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": 75459.5416336954,
-      "required_pages_per_second_upper_estimate": 0.006540641818348943
+      "required_token_positions_per_second": 75435.13873466507,
+      "required_pages_per_second_upper_estimate": 0.0065385266369095765
     },
     {
       "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": 9432.442704211924,
-      "required_pages_per_second_upper_estimate": 0.003762742422296124
+      "required_token_positions_per_second": 9429.392341833134,
+      "required_pages_per_second_upper_estimate": 0.003761525587136243
     },
     {
       "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": 18864.88540842385,
-      "required_pages_per_second_upper_estimate": 0.004364059228595241
+      "required_token_positions_per_second": 18858.78468366627,
+      "required_pages_per_second_upper_estimate": 0.0043626479332916634
     },
     {
       "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": 28297.32811263577,
-      "required_pages_per_second_upper_estimate": 0.0017231588174528606
+      "required_token_positions_per_second": 28288.1770254994,
+      "required_pages_per_second_upper_estimate": 0.0017226015642582994
     },
     {
       "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": 28297.32811263577,
-      "required_pages_per_second_upper_estimate": 0.0021784066501283894
+      "required_token_positions_per_second": 28288.1770254994,
+      "required_pages_per_second_upper_estimate": 0.0021777021741088013
     },
     {
       "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": 28297.32811263577,
-      "required_pages_per_second_upper_estimate": 0.3513581100939415
+      "required_token_positions_per_second": 28288.1770254994,
+      "required_pages_per_second_upper_estimate": 0.3512444842184263
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09865101757813477,
-    "producer_p95_seconds": 0.3212924279978324,
-    "gpu_update_seconds": 0.9770533772640597,
-    "producer_batches_per_second": 10.136742879595417,
-    "gpu_batches_per_second": 1.0234855364813285,
-    "producer_headroom": 9.904138864966113,
+    "producer_batch_seconds": 0.12283731911728069,
+    "producer_p95_seconds": 0.2907117279974045,
+    "gpu_update_seconds": 0.9773694492606457,
+    "producer_batches_per_second": 8.140848458645012,
+    "gpu_batches_per_second": 1.0231545509801576,
+    "producer_headroom": 7.956616574540252,
     "sustainable": true,
-    "buffer_seconds": 60.5773093903717,
+    "buffer_seconds": 60.596905854160035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526154.4247739561,
-    "upload_bytes_per_second": 3627485.5486378567,
+    "generation_bytes_per_second": 523736.34410499054,
+    "upload_bytes_per_second": 3575845.4185311226,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.9004913083459,
+    "estimated_full_shards_per_hour": 112.38162748682296,
     "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": 1.207310141000562,
-      "bytes": 13710377,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.9586756399985461,
       "bytes": 1163538,
@@ -562,6 +556,12 @@
       "bytes": 1242501,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.202817702996981,
+      "bytes": 7607811,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 217900, "loss": 0.6120729297399521, "lr": 0.0003, "grad_norm": 0.1569615759045584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31893514828, "tokens_per_second": 149033.36150266792, "data_wait_s": 1.341800088994205e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218000, "loss": 0.606307540088892, "lr": 0.0003, "grad_norm": 0.14625266571941004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31908214211, "tokens_per_second": 148565.29103573423, "data_wait_s": 1.2604999938048422e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218100, "loss": 0.6077547408640385, "lr": 0.0003, "grad_norm": 0.1613877905739051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31922814983, "tokens_per_second": 147908.95578827965, "data_wait_s": 1.1562999134184793e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 218200, "loss": 0.5977561417967081, "lr": 0.0003, "grad_norm": 0.15101460923973592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31937438736, "tokens_per_second": 147734.543121157, "data_wait_s": 2.2462998458649963e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a5aaa71072fcb24b20c56b365ef8c37d01acc073971f78594ccf16fad9dd914f
-size 492555
+oid sha256:d23de8a379142f29b79c0c5869caaadf0f2c78deaf0a1a472fdfd995f242f854
+size 494673