Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76124.22362125282,
-      "required_pages_per_second_upper_estimate": 0.005266450422652285
+      "required_token_positions_per_second": 76505.58002849238,
+      "required_pages_per_second_upper_estimate": 0.005292833543773894
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9515.527952656603,
-      "required_pages_per_second_upper_estimate": 0.003983952846957963
+      "required_token_positions_per_second": 9563.197503561547,
+      "required_pages_per_second_upper_estimate": 0.004003911092468443
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19031.055905313206,
-      "required_pages_per_second_upper_estimate": 0.003963667062382368
+      "required_token_positions_per_second": 19126.395007123094,
+      "required_pages_per_second_upper_estimate": 0.003983523683028177
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28546.58385796981,
-      "required_pages_per_second_upper_estimate": 0.0017113241141755724
+      "required_token_positions_per_second": 28689.59251068464,
+      "required_pages_per_second_upper_estimate": 0.0017198972645442573
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28546.58385796981,
-      "required_pages_per_second_upper_estimate": 0.0022193713318046844
+      "required_token_positions_per_second": 28689.59251068464,
+      "required_pages_per_second_upper_estimate": 0.002230489625524677
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28546.58385796981,
-      "required_pages_per_second_upper_estimate": 0.36616492679634444
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28689.59251068464,
+      "required_pages_per_second_upper_estimate": 0.36799928824264233
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07783344999998576,
-    "producer_p95_seconds": 0.19317685299938603,
-    "gpu_update_seconds": 0.9685221929727,
-    "producer_batches_per_second": 12.847946480596491,
-    "gpu_batches_per_second": 1.0325008629184682,
-    "producer_headroom": 12.443521300583196,
+    "producer_batch_seconds": 0.10589637207023372,
+    "producer_p95_seconds": 0.19007688599958783,
+    "gpu_update_seconds": 0.9636944125192183,
+    "producer_batches_per_second": 9.443194138291814,
+    "gpu_batches_per_second": 1.0376733402302025,
+    "producer_headroom": 9.100353427406056,
     "sustainable": true,
-    "buffer_seconds": 60.048375964307404,
+    "buffer_seconds": 60.71274798871075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348039.40157982556,
-    "upload_bytes_per_second": 4733492.384435788,
+    "generation_bytes_per_second": 329743.2298169648,
+    "upload_bytes_per_second": 4730993.788843567,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.6811536364181,
+    "estimated_full_shards_per_hour": 70.75522108918865,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 1181953,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.210031186001288,
+      "bytes": 5583238,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 729000, "loss": 0.5009329319000244, "lr": 5e-05, "grad_norm": 0.20051109723056734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106705132571, "tokens_per_second": 151160.04991436226, "data_wait_s": 1.3786999261355959e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 729100, "loss": 0.49145078659057617, "lr": 5e-05, "grad_norm": 0.2316350361631726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106719782811, "tokens_per_second": 150992.7338657109, "data_wait_s": 1.003099896479398e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 729200, "loss": 0.5399318337440491, "lr": 5e-05, "grad_norm": 0.215438854976121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106734428913, "tokens_per_second": 151357.1710056955, "data_wait_s": 1.3409000530373305e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 729300, "loss": 0.5082528591156006, "lr": 5e-05, "grad_norm": 0.21188878347115842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106749031340, "tokens_per_second": 141508.57364701325, "data_wait_s": 9.015999239636585e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a33b3625b13ebcdfccf9272289780f7b0efd5b02977863c03987c7946feeaedb
-size 195804
+oid sha256:6fd2ea1828fa1123ec7f737d6f782d8c18049d57ee5941457ce58a634494a69f
+size 197922