimate": 0.004329089868724777
+      "required_token_positions_per_second": 18822.30899839628,
+      "required_pages_per_second_upper_estimate": 0.004354209925454587
     },
     {
       "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": 28070.58062862835,
-      "required_pages_per_second_upper_estimate": 0.0017093510853289165
+      "required_token_positions_per_second": 28233.463497594417,
+      "required_pages_per_second_upper_estimate": 0.0017192697974685819
     },
     {
       "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": 28070.58062862835,
-      "required_pages_per_second_upper_estimate": 0.0021609510011322895
+      "required_token_positions_per_second": 28233.463497594417,
+      "required_pages_per_second_upper_estimate": 0.002173490175273937
     },
     {
       "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": 28070.58062862835,
-      "required_pages_per_second_upper_estimate": 0.34854266521758137
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28233.463497594417,
+      "required_pages_per_second_upper_estimate": 0.3505651253162449
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07897261444520609,
-    "producer_p95_seconds": 0.3008080700019491,
-    "gpu_update_seconds": 0.9849457824111635,
-    "producer_batches_per_second": 12.662617377240743,
-    "gpu_batches_per_second": 1.0152843109312917,
-    "producer_headroom": 12.471991579999578,
+    "producer_batch_seconds": 0.09746277219517196,
+    "producer_p95_seconds": 0.2983498219982721,
+    "gpu_update_seconds": 0.9792634900197668,
+    "producer_batches_per_second": 10.260327892146057,
+    "gpu_batches_per_second": 1.0211756184025758,
+    "producer_headroom": 10.047564500410104,
     "sustainable": true,
-    "buffer_seconds": 60.08169272708097,
+    "buffer_seconds": 60.714336381225536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369447.29963066,
-    "upload_bytes_per_second": 4127713.71209437,
+    "generation_bytes_per_second": 368183.7058667921,
+    "upload_bytes_per_second": 4121481.18313686,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.27479020776605,
+    "estimated_full_shards_per_hour": 79.00365240099738,
     "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": 0.8377171489992179,
-      "bytes": 5823773,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.6930653950003034,
       "bytes": 12802437,
@@ -562,6 +556,12 @@
       "bytes": 6675780,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0791062770003919,
+      "bytes": 6176898,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -208,3 +208,4 @@
 {"step": 105800, "loss": 0.5691022537648678, "lr": 0.0003, "grad_norm": 0.13232588957847485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15482144994, "tokens_per_second": 147517.791146581, "data_wait_s": 1.3156997738406062e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105900, "loss": 0.6305570360273123, "lr": 0.0003, "grad_norm": 0.13164483919808131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15496873678, "tokens_per_second": 136118.9266714003, "data_wait_s": 1.3787997886538506e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106000, "loss": 0.6038626618683338, "lr": 0.0003, "grad_norm": 0.13520886709246546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15511593792, "tokens_per_second": 149484.15233918777, "data_wait_s": 1.5450998034793884e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 106100, "loss": 0.6192547734826803, "lr": 0.0003, "grad_norm": 0.1329948030528072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15526298252, "tokens_per_second": 149190.17782175785, "data_wait_s": 1.518700082669966e-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 sha256:8e249ad54ec700a20fcafc32fcd61f71c03b5fd1340443ae786a9ecb55b07dad
-size 448077
+oid sha256:76adebf89f2d1df0a39f8066bade20a1dfa049dbd2428931e0388f52e64b60b1
+size 450195