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": 75679.87601902093,
-      "required_pages_per_second_upper_estimate": 0.006559739844436566
+      "required_token_positions_per_second": 75706.37746963155,
+      "required_pages_per_second_upper_estimate": 0.006562036923008182
     },
     {
       "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": 9459.984502377616,
-      "required_pages_per_second_upper_estimate": 0.0037737292573709973
+      "required_token_positions_per_second": 9463.297183703944,
+      "required_pages_per_second_upper_estimate": 0.003775050735481069
     },
     {
       "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": 18919.969004755232,
-      "required_pages_per_second_upper_estimate": 0.004376801849168321
+      "required_token_positions_per_second": 18926.594367407888,
+      "required_pages_per_second_upper_estimate": 0.004378334510215613
     },
     {
       "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": 28379.95350713285,
-      "required_pages_per_second_upper_estimate": 0.0017281902704757907
+      "required_token_positions_per_second": 28389.891551111832,
+      "required_pages_per_second_upper_estimate": 0.0017287954452132237
     },
     {
       "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": 28379.95350713285,
-      "required_pages_per_second_upper_estimate": 0.0021847673817185055
+      "required_token_positions_per_second": 28389.891551111832,
+      "required_pages_per_second_upper_estimate": 0.002185532439854274
     },
     {
       "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": 28379.95350713285,
-      "required_pages_per_second_upper_estimate": 0.3523840409641885
+      "required_token_positions_per_second": 28389.891551111832,
+      "required_pages_per_second_upper_estimate": 0.35250743820991387
     }
   ],
   "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": 0.04573885890625817,
-    "producer_p95_seconds": 0.1495671850002509,
-    "gpu_update_seconds": 0.9742087841352917,
-    "producer_batches_per_second": 21.863247661020594,
-    "gpu_batches_per_second": 1.0264740128447936,
-    "producer_headroom": 21.299367921091633,
+    "producer_batch_seconds": 0.09280113936720369,
+    "producer_p95_seconds": 0.1521793259998958,
+    "gpu_update_seconds": 0.9738677567761692,
+    "producer_batches_per_second": 10.77572976817787,
+    "gpu_batches_per_second": 1.0268334617734314,
+    "producer_headroom": 10.494135776961572,
     "sustainable": true,
-    "buffer_seconds": 60.40094461638808,
+    "buffer_seconds": 60.37980092012249,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 176,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 593174.6967376038,
-    "upload_bytes_per_second": 2734484.827849634,
+    "generation_bytes_per_second": 610804.7520996467,
+    "upload_bytes_per_second": 2733373.1832623743,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.64074075983088,
+    "estimated_full_shards_per_hour": 65.53224049683595,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 7369438,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9484031940000932,
+      "bytes": 2544896,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 1700, "loss": 0.9110855236649513, "lr": 0.0003, "grad_norm": 0.2608335000330014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 249356749, "tokens_per_second": 150226.706145175, "data_wait_s": 1.4890999864292098e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1800, "loss": 0.9287391342222691, "lr": 0.0003, "grad_norm": 0.23552261001673377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 263963539, "tokens_per_second": 149206.06424722384, "data_wait_s": 1.4887000361341052e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1900, "loss": 0.9189868532121181, "lr": 0.0003, "grad_norm": 0.23279412071070146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 278648795, "tokens_per_second": 148790.8751210036, "data_wait_s": 1.0176000159844989e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 2000, "loss": 0.9953780174255371, "lr": 0.0003, "grad_norm": 0.2292280217590217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 293112776, "tokens_per_second": 146925.25477945834, "data_wait_s": 1.2533999779407168e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12d072ea19129dc796df73170a653ad0a506c0eb8403b3d76a118549ea420556
-size 39703
+oid sha256:68952f060804efa87ed3beee65d5d372514a5a789578e75115ef4fbf053561b8
+size 41786