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": 74640.87208849566,
-      "required_pages_per_second_upper_estimate": 0.006469681617070022
+      "required_token_positions_per_second": 74718.25356888806,
+      "required_pages_per_second_upper_estimate": 0.006476388847668867
     },
     {
       "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": 9330.109011061957,
-      "required_pages_per_second_upper_estimate": 0.003721919982073543
+      "required_token_positions_per_second": 9339.781696111007,
+      "required_pages_per_second_upper_estimate": 0.0037257785607591382
     },
     {
       "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": 18660.218022123914,
-      "required_pages_per_second_upper_estimate": 0.004316713030797725
+      "required_token_positions_per_second": 18679.563392222015,
+      "required_pages_per_second_upper_estimate": 0.004321188241703035
     },
     {
       "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": 27990.32703318587,
-      "required_pages_per_second_upper_estimate": 0.0017044640624245436
+      "required_token_positions_per_second": 28019.345088333022,
+      "required_pages_per_second_upper_estimate": 0.001706231109737036
     },
     {
       "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": 27990.32703318587,
-      "required_pages_per_second_upper_estimate": 0.0021547728571989565
+      "required_token_positions_per_second": 28019.345088333022,
+      "required_pages_per_second_upper_estimate": 0.002157006747411302
     },
     {
       "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": 27990.32703318587,
-      "required_pages_per_second_upper_estimate": 0.3475461841536917
+      "required_token_positions_per_second": 28019.345088333022,
+      "required_pages_per_second_upper_estimate": 0.34790649128143614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11815580010130589,
-    "producer_p95_seconds": 0.3783232269997825,
-    "gpu_update_seconds": 0.9877698094495287,
-    "producer_batches_per_second": 8.463401704720441,
-    "gpu_batches_per_second": 1.0123816201239102,
-    "producer_headroom": 8.359892689166527,
+    "producer_batch_seconds": 0.12060556768727793,
+    "producer_p95_seconds": 0.3898110660011298,
+    "gpu_update_seconds": 0.9867468319776094,
+    "producer_batches_per_second": 8.291491173881228,
+    "gpu_batches_per_second": 1.0134311736231563,
+    "producer_headroom": 8.181602648197611,
     "sustainable": true,
-    "buffer_seconds": 60.25395837642125,
+    "buffer_seconds": 60.19155675063417,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302671.5015518056,
-    "upload_bytes_per_second": 4080763.6462986646,
+    "generation_bytes_per_second": 313859.1557318924,
+    "upload_bytes_per_second": 4032437.855870381,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.94625840106607,
+    "estimated_full_shards_per_hour": 67.34686855285243,
     "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": 0.8605145479996281,
-      "bytes": 5751347,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.0857413720004843,
       "bytes": 7211440,
@@ -562,6 +556,12 @@
       "bytes": 8199005,
       "pages": 47,
       "shards": 3
+    },
+    {
+      "seconds": 0.9340340310009196,
+      "bytes": 958542,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 349200, "loss": 0.5608677826821804, "lr": 0.0003, "grad_norm": 0.1469427349181762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51113978707, "tokens_per_second": 149871.6101259567, "data_wait_s": 1.3037999451626092e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349300, "loss": 0.583791222423315, "lr": 0.0003, "grad_norm": 0.15534844741331277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51128551151, "tokens_per_second": 147342.36753192105, "data_wait_s": 1.9553001038730145e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349400, "loss": 0.6577212065458298, "lr": 0.0003, "grad_norm": 0.21542801983961626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51143129490, "tokens_per_second": 148095.95266899132, "data_wait_s": 1.410899858456105e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 349500, "loss": 0.6470698090270162, "lr": 0.0003, "grad_norm": 0.16589583725746204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51157762813, "tokens_per_second": 148381.10803881442, "data_wait_s": 1.894300658022985e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72b1bba7147283aee166500def8339a6cd32b01da85863d6967204b72efb187c
-size 733455
+oid sha256:06840578ad5a910479aa6db47337643a50ddf588ffe1028e9a0bdfd4bfd99741
+size 735573