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": 74792.2489180646,
-      "required_pages_per_second_upper_estimate": 0.006482802576995991
+      "required_token_positions_per_second": 74183.00515407824,
+      "required_pages_per_second_upper_estimate": 0.006429994871647857
     },
     {
       "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": 9349.031114758074,
-      "required_pages_per_second_upper_estimate": 0.0037294682921485854
+      "required_token_positions_per_second": 9272.87564425978,
+      "required_pages_per_second_upper_estimate": 0.0036990887363410646
     },
     {
       "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": 18698.06222951615,
-      "required_pages_per_second_upper_estimate": 0.0043254676221426164
+      "required_token_positions_per_second": 18545.75128851956,
+      "required_pages_per_second_upper_estimate": 0.004290233166524065
     },
     {
       "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": 28047.09334427422,
-      "required_pages_per_second_upper_estimate": 0.0017079208329400073
+      "required_token_positions_per_second": 27818.62693277934,
+      "required_pages_per_second_upper_estimate": 0.0016940084271505921
     },
     {
       "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": 28047.09334427422,
-      "required_pages_per_second_upper_estimate": 0.002159142884965744
+      "required_token_positions_per_second": 27818.62693277934,
+      "required_pages_per_second_upper_estimate": 0.0021415549081733637
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28047.09334427422,
-      "required_pages_per_second_upper_estimate": 0.34825103175278715
+      "required_token_positions_per_second": 27818.62693277934,
+      "required_pages_per_second_upper_estimate": 0.34541424354991296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2707074414999582,
-    "producer_p95_seconds": 0.1927194209993104,
-    "gpu_update_seconds": 0.9857705987791532,
-    "producer_batches_per_second": 3.6940247909666515,
-    "gpu_batches_per_second": 1.0144347997784369,
-    "producer_headroom": 3.6414610300962322,
+    "producer_batch_seconds": 0.35113921403119974,
+    "producer_p95_seconds": 0.20364820800023153,
+    "gpu_update_seconds": 0.9938664502316508,
+    "producer_batches_per_second": 2.8478733221495083,
+    "gpu_batches_per_second": 1.006171402371938,
+    "producer_headroom": 2.83040574939415,
     "sustainable": true,
-    "buffer_seconds": 60.13200652552835,
+    "buffer_seconds": 60.6258534641307,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 410360.88458604633,
-    "upload_bytes_per_second": 5888719.806983413,
+    "generation_bytes_per_second": 374390.1606640173,
+    "upload_bytes_per_second": 5923168.641723051,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.0538931196789,
+    "estimated_full_shards_per_hour": 80.33541312160864,
     "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": [
@@ -514,6 +514,12 @@
       "bytes": 6393614,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0277372999998988,
+      "bytes": 8122954,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 88400, "loss": 0.6176491789519787, "lr": 0.0003, "grad_norm": 0.12655935267375543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12937137368, "tokens_per_second": 149196.44004689562, "data_wait_s": 1.203600004373584e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 88500, "loss": 0.6558070499449968, "lr": 0.0003, "grad_norm": 0.1680909194797863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12951962743, "tokens_per_second": 150960.81700967898, "data_wait_s": 7.649000508536119e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 88600, "loss": 0.6345845386385918, "lr": 0.0003, "grad_norm": 0.13977763920575317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12966471919, "tokens_per_second": 144134.3117616288, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 88700, "loss": 0.6238090228289366, "lr": 0.0003, "grad_norm": 0.13314721013217592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12980961945, "tokens_per_second": 147591.8424126812, "data_wait_s": 1.3113000022713095e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6509e28340b3d9c139d237922831a6464f10d7593f97bcee23eca07c65e5cd16
-size 142854
+oid sha256:bf8a9e8503cbdc747cf4f4da9268e6b6c128cd7daac4ef5f29443895d515f0ef
+size 144972