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": 75087.78338724484,
-      "required_pages_per_second_upper_estimate": 0.0065084187557592705
+      "required_token_positions_per_second": 76099.32439352323,
+      "required_pages_per_second_upper_estimate": 0.00659609656645624
     },
     {
       "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": 9385.972923405605,
-      "required_pages_per_second_upper_estimate": 0.0037442049319473454
+      "required_token_positions_per_second": 9512.415549190404,
+      "required_pages_per_second_upper_estimate": 0.0037946447858586262
     },
     {
       "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": 18771.94584681121,
-      "required_pages_per_second_upper_estimate": 0.0043425592966965755
+      "required_token_positions_per_second": 19024.831098380808,
+      "required_pages_per_second_upper_estimate": 0.004401059849018782
     },
     {
       "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": 28157.918770216816,
-      "required_pages_per_second_upper_estimate": 0.0017146695199238386
+      "required_token_positions_per_second": 28537.24664757121,
+      "required_pages_per_second_upper_estimate": 0.0017377685974751579
     },
     {
       "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": 28157.918770216816,
-      "required_pages_per_second_upper_estimate": 0.002167674533039219
+      "required_token_positions_per_second": 28537.24664757121,
+      "required_pages_per_second_upper_estimate": 0.002196876243084731
     },
     {
       "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": 28157.918770216816,
-      "required_pages_per_second_upper_estimate": 0.34962711263415347
+      "required_token_positions_per_second": 28537.24664757121,
+      "required_pages_per_second_upper_estimate": 0.3543370953421559
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.11164432211734265,
-    "producer_p95_seconds": 0.23870455399992352,
-    "gpu_update_seconds": 0.9818907507199656,
-    "producer_batches_per_second": 8.957016183491714,
-    "gpu_batches_per_second": 1.0184432425570318,
-    "producer_headroom": 8.794811344619559,
+    "producer_batch_seconds": 0.07655757511720651,
+    "producer_p95_seconds": 0.18906632299876946,
+    "gpu_update_seconds": 0.9688390874370884,
+    "producer_batches_per_second": 13.062064707104959,
+    "gpu_batches_per_second": 1.0321631455284725,
+    "producer_headroom": 12.655038850875767,
     "sustainable": true,
-    "buffer_seconds": 60.877226544637864,
+    "buffer_seconds": 60.068023421099475,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329406.045372402,
-    "upload_bytes_per_second": 5594572.579222196,
+    "generation_bytes_per_second": 316130.58627826837,
+    "upload_bytes_per_second": 5592055.152016832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.68286915663762,
+    "estimated_full_shards_per_hour": 67.83426467190779,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 12703928,
       "pages": 49,
       "shards": 4
+    },
+    {
+      "seconds": 1.1181347120000282,
+      "bytes": 6085437,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -68,3 +68,4 @@
 {"step": 287000, "loss": 0.5955741610378027, "lr": 0.0003, "grad_norm": 0.15505631039645046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42008581843, "tokens_per_second": 150827.1468653094, "data_wait_s": 1.0313000530004501e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287100, "loss": 0.5854803398251534, "lr": 0.0003, "grad_norm": 0.14064032562833925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42023205706, "tokens_per_second": 149778.58866883282, "data_wait_s": 1.8060000002151355e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287200, "loss": 0.619665365666151, "lr": 0.0003, "grad_norm": 0.17011384335844504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42037862083, "tokens_per_second": 150396.25181598187, "data_wait_s": 1.2086999049643055e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 287300, "loss": 0.6012470815330744, "lr": 0.0003, "grad_norm": 0.13943662973821655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42052528223, "tokens_per_second": 151016.36593036633, "data_wait_s": 1.1636999261099845e-05, "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:cd837f1453428b0f3d4fc3b80dcbfb0594f4bc8bfb15ae45477e2264b3358b12
-size 149208
+oid sha256:312a9f40db7fe33c887e14f7249df8806a5e7213eb42bd8034998c76db839a94
+size 151326