From 4b85605edf7085a337f3e8a8a530a5c096d91c6d Mon Sep 17 00:00:00 2001 From: biondizzle Date: Tue, 19 May 2026 15:28:17 +0000 Subject: [PATCH] Fix fp8 amax in decode test --- tests/test_decode_attention_b200.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/test_decode_attention_b200.py b/tests/test_decode_attention_b200.py index 7c12f123..c62be70d 100644 --- a/tests/test_decode_attention_b200.py +++ b/tests/test_decode_attention_b200.py @@ -292,7 +292,7 @@ def test_prefill_decode(layer_id, compress_ratio): kv_cache_bf16[:N_PREFILL] = kv_rope_prefill print(f" Prefill: {N_PREFILL} tokens written to KV cache") - print(f" KV cache fp8 amax: {kv_fp8_prefill.amax():.4f}") + print(f" KV cache fp8 amax: {kv_fp8_prefill.float().abs().max():.4f}") print(f" KV BF16 amax: {kv_rope_prefill.amax():.4f}") # Verify roundtrip: read back and compare