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": 74374.54918931957,
-      "required_pages_per_second_upper_estimate": 0.006446597423158656
+      "required_token_positions_per_second": 74748.99226094646,
+      "required_pages_per_second_upper_estimate": 0.006479053199589985
     },
     {
       "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": 9296.818648664947,
-      "required_pages_per_second_upper_estimate": 0.003708639958778102
+      "required_token_positions_per_second": 9343.624032618307,
+      "required_pages_per_second_upper_estimate": 0.00372731132623995
     },
     {
       "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": 18593.637297329893,
-      "required_pages_per_second_upper_estimate": 0.004301310751897364
+      "required_token_positions_per_second": 18687.248065236614,
+      "required_pages_per_second_upper_estimate": 0.004322965955559337
     },
     {
       "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": 27890.455945994843,
-      "required_pages_per_second_upper_estimate": 0.0016983824371977088
+      "required_token_positions_per_second": 28030.87209785492,
+      "required_pages_per_second_upper_estimate": 0.0017069330441393746
     },
     {
       "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": 27890.455945994843,
-      "required_pages_per_second_upper_estimate": 0.0021470845044461277
+      "required_token_positions_per_second": 28030.87209785492,
+      "required_pages_per_second_upper_estimate": 0.0021578941285131025
     },
     {
       "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": 27890.455945994843,
-      "required_pages_per_second_upper_estimate": 0.34630611949780654
+      "required_token_positions_per_second": 28030.87209785492,
+      "required_pages_per_second_upper_estimate": 0.34804961816127894
     }
   ],
   "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": 0.12687348643729024,
-    "producer_p95_seconds": 0.3254512290004641,
-    "gpu_update_seconds": 0.9913068489642365,
-    "producer_batches_per_second": 7.88186742621178,
-    "gpu_batches_per_second": 1.0087693846207626,
-    "producer_headroom": 7.813349162231856,
+    "producer_batch_seconds": 0.0814920806639634,
+    "producer_p95_seconds": 0.33757735499966657,
+    "gpu_update_seconds": 0.9863410565137494,
+    "producer_batches_per_second": 12.271131033254004,
+    "gpu_batches_per_second": 1.0138480938170906,
+    "producer_headroom": 12.103520347958412,
     "sustainable": true,
-    "buffer_seconds": 60.46971778681843,
+    "buffer_seconds": 60.16680444733871,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394566.2724391668,
-    "upload_bytes_per_second": 4154281.72718529,
+    "generation_bytes_per_second": 406285.58623943786,
+    "upload_bytes_per_second": 4171605.209879654,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.66473703271153,
+    "estimated_full_shards_per_hour": 87.1794289625869,
     "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": 0.8924994240005617,
-      "bytes": 5699054,
-      "pages": 19,
-      "shards": 2
-    },
     {
       "seconds": 1.2686124620013288,
       "bytes": 11642195,
@@ -562,6 +556,12 @@
       "bytes": 6689673,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 0.9397628990009252,
+      "bytes": 7665294,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -223,3 +223,4 @@
 {"step": 107300, "loss": 0.6110148895531893, "lr": 0.0003, "grad_norm": 0.12673179723575592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15701488159, "tokens_per_second": 149519.10953555114, "data_wait_s": 2.0088002202101052e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107400, "loss": 0.6278372518718243, "lr": 0.0003, "grad_norm": 0.134091514394105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15716053193, "tokens_per_second": 147333.55900004302, "data_wait_s": 2.595000114524737e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107500, "loss": 0.620902725495398, "lr": 0.0003, "grad_norm": 0.12747942690273625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15730833505, "tokens_per_second": 150345.47894324994, "data_wait_s": 1.1673997505567968e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 107600, "loss": 0.6109723132103682, "lr": 0.0003, "grad_norm": 0.1507481802760825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15745501220, "tokens_per_second": 149072.38841500712, "data_wait_s": 1.1003001418430358e-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 sha256:89f1d77082916ca1a0ba46392efb1e33aabc078c56133a434a9a22d9d27f0917
-size 479847
+oid sha256:c9a979632dd5a365981ffcb34b3d69db526e604b21fa7f6b2891419b3a37fc4a
+size 481965