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": 75434.92562306488,
-      "required_pages_per_second_upper_estimate": 0.006538508164936198
+      "required_token_positions_per_second": 75708.4334147349,
+      "required_pages_per_second_upper_estimate": 0.006562215127118996
     },
     {
       "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": 9429.36570288311,
-      "required_pages_per_second_upper_estimate": 0.0037615149604607904
+      "required_token_positions_per_second": 9463.554176841862,
+      "required_pages_per_second_upper_estimate": 0.0037751532538861745
     },
     {
       "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": 18858.73140576622,
-      "required_pages_per_second_upper_estimate": 0.0043626356083871455
+      "required_token_positions_per_second": 18927.108353683725,
+      "required_pages_per_second_upper_estimate": 0.004378453411894678
     },
     {
       "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": 28288.097108649326,
-      "required_pages_per_second_upper_estimate": 0.0017225966977414213
+      "required_token_positions_per_second": 28390.662530525587,
+      "required_pages_per_second_upper_estimate": 0.001728842393814505
     },
     {
       "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": 28288.097108649326,
-      "required_pages_per_second_upper_estimate": 0.0021776960218884587
+      "required_token_positions_per_second": 28390.662530525587,
+      "required_pages_per_second_upper_estimate": 0.0021855917919837526
     },
     {
       "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": 28288.097108649326,
-      "required_pages_per_second_upper_estimate": 0.35124349191861287
+      "required_token_positions_per_second": 28390.662530525587,
+      "required_pages_per_second_upper_estimate": 0.3525170111939306
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10696796451537693,
+    "producer_batch_seconds": 0.13472059385958346,
     "producer_p95_seconds": 0.35885942399909254,
-    "gpu_update_seconds": 0.9773722104321534,
-    "producer_batches_per_second": 9.348593333813016,
-    "gpu_batches_per_second": 1.0231516604690873,
-    "producer_headroom": 9.137055331100122,
+    "gpu_update_seconds": 0.9738413103347948,
+    "producer_batches_per_second": 7.4227701300239195,
+    "gpu_batches_per_second": 1.0268613473135701,
+    "producer_headroom": 7.228600189736469,
     "sustainable": true,
-    "buffer_seconds": 60.59707704679351,
+    "buffer_seconds": 60.37816124075728,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262317.8338768617,
-    "upload_bytes_per_second": 5764440.892628973,
+    "generation_bytes_per_second": 271348.86481280223,
+    "upload_bytes_per_second": 5722400.73628493,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.28730070332898,
+    "estimated_full_shards_per_hour": 58.22514971054125,
     "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": 1.4423049569995783,
-      "bytes": 10949179,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.3191998570000578,
       "bytes": 13241805,
@@ -562,6 +556,12 @@
       "bytes": 6743722,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.11581381400174,
+      "bytes": 5950902,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -268,3 +268,4 @@
 {"step": 307000, "loss": 0.6174443885684013, "lr": 0.0003, "grad_norm": 0.23231451502623285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44936485048, "tokens_per_second": 141820.6529797209, "data_wait_s": 1.0821997420862317e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 307100, "loss": 0.5371118113398552, "lr": 0.0003, "grad_norm": 0.19132379531914268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44951135665, "tokens_per_second": 131276.77808792147, "data_wait_s": 1.3419004972092807e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 307200, "loss": 0.6113442908972502, "lr": 0.0003, "grad_norm": 0.14900561915782579, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44965784277, "tokens_per_second": 150229.627348448, "data_wait_s": 1.150499883806333e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 307300, "loss": 0.5767037272453308, "lr": 0.0003, "grad_norm": 0.14764916325513514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44980404567, "tokens_per_second": 149316.40687751726, "data_wait_s": 9.604998922441155e-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:f76099a9f333aa078f32400f8c2d5ee81fc43cccac5a437eecf8abf7039cb4a4
-size 575940
+oid sha256:6ae27a2df69db284df1426512885e71703ad8682d9524fb20b56039932368664
+size 578058