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": 73979.77825498128,
-      "required_pages_per_second_upper_estimate": 0.006412379679107999
+      "required_token_positions_per_second": 74314.32767331363,
+      "required_pages_per_second_upper_estimate": 0.006441377574781298
     },
     {
       "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": 9247.47228187266,
-      "required_pages_per_second_upper_estimate": 0.003688954955270728
+      "required_token_positions_per_second": 9289.290959164204,
+      "required_pages_per_second_upper_estimate": 0.0037056370508872683
     },
     {
       "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": 18494.94456374532,
-      "required_pages_per_second_upper_estimate": 0.00427847992491536
+      "required_token_positions_per_second": 18578.581918328407,
+      "required_pages_per_second_upper_estimate": 0.004297827954930991
     },
     {
       "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": 27742.41684561798,
-      "required_pages_per_second_upper_estimate": 0.0016893676326858909
+      "required_token_positions_per_second": 27867.87287749261,
+      "required_pages_per_second_upper_estimate": 0.0016970072468101263
     },
     {
       "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": 27742.41684561798,
-      "required_pages_per_second_upper_estimate": 0.0021356880447006587
+      "required_token_positions_per_second": 27867.87287749261,
+      "required_pages_per_second_upper_estimate": 0.002145345997318891
     },
     {
       "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": 27742.41684561798,
-      "required_pages_per_second_upper_estimate": 0.34446796932612317
+      "required_token_positions_per_second": 27867.87287749261,
+      "required_pages_per_second_upper_estimate": 0.34602571336767707
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.35496164527336305,
-    "producer_p95_seconds": 0.29834150800161297,
-    "gpu_update_seconds": 0.9965966611292955,
-    "producer_batches_per_second": 2.817205783542839,
-    "gpu_batches_per_second": 1.0034149611406966,
-    "producer_headroom": 2.807617877592934,
+    "producer_batch_seconds": 0.11298212166420285,
+    "producer_p95_seconds": 0.3022184830006154,
+    "gpu_update_seconds": 0.992110166482416,
+    "producer_batches_per_second": 8.850957879620339,
+    "gpu_batches_per_second": 1.0079525780343104,
+    "producer_headroom": 8.781125295478986,
     "sustainable": true,
-    "buffer_seconds": 60.79239632888702,
+    "buffer_seconds": 60.518720155427374,
     "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": 421807.3565997709,
-    "upload_bytes_per_second": 5924549.875533763,
+    "generation_bytes_per_second": 403951.74740113114,
+    "upload_bytes_per_second": 6044848.7654904965,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.51003955359312,
+    "estimated_full_shards_per_hour": 86.67864147687388,
     "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": 0.9827529990016046,
-      "bytes": 788256,
-      "pages": 22,
-      "shards": 1
-    },
     {
       "seconds": 1.1497582369993324,
       "bytes": 10983620,
@@ -562,6 +556,12 @@
       "bytes": 7614945,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0503759480016015,
+      "bytes": 9758074,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 98700, "loss": 0.6326380763202906, "lr": 0.0003, "grad_norm": 0.13642060506661843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14442954752, "tokens_per_second": 148356.8771632658, "data_wait_s": 1.5718000213382766e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98800, "loss": 0.5924640372395515, "lr": 0.0003, "grad_norm": 0.12789388008356628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14457637360, "tokens_per_second": 146134.87580917391, "data_wait_s": 1.3787001080345362e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98900, "loss": 0.6071541821584105, "lr": 0.0003, "grad_norm": 0.13787328096811033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14472080736, "tokens_per_second": 146747.2228629315, "data_wait_s": 1.5404002624563873e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 99000, "loss": 0.5780111197382212, "lr": 0.0003, "grad_norm": 0.13925460440210524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14486645414, "tokens_per_second": 148265.99686362976, "data_wait_s": 1.2526001228252426e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e9a05252c89966d5941ecbc7109d624e9115711a3fec446c7f9e722b551e51e
-size 296133
+oid sha256:3da3a0db060d494786f6b2d5c34de583cb421002f2e6afe1142d12aadfc9f2cc
+size 298251