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": 76275.16198051495,
-      "required_pages_per_second_upper_estimate": 0.006611337722840392
+      "required_token_positions_per_second": 76058.98654815045,
+      "required_pages_per_second_upper_estimate": 0.0065926001842546616
     },
     {
       "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": 9534.39524756437,
-      "required_pages_per_second_upper_estimate": 0.003803412816165777
+      "required_token_positions_per_second": 9507.373318518807,
+      "required_pages_per_second_upper_estimate": 0.003792633364655659
     },
     {
       "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": 19068.79049512874,
-      "required_pages_per_second_upper_estimate": 0.004411229081797457
+      "required_token_positions_per_second": 19014.746637037613,
+      "required_pages_per_second_upper_estimate": 0.004398726986367492
     },
     {
       "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": 28603.185742693106,
-      "required_pages_per_second_upper_estimate": 0.0017417839424123862
+      "required_token_positions_per_second": 28522.11995555642,
+      "required_pages_per_second_upper_estimate": 0.0017368474612950767
     },
     {
       "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": 28603.185742693106,
-      "required_pages_per_second_upper_estimate": 0.0022019524171583123
+      "required_token_positions_per_second": 28522.11995555642,
+      "required_pages_per_second_upper_estimate": 0.002195711748460069
     },
     {
       "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": 28603.185742693106,
-      "required_pages_per_second_upper_estimate": 0.35515583821961466
+      "required_token_positions_per_second": 28522.11995555642,
+      "required_pages_per_second_upper_estimate": 0.35414927245311373
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.17059740275783497,
-    "producer_p95_seconds": 0.3385095449993969,
-    "gpu_update_seconds": 0.9666056168957643,
-    "producer_batches_per_second": 5.861753953074607,
-    "gpu_batches_per_second": 1.0345480954388422,
-    "producer_headroom": 5.666004295902866,
+    "producer_batch_seconds": 0.1823512239845968,
+    "producer_p95_seconds": 0.37393767799949273,
+    "gpu_update_seconds": 0.9693529107612447,
+    "producer_batches_per_second": 5.483922609066062,
+    "gpu_batches_per_second": 1.0316160284851135,
+    "producer_headroom": 5.315856343487587,
     "sustainable": true,
-    "buffer_seconds": 60.89615386443315,
+    "buffer_seconds": 60.099880467197174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297938.15925174695,
-    "upload_bytes_per_second": 5716825.417728054,
+    "generation_bytes_per_second": 298832.5282702654,
+    "upload_bytes_per_second": 5725627.603928612,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.9305933300429,
+    "estimated_full_shards_per_hour": 64.12250410157176,
     "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": 1.0644437480004854,
-      "bytes": 8538245,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.3977851220006414,
       "bytes": 6073631,
@@ -562,6 +556,12 @@
       "bytes": 1116619,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4015686020065914,
+      "bytes": 11114007,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -285,3 +285,4 @@
 {"step": 308700, "loss": 0.5932493638247252, "lr": 0.0003, "grad_norm": 0.1425984820295111, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45185539764, "tokens_per_second": 152588.452815944, "data_wait_s": 1.1445001291576773e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 308800, "loss": 0.6029892917722464, "lr": 0.0003, "grad_norm": 0.1617831479673141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45200249080, "tokens_per_second": 150909.29004143534, "data_wait_s": 1.0438001481816173e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 308900, "loss": 0.6228929907083511, "lr": 0.0003, "grad_norm": 0.16690376191349374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45214920191, "tokens_per_second": 150615.64277173474, "data_wait_s": 1.103100657928735e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 309000, "loss": 0.601045886054635, "lr": 0.0003, "grad_norm": 0.1612058812513313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45229451852, "tokens_per_second": 149826.7964951862, "data_wait_s": 9.901996236294508e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6279744a68ee5321deafddfd5976d250a7c4c68c2943fe64d479cd579061608
-size 611946
+oid sha256:46bc6f57f8edfeba5a8dd4d453a90ff0021593f97b20a55ea747fcc2224d8f3b
+size 614064