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": 74724.06776629508,
-      "required_pages_per_second_upper_estimate": 0.0064768928075641585
+      "required_token_positions_per_second": 74778.5312429757,
+      "required_pages_per_second_upper_estimate": 0.006481613563686415
     },
     {
       "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": 9340.508470786885,
-      "required_pages_per_second_upper_estimate": 0.003726068482043595
+      "required_token_positions_per_second": 9347.316405371963,
+      "required_pages_per_second_upper_estimate": 0.003728784268937276
     },
     {
       "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": 18681.01694157377,
-      "required_pages_per_second_upper_estimate": 0.004321524494763969
+      "required_token_positions_per_second": 18694.632810743926,
+      "required_pages_per_second_upper_estimate": 0.004324674286465372
     },
     {
       "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": 28021.525412360654,
-      "required_pages_per_second_upper_estimate": 0.0017063638800310426
+      "required_token_positions_per_second": 28041.94921611589,
+      "required_pages_per_second_upper_estimate": 0.0017076075825243195
     },
     {
       "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": 28021.525412360654,
-      "required_pages_per_second_upper_estimate": 0.0021571745947904725
+      "required_token_positions_per_second": 28041.94921611589,
+      "required_pages_per_second_upper_estimate": 0.0021587468757402577
     },
     {
       "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": 28021.525412360654,
-      "required_pages_per_second_upper_estimate": 0.3479335636087842
+      "required_token_positions_per_second": 28041.94921611589,
+      "required_pages_per_second_upper_estimate": 0.34818715889735014
     }
   ],
   "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": 0.11981086477328518,
-    "producer_p95_seconds": 0.39368333099992014,
-    "gpu_update_seconds": 0.9866700542934794,
-    "producer_batches_per_second": 8.34648845822349,
-    "gpu_batches_per_second": 1.0135100337225351,
-    "producer_headroom": 8.23523022023527,
+    "producer_batch_seconds": 0.407925182328313,
+    "producer_p95_seconds": 0.4177976270002546,
+    "gpu_update_seconds": 0.985951432509924,
+    "producer_batches_per_second": 2.4514299271555235,
+    "gpu_batches_per_second": 1.0142487419023396,
+    "producer_headroom": 2.4169908483766873,
     "sustainable": true,
-    "buffer_seconds": 60.18687331190225,
+    "buffer_seconds": 60.143037383105366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378859.0984405867,
-    "upload_bytes_per_second": 5726434.165336837,
+    "generation_bytes_per_second": 339903.83038972184,
+    "upload_bytes_per_second": 5677781.539069274,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.29434313691331,
+    "estimated_full_shards_per_hour": 72.93544944542639,
     "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": 1.346531524999591,
-      "bytes": 7336381,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.4059662229992682,
       "bytes": 6256203,
@@ -562,6 +556,12 @@
       "bytes": 5837048,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8334624849958345,
+      "bytes": 871180,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -374,3 +374,4 @@
 {"step": 232400, "loss": 0.6167115028947592, "lr": 0.0003, "grad_norm": 0.1588070864344922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34016744106, "tokens_per_second": 148924.43527422822, "data_wait_s": 1.2810996850021183e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232500, "loss": 0.5852804183959961, "lr": 0.0003, "grad_norm": 0.15621871718730998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34031515048, "tokens_per_second": 150568.59915837742, "data_wait_s": 9.556999430060387e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 232600, "loss": 0.5765535496175289, "lr": 0.0003, "grad_norm": 0.16614487692741248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34046224255, "tokens_per_second": 137174.81873389723, "data_wait_s": 8.55699909152463e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 232700, "loss": 0.5782700162380934, "lr": 0.0003, "grad_norm": 0.15028562898498768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34060838035, "tokens_per_second": 147250.36277060804, "data_wait_s": 2.0628998754546046e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6961baa2409fb27150d10e39fda930831e343231cf95bedd86caf0b79abbf2f
-size 802014
+oid sha256:3d379a825dc3e2acd864b7181b777a22cbf6a0a084bec8b8e29ca622bd1b6ffc
+size 804132