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": 74338.4573752969,
-      "required_pages_per_second_upper_estimate": 0.006443469076193037
+      "required_token_positions_per_second": 75781.23834021771,
+      "required_pages_per_second_upper_estimate": 0.0065685256735374484
     },
     {
       "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": 9292.307171912113,
-      "required_pages_per_second_upper_estimate": 0.0037068402632488085
+      "required_token_positions_per_second": 9472.654792527213,
+      "required_pages_per_second_upper_estimate": 0.003778783625549391
     },
     {
       "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": 18584.614343824225,
-      "required_pages_per_second_upper_estimate": 0.004299223450402385
+      "required_token_positions_per_second": 18945.309585054427,
+      "required_pages_per_second_upper_estimate": 0.004382663946441559
     },
     {
       "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": 27876.92151573634,
-      "required_pages_per_second_upper_estimate": 0.0016975582614046337
+      "required_token_positions_per_second": 28417.96437758164,
+      "required_pages_per_second_upper_estimate": 0.0017305049330584975
     },
     {
       "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": 27876.92151573634,
-      "required_pages_per_second_upper_estimate": 0.002146042586539127
+      "required_token_positions_per_second": 28417.96437758164,
+      "required_pages_per_second_upper_estimate": 0.002187693563746477
     },
     {
       "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": 27876.92151573634,
-      "required_pages_per_second_upper_estimate": 0.34613806717082013
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28417.96437758164,
+      "required_pages_per_second_upper_estimate": 0.35285600876096257
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.37585391521879785,
-    "producer_p95_seconds": 0.2003837560005195,
-    "gpu_update_seconds": 0.9917881350131464,
-    "producer_batches_per_second": 2.6606081764982137,
-    "gpu_batches_per_second": 1.0082798580633803,
-    "producer_headroom": 2.638759621369892,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.49907623580193,
+    "buffer_seconds": 60.32015443556116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292031.53345234983,
-    "upload_bytes_per_second": 5770384.431542097,
+    "generation_bytes_per_second": 313217.25650602847,
+    "upload_bytes_per_second": 5805391.209229734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.6631689327037,
+    "estimated_full_shards_per_hour": 67.20913192163124,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 6295663,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0289003750003758,
+      "bytes": 6769832,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -72,3 +72,4 @@
 {"step": 92200, "loss": 0.6220071911811829, "lr": 0.0003, "grad_norm": 0.13366662895182727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13492718975, "tokens_per_second": 147595.04661729187, "data_wait_s": 3.900800038536545e-05, "gpu_peak_gb": 85.46991729736328}
 {"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}

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:f89d2b7565a6d8d928f53fb1aee8f2dff19be01823729cc6cf047ee47987b437
-size 157680
+oid sha256:5245ceaf8bea67e45137fa3e8b06190e7a71e0dc1ca6f20dd820af53c057a261
+size 159798