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": 75737.44431673342,
-      "required_pages_per_second_upper_estimate": 0.0065647297185767685
+      "required_token_positions_per_second": 74552.46993637337,
+      "required_pages_per_second_upper_estimate": 0.006462019142577268
     },
     {
       "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": 9467.180539591678,
-      "required_pages_per_second_upper_estimate": 0.003776599864206031
+      "required_token_positions_per_second": 9319.05874204667,
+      "required_pages_per_second_upper_estimate": 0.0037175118645471003
     },
     {
       "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": 18934.361079183356,
-      "required_pages_per_second_upper_estimate": 0.0043801312022954085
+      "required_token_positions_per_second": 18638.11748409334,
+      "required_pages_per_second_upper_estimate": 0.004311600460280537
     },
     {
       "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": 28401.541618775034,
-      "required_pages_per_second_upper_estimate": 0.0017295048732107316
+      "required_token_positions_per_second": 27957.176226140014,
+      "required_pages_per_second_upper_estimate": 0.0017024453522042934
     },
     {
       "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": 28401.541618775034,
-      "required_pages_per_second_upper_estimate": 0.0021864292943124387
+      "required_token_positions_per_second": 27957.176226140014,
+      "required_pages_per_second_upper_estimate": 0.0021522208163052593
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28401.541618775034,
-      "required_pages_per_second_upper_estimate": 0.35265209306002254
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27957.176226140014,
+      "required_pages_per_second_upper_estimate": 0.3471345620787963
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10468656004655941,
-    "producer_p95_seconds": 0.34765762299866765,
-    "gpu_update_seconds": 0.9734682846132232,
-    "producer_batches_per_second": 9.552324573042132,
-    "gpu_batches_per_second": 1.0272548328549997,
-    "producer_headroom": 9.298885016188063,
+    "producer_batch_seconds": 0.44542471855470467,
+    "producer_p95_seconds": 0.333499052998377,
+    "gpu_update_seconds": 0.9889410781818897,
+    "producer_batches_per_second": 2.245048283904759,
+    "gpu_batches_per_second": 1.0111825891977726,
+    "producer_headroom": 2.220220470455174,
     "sustainable": true,
-    "buffer_seconds": 60.35503364601984,
+    "buffer_seconds": 60.32540576909527,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409337.6307784296,
-    "upload_bytes_per_second": 4984805.126951665,
+    "generation_bytes_per_second": 364225.7773524917,
+    "upload_bytes_per_second": 4986447.894074069,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.83432667269389,
+    "estimated_full_shards_per_hour": 78.15437307768882,
     "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": 1.5007521030020143,
-      "bytes": 7605566,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8186183350007923,
       "bytes": 971258,
@@ -562,6 +556,12 @@
       "bytes": 6144689,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3160131810000166,
+      "bytes": 6821427,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -313,3 +313,4 @@
 {"step": 386300, "loss": 0.5637060552835464, "lr": 0.0003, "grad_norm": 0.16392651726264335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56543365074, "tokens_per_second": 148497.18542614067, "data_wait_s": 1.1593001545406878e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 386400, "loss": 0.5703699178993702, "lr": 0.0003, "grad_norm": 0.18797698578785957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56558064033, "tokens_per_second": 149667.19114507549, "data_wait_s": 1.2093001714674756e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 386500, "loss": 0.5747383907437325, "lr": 0.0003, "grad_norm": 0.16812191695093157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56572755382, "tokens_per_second": 149876.05576409269, "data_wait_s": 9.35700154514052e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 386600, "loss": 0.6133938077837229, "lr": 0.0003, "grad_norm": 0.15190493584300963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56587505244, "tokens_per_second": 144101.1346329765, "data_wait_s": 8.521001291228458e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99079ae9d0ee5bd4e4eafa36d81797751aec6d8228f45a8699345fa814df2e70
-size 672033
+oid sha256:e7a9758d078e89f91f9f9dc9e0b9721dc5221d1a75dd17ca50fb12c492656ec2
+size 674151