fix: Clear kv cache to avoid kv bug when image is evaluated first

2024-05-10 02:38:10 -04:00 · 2024-05-10 02:38:10 -04:00 · ac55d0a175
commit ac55d0a175
parent 4badac3a60
1 changed files with 1 additions and 0 deletions
--- a/llama_cpp/llama_chat_format.py
+++ b/llama_cpp/llama_chat_format.py
@ -2637,6 +2637,7 @@ class Llava15ChatHandler:

        # Evaluate prompt
        llama.reset()
+        llama._ctx.kv_cache_clear()
        for type_, value in split_text:
            if type_ == "text":
                tokens = llama.tokenize(value.encode("utf8"), add_bos=False, special=True)