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": 75074.42347959046,
-      "required_pages_per_second_upper_estimate": 0.006507260752824165
+      "required_token_positions_per_second": 75317.80791592119,
+      "required_pages_per_second_upper_estimate": 0.006528356698913098
     },
     {
       "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": 9384.302934948808,
-      "required_pages_per_second_upper_estimate": 0.0037435387485833765
+      "required_token_positions_per_second": 9414.725989490149,
+      "required_pages_per_second_upper_estimate": 0.0037556749599051177
     },
     {
       "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": 18768.605869897616,
-      "required_pages_per_second_upper_estimate": 0.004341786651819201
+      "required_token_positions_per_second": 18829.451978980298,
+      "required_pages_per_second_upper_estimate": 0.004355862328300535
     },
     {
       "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": 28152.908804846425,
-      "required_pages_per_second_upper_estimate": 0.0017143644393180387
+      "required_token_positions_per_second": 28244.17796847045,
+      "required_pages_per_second_upper_estimate": 0.001719922252530448
     },
     {
       "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": 28152.908804846425,
-      "required_pages_per_second_upper_estimate": 0.002167288852036535
+      "required_token_positions_per_second": 28244.17796847045,
+      "required_pages_per_second_upper_estimate": 0.0021743150049015275
     },
     {
       "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": 28152.908804846425,
-      "required_pages_per_second_upper_estimate": 0.3495649056315287
+      "required_token_positions_per_second": 28244.17796847045,
+      "required_pages_per_second_upper_estimate": 0.35069816318549796
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.19555226523439728,
-    "producer_p95_seconds": 0.19411439200030145,
-    "gpu_update_seconds": 0.9820654835936702,
-    "producer_batches_per_second": 5.113722404602971,
-    "gpu_batches_per_second": 1.0182620372123272,
-    "producer_headroom": 5.022010266240202,
+    "producer_batch_seconds": 0.08166430721096418,
+    "producer_p95_seconds": 0.19605706099991949,
+    "gpu_update_seconds": 0.9788920049598904,
+    "producer_batches_per_second": 12.245251740355679,
+    "gpu_batches_per_second": 1.021563149901275,
+    "producer_headroom": 11.986779027355357,
     "sustainable": true,
-    "buffer_seconds": 60.888059982807555,
+    "buffer_seconds": 60.69130430751321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 510861.4419121814,
-    "upload_bytes_per_second": 5681493.436906536,
+    "generation_bytes_per_second": 509183.7536052412,
+    "upload_bytes_per_second": 5562314.262016217,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.61897318862992,
+    "estimated_full_shards_per_hour": 109.25898033254553,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 10837386,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.9611707320000278,
+      "bytes": 976408,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -63,3 +63,4 @@
 {"step": 361300, "loss": 0.5114772878587246, "lr": 0.0003, "grad_norm": 0.18071435522045554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52884222573, "tokens_per_second": 147260.77314922726, "data_wait_s": 1.0229000508843455e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361400, "loss": 0.5893243551254272, "lr": 0.0003, "grad_norm": 0.16073133032472547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52898796627, "tokens_per_second": 148259.68403419337, "data_wait_s": 1.431600048817927e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361500, "loss": 0.5872211307287216, "lr": 0.0003, "grad_norm": 0.17403485592445106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52913454302, "tokens_per_second": 149950.54574341545, "data_wait_s": 1.1433000508986879e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 361600, "loss": 0.6104452330619097, "lr": 0.0003, "grad_norm": 0.18630582541533516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52928217372, "tokens_per_second": 150174.26182302722, "data_wait_s": 1.39579997266992e-05, "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 sha256:3b30a83a543cf23f585092021a28f07fcd3c9aaf7f17962a11e2b2ef8ee9c89a
-size 138618
+oid sha256:9290454ce7a8dfd896a44acd7bf6ef49e2b83eb47cf02f6c3586f298330cd642
+size 140736