perf.py: Error out if test length > cache size

2026-04-20 14:29:51 +00:00 · 2026-02-17 20:04:13 +01:00
parent 3f9c053227
commit b2b6f37e12
1 changed files with 3 additions and 0 deletions
--- a/eval/perf.py
+++ b/eval/perf.py
@@ -119,6 +119,9 @@ def measure_generate(args, model, cache, warmup = False):
@torch.inference_mode()
 def main(args):

+    assert args.max_length <= args.cache_size, \
+        "max_length cannot exceed cache size"
+
    model, config, cache, tokenizer = model_init.init(args, max_chunk_size = args.chunk_size)
    bpw_layer, bpw_head, vram_bits = model.get_storage_info()