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": 75596.35244143053,
-      "required_pages_per_second_upper_estimate": 0.006552500232419602
+      "required_token_positions_per_second": 75990.2426011391,
+      "required_pages_per_second_upper_estimate": 0.006586641633157659
     },
     {
       "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": 9449.544055178816,
-      "required_pages_per_second_upper_estimate": 0.0037695644068847997
+      "required_token_positions_per_second": 9498.780325142388,
+      "required_pages_per_second_upper_estimate": 0.003789205491121106
     },
     {
       "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": 18899.088110357632,
-      "required_pages_per_second_upper_estimate": 0.0043719714217406265
+      "required_token_positions_per_second": 18997.560650284777,
+      "required_pages_per_second_upper_estimate": 0.00439475131079526
     },
     {
       "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": 28348.632165536448,
-      "required_pages_per_second_upper_estimate": 0.0017262829651029492
+      "required_token_positions_per_second": 28496.340975427167,
+      "required_pages_per_second_upper_estimate": 0.001735277656657061
     },
     {
       "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": 28348.632165536448,
-      "required_pages_per_second_upper_estimate": 0.0021823561781393916
+      "required_token_positions_per_second": 28496.340975427167,
+      "required_pages_per_second_upper_estimate": 0.002193727211208935
     },
     {
       "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": 28348.632165536448,
-      "required_pages_per_second_upper_estimate": 0.3519951347273483
+      "required_token_positions_per_second": 28496.340975427167,
+      "required_pages_per_second_upper_estimate": 0.3538291837966049
     }
   ],
   "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": 0.5399337224299643,
-    "producer_p95_seconds": 0.39453778200550005,
-    "gpu_update_seconds": 0.975285150922089,
-    "producer_batches_per_second": 1.8520791690867422,
-    "gpu_batches_per_second": 1.025341151820618,
-    "producer_headroom": 1.8063053119424206,
+    "producer_batch_seconds": 0.12312449342209675,
+    "producer_p95_seconds": 0.4103311769940774,
+    "gpu_update_seconds": 0.9702298278870713,
+    "producer_batches_per_second": 8.121860827250586,
+    "gpu_batches_per_second": 1.0306836290302073,
+    "producer_headroom": 7.880071632546083,
     "sustainable": true,
-    "buffer_seconds": 60.467679357169516,
+    "buffer_seconds": 60.154249328998425,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246726.6696211935,
-    "upload_bytes_per_second": 5850956.289891373,
+    "generation_bytes_per_second": 246258.1079768137,
+    "upload_bytes_per_second": 5895426.093877118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.941799797791035,
+    "estimated_full_shards_per_hour": 52.84125737646397,
     "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": 1.351508649000607,
-      "bytes": 6405900,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9840362190006999,
       "bytes": 2390157,
@@ -562,6 +556,12 @@
       "bytes": 2340987,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 1.4316346049963613,
+      "bytes": 10012608,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 152800, "loss": 0.6182908602058887, "lr": 0.0003, "grad_norm": 0.13115770544587926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22362109494, "tokens_per_second": 150142.12544732486, "data_wait_s": 1.6645004507154226e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152900, "loss": 0.6384348757565022, "lr": 0.0003, "grad_norm": 0.1458762184071941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22376662281, "tokens_per_second": 135295.5858555299, "data_wait_s": 1.1884003470186144e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 153000, "loss": 0.6109720841050148, "lr": 0.0003, "grad_norm": 0.1675781607171175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22391140179, "tokens_per_second": 147429.25054257218, "data_wait_s": 9.397001122124493e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 153100, "loss": 0.5912684127688408, "lr": 0.0003, "grad_norm": 0.14043988631070992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22405728801, "tokens_per_second": 149638.74641906188, "data_wait_s": 9.56199801294133e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8def7c9e4c4b3ff6f4a59e18778d139fca4cef6b867dc90de19fcbab1f36f2c
-size 810486
+oid sha256:89df9902108287d42f8548f2bc1dacba862978fc744233cbe8def1ad86c48c82
+size 812604