Fix start of assistant message detection

2025-04-18 17:46:49 +00:00
parent 23b8c3c54d
commit 33aaaf4455
2 changed files with 47 additions and 17 deletions
--- a/sia/llm_engine/qwq_llm_engine.py
+++ b/sia/llm_engine/qwq_llm_engine.py
@@ -27,9 +27,6 @@ class QwQLlmEngine(LlmEngine):
        """
        self._temperature = temperature

-        with open('/root/sia/qwq_tokenizer_config.json', 'r') as f:
-            tokenizer_config = json.load(f)
-
        quantization_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_compute_dtype=torch.bfloat16,
@@ -47,7 +44,6 @@ class QwQLlmEngine(LlmEngine):

        self._tokenizer = AutoTokenizer.from_pretrained(
            model_path,
-            **tokenizer_config,
        )

        self._pipeline = pipeline(