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": 75289.67055845342,
-      "required_pages_per_second_upper_estimate": 0.006525917824081273
+      "required_token_positions_per_second": 75221.08669729417,
+      "required_pages_per_second_upper_estimate": 0.006519973148820196
     },
     {
       "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": 9411.208819806678,
-      "required_pages_per_second_upper_estimate": 0.003754271908332008
+      "required_token_positions_per_second": 9402.635837161772,
+      "required_pages_per_second_upper_estimate": 0.003750852017377442
     },
     {
       "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": 18822.417639613355,
-      "required_pages_per_second_upper_estimate": 0.004354235057688137
+      "required_token_positions_per_second": 18805.271674323543,
+      "required_pages_per_second_upper_estimate": 0.004350268640377021
     },
     {
       "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": 28233.626459420033,
-      "required_pages_per_second_upper_estimate": 0.0017192797209887672
+      "required_token_positions_per_second": 28207.90751148531,
+      "required_pages_per_second_upper_estimate": 0.001717713571996434
     },
     {
       "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": 28233.626459420033,
-      "required_pages_per_second_upper_estimate": 0.0021735027205263798
+      "required_token_positions_per_second": 28207.90751148531,
+      "required_pages_per_second_upper_estimate": 0.0021715228047197604
     },
     {
       "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": 28233.626459420033,
-      "required_pages_per_second_upper_estimate": 0.3505671487567209
+      "required_token_positions_per_second": 28207.90751148531,
+      "required_pages_per_second_upper_estimate": 0.3502478054991533
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.18123235121094083,
+    "producer_batch_seconds": 0.0760199385312319,
     "producer_p95_seconds": 0.17086536099986915,
-    "gpu_update_seconds": 0.9792578378033814,
-    "producer_batches_per_second": 5.517778659926313,
-    "gpu_batches_per_second": 1.0211815125658288,
-    "producer_headroom": 5.403327999997081,
+    "gpu_update_seconds": 0.9801506896157633,
+    "producer_batches_per_second": 13.154443680445253,
+    "gpu_batches_per_second": 1.0202512844142546,
+    "producer_headroom": 12.893337044900134,
     "sustainable": true,
-    "buffer_seconds": 60.71398594380965,
+    "buffer_seconds": 60.76934275617733,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 614816.2296682058,
-    "upload_bytes_per_second": 5626894.744196676,
+    "generation_bytes_per_second": 610145.5544581906,
+    "upload_bytes_per_second": 5602022.431596586,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.9626253487331,
+    "estimated_full_shards_per_hour": 65.46151626257557,
     "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": [
@@ -322,6 +322,12 @@
       "bytes": 5950611,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.4311316419998548,
+      "bytes": 7407679,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 84500, "loss": 0.6755691282451153, "lr": 0.0003, "grad_norm": 0.15844106523682092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12365969201, "tokens_per_second": 147706.36542072735, "data_wait_s": 1.2414000138960546e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84600, "loss": 0.5803650245070457, "lr": 0.0003, "grad_norm": 0.12690611449244352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12380705618, "tokens_per_second": 149123.07797826442, "data_wait_s": 1.2340000012045493e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84700, "loss": 0.6427894756197929, "lr": 0.0003, "grad_norm": 0.14702141787082534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12395361653, "tokens_per_second": 147233.25360334205, "data_wait_s": 1.1217999599466566e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 84800, "loss": 0.6267599500715733, "lr": 0.0003, "grad_norm": 0.1399491145580503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12410053275, "tokens_per_second": 149362.03363209727, "data_wait_s": 1.4009000096848467e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d3fd3c4c2873f172a9aa3a7f60eb3782fdce1c2b6f969fea52b191e9eaa2899
-size 59469
+oid sha256:cabf22cae536de552a9137bd764b476dd5f3f329e571c9bd3932ba408a317d95
+size 61587