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": 75648.53044063006,
-      "required_pages_per_second_upper_estimate": 0.006557022889146287
+      "required_token_positions_per_second": 76147.23347458833,
+      "required_pages_per_second_upper_estimate": 0.0066002492041785064
     },
     {
       "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": 9456.066305078757,
-      "required_pages_per_second_upper_estimate": 0.0037721662298862125
+      "required_token_positions_per_second": 9518.40418432354,
+      "required_pages_per_second_upper_estimate": 0.003797033741951309
     },
     {
       "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": 18912.132610157514,
-      "required_pages_per_second_upper_estimate": 0.004374989037193974
+      "required_token_positions_per_second": 19036.80836864708,
+      "required_pages_per_second_upper_estimate": 0.0044038305797021275
     },
     {
       "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": 28368.19891523627,
-      "required_pages_per_second_upper_estimate": 0.0017274744775008664
+      "required_token_positions_per_second": 28555.21255297062,
+      "required_pages_per_second_upper_estimate": 0.0017388626268541613
     },
     {
       "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": 28368.19891523627,
-      "required_pages_per_second_upper_estimate": 0.0021838624806955137
+      "required_token_positions_per_second": 28555.21255297062,
+      "required_pages_per_second_upper_estimate": 0.002198259308215187
     },
     {
       "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": 28368.19891523627,
-      "required_pages_per_second_upper_estimate": 0.35223808827292136
+      "required_token_positions_per_second": 28555.21255297062,
+      "required_pages_per_second_upper_estimate": 0.3545601717591991
     }
   ],
   "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": 0.34449968139102793,
-    "producer_p95_seconds": 0.3883891019941075,
-    "gpu_update_seconds": 0.9746124553980952,
-    "producer_batches_per_second": 2.9027603043410064,
-    "gpu_batches_per_second": 1.0260488612281637,
-    "producer_headroom": 2.8290663476459104,
+    "producer_batch_seconds": 0.10045728039040114,
+    "producer_p95_seconds": 0.36006088399881264,
+    "gpu_update_seconds": 0.9682295289769698,
+    "producer_batches_per_second": 9.954480114470147,
+    "gpu_batches_per_second": 1.0328129540281619,
+    "producer_headroom": 9.638221592444044,
     "sustainable": true,
-    "buffer_seconds": 60.4259722346819,
+    "buffer_seconds": 60.03023079657213,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 500161.19285656407,
-    "upload_bytes_per_second": 4436752.9953285,
+    "generation_bytes_per_second": 502395.19192042056,
+    "upload_bytes_per_second": 4440030.5298418375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.32294883034412,
+    "estimated_full_shards_per_hour": 107.80231302461112,
     "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": 1.033323986994219,
-      "bytes": 6001182,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 3.9172589509980753,
       "bytes": 5890320,
@@ -562,6 +556,12 @@
       "bytes": 11718816,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.2345870930003002,
+      "bytes": 7207141,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 196600, "loss": 0.5985371414572, "lr": 0.0003, "grad_norm": 0.14992750001195332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28774382558, "tokens_per_second": 150662.8953278723, "data_wait_s": 1.0135001502931118e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196700, "loss": 0.6548292003571987, "lr": 0.0003, "grad_norm": 0.16110462392309002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28789175919, "tokens_per_second": 151406.54491275083, "data_wait_s": 1.012500433716923e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196800, "loss": 0.6592029295861721, "lr": 0.0003, "grad_norm": 0.15857606671365335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28803846116, "tokens_per_second": 150832.67363795283, "data_wait_s": 1.2368996976874769e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196900, "loss": 0.5936013609170914, "lr": 0.0003, "grad_norm": 0.13963440142091707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28818393057, "tokens_per_second": 148291.17903467067, "data_wait_s": 1.1382995580788702e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:689388e8b9ab172ed14c21de6ab4bf8227753032fbd406646208866cb966ca64
-size 891753
+oid sha256:ce1bdbefca7e95f2081116603056cf658ecda64476bdeffec3fd87043e210da6
+size 893871