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": 74326.08333929114,
-      "required_pages_per_second_upper_estimate": 0.006442396526113766
+      "required_token_positions_per_second": 74708.61055118714,
+      "required_pages_per_second_upper_estimate": 0.006475553015334556
     },
     {
       "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": 9290.760417411393,
-      "required_pages_per_second_upper_estimate": 0.003706223239752431
+      "required_token_positions_per_second": 9338.576318898393,
+      "required_pages_per_second_upper_estimate": 0.003725297717767031
     },
     {
       "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": 18581.520834822786,
-      "required_pages_per_second_upper_estimate": 0.004298507821538801
+      "required_token_positions_per_second": 18677.152637796786,
+      "required_pages_per_second_upper_estimate": 0.004320630556094588
     },
     {
       "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": 27872.28125223418,
-      "required_pages_per_second_upper_estimate": 0.0016972756937029332
+      "required_token_positions_per_second": 28015.728956695177,
+      "required_pages_per_second_upper_estimate": 0.0017060109062926676
     },
     {
       "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": 27872.28125223418,
-      "required_pages_per_second_upper_estimate": 0.0021456853662096594
+      "required_token_positions_per_second": 28015.728956695177,
+      "required_pages_per_second_upper_estimate": 0.0021567283675877277
     },
     {
       "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": 27872.28125223418,
-      "required_pages_per_second_upper_estimate": 0.34608045062808623
+      "required_token_positions_per_second": 28015.728956695177,
+      "required_pages_per_second_upper_estimate": 0.3478615910289082
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08297353964064769,
-    "producer_p95_seconds": 0.22251176400095574,
-    "gpu_update_seconds": 0.9919532509662731,
-    "producer_batches_per_second": 12.052034953925391,
-    "gpu_batches_per_second": 1.0081120244587016,
-    "producer_headroom": 11.95505525330545,
+    "producer_batch_seconds": 0.08303451996846434,
+    "producer_p95_seconds": 0.2721742869980517,
+    "gpu_update_seconds": 0.9868741963750581,
+    "producer_batches_per_second": 12.043183971916617,
+    "gpu_batches_per_second": 1.0133003818249124,
+    "producer_headroom": 11.885107504082193,
     "sustainable": true,
-    "buffer_seconds": 60.509148308942656,
+    "buffer_seconds": 60.19932597887855,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 514136.99316793267,
-    "upload_bytes_per_second": 5960134.521493151,
+    "generation_bytes_per_second": 494022.765644503,
+    "upload_bytes_per_second": 5947747.0696020685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.32183023718343,
+    "estimated_full_shards_per_hour": 106.00578524590794,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 12637313,
       "pages": 46,
       "shards": 4
+    },
+    {
+      "seconds": 1.1576985259998764,
+      "bytes": 6035027,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -123,3 +123,4 @@
 {"step": 97300, "loss": 0.6078848466277122, "lr": 0.0003, "grad_norm": 0.145436077212289, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14238146498, "tokens_per_second": 148191.2769510072, "data_wait_s": 1.0234001820208505e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 97400, "loss": 0.6127508264034986, "lr": 0.0003, "grad_norm": 0.12496944282964012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14252830089, "tokens_per_second": 149580.05692383242, "data_wait_s": 8.546001481590793e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 97500, "loss": 0.5998757071793079, "lr": 0.0003, "grad_norm": 0.1421152956588826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14267492681, "tokens_per_second": 148811.02533675297, "data_wait_s": 7.196998922154307e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 97600, "loss": 0.7062435783445835, "lr": 0.0003, "grad_norm": 0.16340439951346944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14282107337, "tokens_per_second": 147795.30343363353, "data_wait_s": 1.7122998542618006e-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 sha256:0035e4890b46411fa5512ee21250dbc01716f4ce0452cb7f8a357d54dea6eff9
-size 266481
+oid sha256:280fe04f2c3da2bad48f4f9fddfa0aa8180949b0cb8f5d1b09e755bbdd335d4b
+size 268599