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": 75781.23834021771,
-      "required_pages_per_second_upper_estimate": 0.0065685256735374484
+      "required_token_positions_per_second": 75673.83612372458,
+      "required_pages_per_second_upper_estimate": 0.0065592163216202535
     },
     {
       "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": 9472.654792527213,
-      "required_pages_per_second_upper_estimate": 0.003778783625549391
+      "required_token_positions_per_second": 9459.229515465573,
+      "required_pages_per_second_upper_estimate": 0.003773428081803723
     },
     {
       "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": 18945.309585054427,
-      "required_pages_per_second_upper_estimate": 0.004382663946441559
+      "required_token_positions_per_second": 18918.459030931146,
+      "required_pages_per_second_upper_estimate": 0.004376452543298751
     },
     {
       "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": 28417.96437758164,
-      "required_pages_per_second_upper_estimate": 0.0017305049330584975
+      "required_token_positions_per_second": 28377.68854639672,
+      "required_pages_per_second_upper_estimate": 0.0017280523462502911
     },
     {
       "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": 28417.96437758164,
-      "required_pages_per_second_upper_estimate": 0.002187693563746477
+      "required_token_positions_per_second": 28377.68854639672,
+      "required_pages_per_second_upper_estimate": 0.0021845930187712258
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28417.96437758164,
-      "required_pages_per_second_upper_estimate": 0.35285600876096257
+      "required_token_positions_per_second": 28377.68854639672,
+      "required_pages_per_second_upper_estimate": 0.3523559177321817
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.057434242812462344,
-    "producer_p95_seconds": 0.18944932900012645,
-    "gpu_update_seconds": 0.9729057167025993,
-    "producer_batches_per_second": 17.411215871083364,
-    "gpu_batches_per_second": 1.0278488273141506,
-    "producer_headroom": 16.93947145572003,
+    "producer_batch_seconds": 0.13377740542965455,
+    "producer_p95_seconds": 0.2133227390004322,
+    "gpu_update_seconds": 0.9742865404557634,
+    "producer_batches_per_second": 7.475103862182762,
+    "gpu_batches_per_second": 1.026392091521872,
+    "producer_headroom": 7.282893081433559,
     "sustainable": true,
-    "buffer_seconds": 60.32015443556116,
+    "buffer_seconds": 60.405765508257325,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313217.25650602847,
-    "upload_bytes_per_second": 5805391.209229734,
+    "generation_bytes_per_second": 321684.99589213706,
+    "upload_bytes_per_second": 5991612.309558541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.20913192163124,
+    "estimated_full_shards_per_hour": 69.02611167500575,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 6769832,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.118036003999805,
+      "bytes": 11128347,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -73,3 +73,4 @@
 {"step": 92300, "loss": 0.6198513470590115, "lr": 0.0003, "grad_norm": 0.13410049734005566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13507492070, "tokens_per_second": 141402.48420216204, "data_wait_s": 6.253554176000762, "gpu_peak_gb": 85.46991729736328}
 {"step": 92400, "loss": 0.7089204117655754, "lr": 0.0003, "grad_norm": 0.14207928107584153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13522378833, "tokens_per_second": 151903.19310876896, "data_wait_s": 1.2139000318711624e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92500, "loss": 0.6128994338214397, "lr": 0.0003, "grad_norm": 0.13664383668330488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13536942201, "tokens_per_second": 148515.6514204053, "data_wait_s": 1.4106999515206553e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 92600, "loss": 0.5336556099355221, "lr": 0.0003, "grad_norm": 0.1350423349771974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13551411504, "tokens_per_second": 147043.00553476688, "data_wait_s": 1.8695998733164743e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5245ceaf8bea67e45137fa3e8b06190e7a71e0dc1ca6f20dd820af53c057a261
-size 159798
+oid sha256:2bdd21edd7686f01adc8492e6b0e202f1cc436c5734749d5543dac926c72e2dc
+size 161916