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": 75261.52950120384,
-      "required_pages_per_second_upper_estimate": 0.006523478628561737
+      "required_token_positions_per_second": 74617.75281290904,
+      "required_pages_per_second_upper_estimate": 0.006467677696857442
     },
     {
       "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": 9407.69118765048,
-      "required_pages_per_second_upper_estimate": 0.0037528686722716133
+      "required_token_positions_per_second": 9327.21910161363,
+      "required_pages_per_second_upper_estimate": 0.003720767153986608
     },
     {
       "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": 18815.38237530096,
-      "required_pages_per_second_upper_estimate": 0.004352607573105887
+      "required_token_positions_per_second": 18654.43820322726,
+      "required_pages_per_second_upper_estimate": 0.004315375971417322
     },
     {
       "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": 28223.073562951442,
-      "required_pages_per_second_upper_estimate": 0.0017186371049605989
+      "required_token_positions_per_second": 27981.657304840886,
+      "required_pages_per_second_upper_estimate": 0.0017039361214548831
     },
     {
       "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": 28223.073562951442,
-      "required_pages_per_second_upper_estimate": 0.0021726903293439407
+      "required_token_positions_per_second": 27981.657304840886,
+      "required_pages_per_second_upper_estimate": 0.002154105437511616
     },
     {
       "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": 28223.073562951442,
-      "required_pages_per_second_upper_estimate": 0.35043611710085354
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27981.657304840886,
+      "required_pages_per_second_upper_estimate": 0.34743853514336126
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07683684628892706,
-    "producer_p95_seconds": 0.28828056400016067,
-    "gpu_update_seconds": 0.9796239923455274,
-    "producer_batches_per_second": 13.01458933178664,
-    "gpu_batches_per_second": 1.0207998250488803,
-    "producer_headroom": 12.749403959942338,
+    "producer_batch_seconds": 0.13528123724216812,
+    "producer_p95_seconds": 0.32201978899684036,
+    "gpu_update_seconds": 0.9880758562223131,
+    "producer_batches_per_second": 7.392008089118015,
+    "gpu_batches_per_second": 1.012068044879951,
+    "producer_headroom": 7.303864721857547,
     "sustainable": true,
-    "buffer_seconds": 60.7366875254227,
+    "buffer_seconds": 60.272627229561095,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 259959.99335736595,
-    "upload_bytes_per_second": 5045240.789449554,
+    "generation_bytes_per_second": 262812.34543405013,
+    "upload_bytes_per_second": 5066114.178175092,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.781363015563336,
+    "estimated_full_shards_per_hour": 56.393411371861724,
     "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": 1.2802494179995847,
-      "bytes": 4852642,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.42962039700069,
       "bytes": 6308465,
@@ -562,6 +556,12 @@
       "bytes": 1024816,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.4308035579997522,
+      "bytes": 7307496,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -241,3 +241,4 @@
 {"step": 261200, "loss": 0.6021216567605734, "lr": 0.0003, "grad_norm": 0.16239392872977046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38232537918, "tokens_per_second": 127206.86807270769, "data_wait_s": 1.2266998965060338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261300, "loss": 0.5932094510644674, "lr": 0.0003, "grad_norm": 0.16197659898215175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38247263359, "tokens_per_second": 150637.85039217872, "data_wait_s": 1.0947998816845939e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261400, "loss": 0.609815489500761, "lr": 0.0003, "grad_norm": 0.16562488191582922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38261832378, "tokens_per_second": 147960.23528569652, "data_wait_s": 1.4483000995824113e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 261500, "loss": 0.5521674687042832, "lr": 0.0003, "grad_norm": 0.14959994957872588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38276317946, "tokens_per_second": 147578.46847866013, "data_wait_s": 1.921099828905426e-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:d15b1ef36414914e68ab556fbc6573bc65dcc7d9839dafa7d719cd73e65becb5
-size 518754
+oid sha256:ef7db463c0f543c047f8d12611d323aec59e161fba7b4111bc10dedc5a91a8fe
+size 520872