Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76407.24462989796,
-      "required_pages_per_second_upper_estimate": 0.005286030472729572
+      "required_token_positions_per_second": 76682.41715121405,
+      "required_pages_per_second_upper_estimate": 0.005305067546242434
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9550.905578737245,
-      "required_pages_per_second_upper_estimate": 0.003998764720229087
+      "required_token_positions_per_second": 9585.302143901756,
+      "required_pages_per_second_upper_estimate": 0.004013165843781508
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19101.81115747449,
-      "required_pages_per_second_upper_estimate": 0.003978403515466085
+      "required_token_positions_per_second": 19170.60428780351,
+      "required_pages_per_second_upper_estimate": 0.003992731310316774
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28652.71673621174,
-      "required_pages_per_second_upper_estimate": 0.0017176866181706544
+      "required_token_positions_per_second": 28755.906431705265,
+      "required_pages_per_second_upper_estimate": 0.0017238726828539494
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28652.71673621174,
-      "required_pages_per_second_upper_estimate": 0.002227622696258799
+      "required_token_positions_per_second": 28755.906431705265,
+      "required_pages_per_second_upper_estimate": 0.002235645241199918
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28652.71673621174,
-      "required_pages_per_second_upper_estimate": 0.3675262854018258
+      "required_token_positions_per_second": 28755.906431705265,
+      "required_pages_per_second_upper_estimate": 0.368849892019154
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16596693808591567,
-    "producer_p95_seconds": 0.23222752800029411,
-    "gpu_update_seconds": 0.9649346780809108,
-    "producer_batches_per_second": 6.025296432728864,
-    "gpu_batches_per_second": 1.0363395810261768,
-    "producer_headroom": 5.814017473657287,
+    "producer_batch_seconds": 0.5597988922030908,
+    "producer_p95_seconds": 0.23329176900006132,
+    "gpu_update_seconds": 0.9614720393413255,
+    "producer_batches_per_second": 1.786355803714609,
+    "gpu_batches_per_second": 1.040071847211562,
+    "producer_headroom": 1.7175311575866978,
     "sustainable": true,
-    "buffer_seconds": 60.790884719097384,
+    "buffer_seconds": 60.57273847850351,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459267.0783814179,
-    "upload_bytes_per_second": 4238930.0030075805,
+    "generation_bytes_per_second": 411992.0391175489,
+    "upload_bytes_per_second": 4234686.9373777155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.54802383024123,
+    "estimated_full_shards_per_hour": 88.40390091080523,
     "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": 1.062077575999865,
-      "bytes": 5281467,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 0.9683684410001661,
       "bytes": 6284844,
@@ -562,6 +556,12 @@
       "bytes": 6015964,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0633229350005422,
+      "bytes": 4861789,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -151,3 +151,4 @@
 {"step": 700100, "loss": 0.5273600816726685, "lr": 8e-05, "grad_norm": 0.20572507000430548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102475568774, "tokens_per_second": 145293.57576594577, "data_wait_s": 1.4528000974678434e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700200, "loss": 0.5707261562347412, "lr": 8e-05, "grad_norm": 0.22813854833965583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102490078033, "tokens_per_second": 150402.06672966364, "data_wait_s": 1.755399898684118e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700300, "loss": 0.49784529209136963, "lr": 8e-05, "grad_norm": 0.21931447414439897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102504529645, "tokens_per_second": 146519.1007752696, "data_wait_s": 1.7926000509760343e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 700400, "loss": 0.551393985748291, "lr": 8e-05, "grad_norm": 0.21495483709769558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102519086586, "tokens_per_second": 152066.0889579606, "data_wait_s": 1.5417999748024158e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24eddb5b3d3319466b94040c2b12f03dfa8d5de2f2e9b3ee6d943b691c35d5e0
-size 326568
+oid sha256:2463c3421be517c5cf523e1ec362edcd2c8709262eb3856530ff0e4869956777
+size 328686