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": 77295.57048567718,
-      "required_pages_per_second_upper_estimate": 0.0066997841458661415
+      "required_token_positions_per_second": 21024.63016988743,
+      "required_pages_per_second_upper_estimate": 0.0018223616566878452
     },
     {
       "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": 9661.946310709647,
-      "required_pages_per_second_upper_estimate": 0.003854294842312768
+      "required_token_positions_per_second": 2628.078771235929,
+      "required_pages_per_second_upper_estimate": 0.001048379915125231
     },
     {
       "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": 19323.892621419294,
-      "required_pages_per_second_upper_estimate": 0.004470242469070695
+      "required_token_positions_per_second": 5256.157542471858,
+      "required_pages_per_second_upper_estimate": 0.0012159195422375621
     },
     {
       "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": 28985.838932128943,
-      "required_pages_per_second_upper_estimate": 0.0017650855140229013
+      "required_token_positions_per_second": 7884.2363137077855,
+      "required_pages_per_second_upper_estimate": 0.00048010862611375197
     },
     {
       "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": 28985.838932128943,
-      "required_pages_per_second_upper_estimate": 0.0022314101189329066
+      "required_token_positions_per_second": 7884.2363137077855,
+      "required_pages_per_second_upper_estimate": 0.0006069503363922014
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28985.838932128943,
-      "required_pages_per_second_upper_estimate": 0.359907110174565
+      "estimated_pages_for_buffer": 7,
+      "required_token_positions_per_second": 7884.2363137077855,
+      "required_pages_per_second_upper_estimate": 0.09789582817472448
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,21 +155,21 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 18,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.5862758550001672,
-    "producer_p95_seconds": 1.1575575899996693,
-    "gpu_update_seconds": 0.953845084999557,
-    "producer_batches_per_second": 1.7056817050733137,
-    "gpu_batches_per_second": 1.0483882715613766,
-    "producer_headroom": 1.626956110957844,
+    "producer_batch_seconds": 0.03220738701584774,
+    "producer_p95_seconds": 0.08623680900018371,
+    "gpu_update_seconds": 3.5067442044995913,
+    "producer_batches_per_second": 31.048777707671448,
+    "gpu_batches_per_second": 0.2851647972261207,
+    "producer_headroom": 108.88012128317295,
     "sustainable": true,
-    "buffer_seconds": 60.09224035497209,
+    "buffer_seconds": 63.12139568099264,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 6,
-    "max_pending_pages": 1433,
+    "upload_pages_per_commit": 20,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -0,0 +1 @@
+{"step": 280300, "loss": 0.5914102615788579, "lr": 0.0003, "grad_norm": 0.15056614637032292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41028209266, "tokens_per_second": 75128.25690262954, "data_wait_s": 1.346200042462442e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2cf593cb94cc2a9bfac9f0d34de5b17ae161774c447b9eb0b2a958dc33309305
-size 165
+oid sha256:f32632e1c5530ab1ed9f10084601a41d24f14de0a78a56a193fec29b96c4b7e8
+size 2283