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": 75434.5597620719,
-      "required_pages_per_second_upper_estimate": 0.006538476453032594
+      "required_token_positions_per_second": 74707.7772132036,
+      "required_pages_per_second_upper_estimate": 0.006475480783710219
     },
     {
       "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": 9429.319970258988,
-      "required_pages_per_second_upper_estimate": 0.003761496717033265
+      "required_token_positions_per_second": 9338.47215165045,
+      "required_pages_per_second_upper_estimate": 0.0037252561638943874
     },
     {
       "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": 18858.639940517976,
-      "required_pages_per_second_upper_estimate": 0.004362614449511698
+      "required_token_positions_per_second": 18676.9443033009,
+      "required_pages_per_second_upper_estimate": 0.004320582361575526
     },
     {
       "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": 28287.959910776965,
-      "required_pages_per_second_upper_estimate": 0.001722588343110816
+      "required_token_positions_per_second": 28015.41645495135,
+      "required_pages_per_second_upper_estimate": 0.0017059918765760386
     },
     {
       "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": 28287.959910776965,
-      "required_pages_per_second_upper_estimate": 0.0021776854600164563
+      "required_token_positions_per_second": 28015.41645495135,
+      "required_pages_per_second_upper_estimate": 0.002156704310338428
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28287.959910776965,
-      "required_pages_per_second_upper_estimate": 0.3512417883802099
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28015.41645495135,
+      "required_pages_per_second_upper_estimate": 0.3478577108031259
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09643186840626683,
-    "producer_p95_seconds": 0.16153355100004774,
-    "gpu_update_seconds": 0.9773769507311428,
-    "producer_batches_per_second": 10.37001581040623,
-    "gpu_batches_per_second": 1.0231466981617825,
-    "producer_headroom": 10.13541443180858,
+    "producer_batch_seconds": 0.3660056987890492,
+    "producer_p95_seconds": 0.16309903100000156,
+    "gpu_update_seconds": 0.9868852045964708,
+    "producer_batches_per_second": 2.73219789557528,
+    "gpu_batches_per_second": 1.013289078955127,
+    "producer_headroom": 2.696365679172857,
     "sustainable": true,
-    "buffer_seconds": 60.59737094533086,
+    "buffer_seconds": 60.19999748038472,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280175.03120112076,
-    "upload_bytes_per_second": 4665296.133160948,
+    "generation_bytes_per_second": 285604.6425620745,
+    "upload_bytes_per_second": 4761158.233335389,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.119039554836434,
+    "estimated_full_shards_per_hour": 61.284107758013505,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 7714712,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9509194960000968,
+      "bytes": 6981450,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 317300, "loss": 0.6179237812757492, "lr": 0.0003, "grad_norm": 0.16623904177407978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46444267414, "tokens_per_second": 147122.58520179737, "data_wait_s": 1.336800005447003e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317400, "loss": 0.5875317845493555, "lr": 0.0003, "grad_norm": 0.1388448377576693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46458853229, "tokens_per_second": 147464.9528725937, "data_wait_s": 1.1488999916764442e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317500, "loss": 0.536181753501296, "lr": 0.0003, "grad_norm": 0.1482592861132939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46473593505, "tokens_per_second": 148694.15235386262, "data_wait_s": 9.55299992710934e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 317600, "loss": 0.5978003963828087, "lr": 0.0003, "grad_norm": 0.165373911410232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46488188880, "tokens_per_second": 149238.6228716883, "data_wait_s": 1.0754999948403565e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1751c851c5d7999343d64baf4944862d3a3f9a0de3e777ad05a22efa07965ed
-size 53115
+oid sha256:c4a7f783e1777740f99118de056ba720f300f94bc0d8d2d228e7dcb2a3ab516d
+size 55233