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": 74883.09354224037,
-      "required_pages_per_second_upper_estimate": 0.006490676758775953
+      "required_token_positions_per_second": 75779.02099599062,
+      "required_pages_per_second_upper_estimate": 0.006568333479759652
     },
     {
       "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": 9360.386692780046,
-      "required_pages_per_second_upper_estimate": 0.0037339982020025714
+      "required_token_positions_per_second": 9472.377624498828,
+      "required_pages_per_second_upper_estimate": 0.0037786730590788366
     },
     {
       "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": 18720.773385560093,
-      "required_pages_per_second_upper_estimate": 0.004330721448390688
+      "required_token_positions_per_second": 18944.755248997655,
+      "required_pages_per_second_upper_estimate": 0.004382535710550806
     },
     {
       "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": 28081.160078340137,
-      "required_pages_per_second_upper_estimate": 0.0017099953183102813
+      "required_token_positions_per_second": 28417.132873496485,
+      "required_pages_per_second_upper_estimate": 0.0017304542988222764
     },
     {
       "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": 28081.160078340137,
-      "required_pages_per_second_upper_estimate": 0.002161765436456894
+      "required_token_positions_per_second": 28417.132873496485,
+      "required_pages_per_second_upper_estimate": 0.0021876295522602514
     },
     {
       "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": 28081.160078340137,
-      "required_pages_per_second_upper_estimate": 0.3486740265758613
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28417.132873496485,
+      "required_pages_per_second_upper_estimate": 0.35284568426309004
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.3936577923827542,
-    "producer_p95_seconds": 0.2776268670004356,
-    "gpu_update_seconds": 0.9845747085543575,
-    "producer_batches_per_second": 2.5402774169593934,
-    "gpu_batches_per_second": 1.0156669588520015,
-    "producer_headroom": 2.5010928974500106,
+    "producer_batch_seconds": 0.09406349160936145,
+    "producer_p95_seconds": 0.22384833799878834,
+    "gpu_update_seconds": 0.9729341845667399,
+    "producer_batches_per_second": 10.63111716236225,
+    "gpu_batches_per_second": 1.027818752658293,
+    "producer_headroom": 10.34337730739639,
     "sustainable": true,
-    "buffer_seconds": 60.059057221815806,
+    "buffer_seconds": 60.32191944313787,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295819.1451632663,
-    "upload_bytes_per_second": 5825619.230898786,
+    "generation_bytes_per_second": 297259.3497845693,
+    "upload_bytes_per_second": 5917001.810635777,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.475902234778324,
+    "estimated_full_shards_per_hour": 63.78493662026223,
     "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": 1.0933355659999506,
-      "bytes": 5690232,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.9681096090007486,
       "bytes": 2564629,
@@ -562,6 +556,12 @@
       "bytes": 979107,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.0416768409995711,
+      "bytes": 11876824,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -152,3 +152,4 @@
 {"step": 130200, "loss": 0.5923965331166983, "lr": 0.0003, "grad_norm": 0.13168526747637468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19054230913, "tokens_per_second": 150204.11931707352, "data_wait_s": 9.613000656827353e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 130300, "loss": 0.6134939007461071, "lr": 0.0003, "grad_norm": 0.14417946800087983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19068832999, "tokens_per_second": 138295.00993779086, "data_wait_s": 9.932000466505997e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 130400, "loss": 0.6093156430870295, "lr": 0.0003, "grad_norm": 0.12806106978875761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19083296183, "tokens_per_second": 147766.4906708138, "data_wait_s": 1.3663999197888188e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 130500, "loss": 0.5987209919840097, "lr": 0.0003, "grad_norm": 0.14051451020549627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19097971148, "tokens_per_second": 150032.96474808763, "data_wait_s": 1.0470999768585898e-05, "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 sha256:6447620cc9e206b91b906e3bb12bc41e455b4825f407e1fc2939cf0b6dd4ea91
-size 328686
+oid sha256:eb0fc8cc151e8318e5b5d881cda55941a26198b693db73ce76acc31d7b86f212
+size 330804