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": 74521.91460269527,
-      "required_pages_per_second_upper_estimate": 0.006459370683695836
+      "required_token_positions_per_second": 75891.19692138766,
+      "required_pages_per_second_upper_estimate": 0.006578056604665783
     },
     {
       "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": 9315.239325336908,
-      "required_pages_per_second_upper_estimate": 0.0037159882421162072
+      "required_token_positions_per_second": 9486.399615173457,
+      "required_pages_per_second_upper_estimate": 0.0037842666408063894
     },
     {
       "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": 18630.478650673816,
-      "required_pages_per_second_upper_estimate": 0.004309833350607807
+      "required_token_positions_per_second": 18972.799230346915,
+      "required_pages_per_second_upper_estimate": 0.004389023192078742
     },
     {
       "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": 27945.717976010725,
-      "required_pages_per_second_upper_estimate": 0.0017017476048882122
+      "required_token_positions_per_second": 28459.198845520375,
+      "required_pages_per_second_upper_estimate": 0.0017330158958154328
     },
     {
       "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": 27945.717976010725,
-      "required_pages_per_second_upper_estimate": 0.0021513387284918402
+      "required_token_positions_per_second": 28459.198845520375,
+      "required_pages_per_second_upper_estimate": 0.002190867907232714
     },
     {
       "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": 27945.717976010725,
-      "required_pages_per_second_upper_estimate": 0.3469922889604868
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28459.198845520375,
+      "required_pages_per_second_upper_estimate": 0.3533680028498749
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10614449128152614,
-    "producer_p95_seconds": 0.33165954000287456,
-    "gpu_update_seconds": 0.9893465619217122,
-    "producer_batches_per_second": 9.4211200970167,
-    "gpu_batches_per_second": 1.010768155961036,
-    "producer_headroom": 9.320752777435022,
+    "producer_batch_seconds": 0.23709400589845586,
+    "producer_p95_seconds": 0.2783397060011339,
+    "gpu_update_seconds": 0.9714960758409381,
+    "producer_batches_per_second": 4.217736320285914,
+    "gpu_batches_per_second": 1.0293402360214257,
+    "producer_headroom": 4.097514284089564,
     "sustainable": true,
-    "buffer_seconds": 60.35014027722445,
+    "buffer_seconds": 60.23275670213816,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406829.53270557406,
-    "upload_bytes_per_second": 5156587.121692639,
+    "generation_bytes_per_second": 392399.50034052937,
+    "upload_bytes_per_second": 5142995.5752351815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.29614721179405,
+    "estimated_full_shards_per_hour": 84.19979818021689,
     "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": 1.3715665409999929,
-      "bytes": 4880519,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2786054230000445,
       "bytes": 6003136,
@@ -562,6 +556,12 @@
       "bytes": 2623268,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 2.5355882220028434,
+      "bytes": 9765836,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 259400, "loss": 0.6004841439425945, "lr": 0.0003, "grad_norm": 0.17627090721865235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37968785192, "tokens_per_second": 149266.07668842596, "data_wait_s": 9.916999260894954e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 259500, "loss": 0.588921282440424, "lr": 0.0003, "grad_norm": 0.14287418928485318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37983546941, "tokens_per_second": 150425.63166043407, "data_wait_s": 1.1239997547818348e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 259600, "loss": 0.5525398049503565, "lr": 0.0003, "grad_norm": 0.14585862086876325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37998285802, "tokens_per_second": 149504.3915322681, "data_wait_s": 1.2092001270502806e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 259700, "loss": 0.6227231193333864, "lr": 0.0003, "grad_norm": 0.14113326448330837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38012973655, "tokens_per_second": 97217.35659860563, "data_wait_s": 1.955900006578304e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3562e10906ed7d20b189f394fc8491af7863fe2de86589f4d0d26ae503b63b3
-size 479847
+oid sha256:8eda5607f881040c3aeb4223b9eef721c7ef50406821c4987c0f69da6c81a8d1
+size 481965