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": 74859.59955597336,
-      "required_pages_per_second_upper_estimate": 0.006488640359591291
+      "required_token_positions_per_second": 75033.46580332551,
+      "required_pages_per_second_upper_estimate": 0.006503710645251801
     },
     {
       "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": 9357.44994449667,
-      "required_pages_per_second_upper_estimate": 0.0037328266892040327
+      "required_token_positions_per_second": 9379.183225415689,
+      "required_pages_per_second_upper_estimate": 0.003741496419904136
     },
     {
       "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": 18714.89988899334,
-      "required_pages_per_second_upper_estimate": 0.004329362718329979
+      "required_token_positions_per_second": 18758.366450831378,
+      "required_pages_per_second_upper_estimate": 0.004339417942425848
     },
     {
       "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": 28072.349833490007,
-      "required_pages_per_second_upper_estimate": 0.0017094588206226907
+      "required_token_positions_per_second": 28137.54967624707,
+      "required_pages_per_second_upper_estimate": 0.001713429148969458
     },
     {
       "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": 28072.349833490007,
-      "required_pages_per_second_upper_estimate": 0.0021610871994200125
+      "required_token_positions_per_second": 28137.54967624707,
+      "required_pages_per_second_upper_estimate": 0.002166106463800167
     },
     {
       "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": 28072.349833490007,
-      "required_pages_per_second_upper_estimate": 0.34856463282081535
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28137.54967624707,
+      "required_pages_per_second_upper_estimate": 0.3493741966580214
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.32710891848431345,
-    "producer_p95_seconds": 0.23389736100034497,
-    "gpu_update_seconds": 0.9848837081324855,
-    "producer_batches_per_second": 3.0570857090463437,
-    "gpu_batches_per_second": 1.0153483012691698,
-    "producer_headroom": 3.0108739092043915,
+    "producer_batch_seconds": 0.09271371146876106,
+    "producer_p95_seconds": 0.20319681000000855,
+    "gpu_update_seconds": 0.98260155266255,
+    "producer_batches_per_second": 10.78589114984292,
+    "gpu_batches_per_second": 1.0177065131744454,
+    "producer_headroom": 10.59823339068491,
     "sustainable": true,
-    "buffer_seconds": 60.077906196081614,
+    "buffer_seconds": 60.9212962650781,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319877.3417543802,
+    "generation_bytes_per_second": 319916.6696652353,
     "upload_bytes_per_second": 4045039.515985068,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.63823117707781,
+    "estimated_full_shards_per_hour": 68.64667003124042,
     "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": [

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

@@ -122,3 +122,4 @@
 {"step": 327200, "loss": 0.5817774944007397, "lr": 0.0003, "grad_norm": 0.13918941547010838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47893097448, "tokens_per_second": 148836.4416416254, "data_wait_s": 1.1533000360941514e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327300, "loss": 0.5769318398088217, "lr": 0.0003, "grad_norm": 0.15720138966766922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47907710928, "tokens_per_second": 148577.17979066292, "data_wait_s": 1.4534000001731329e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327400, "loss": 0.6022498216480017, "lr": 0.0003, "grad_norm": 0.1598802955628392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47922314253, "tokens_per_second": 148855.92447491825, "data_wait_s": 1.4339999324874952e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 327500, "loss": 0.6007689274847507, "lr": 0.0003, "grad_norm": 0.13406897571219664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47936968395, "tokens_per_second": 137695.06942940524, "data_wait_s": 8.6880008893786e-06, "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:fcbbecf113ce0b244f777093941e33971b283460ef8d219eca1aef2a21f8dad1
-size 264363
+oid sha256:90153b64de14f3e9f74a6ca276ca2e6bfe2237b069a3dd5e6f89db0ad544881c
+size 266481