Open-source models suitable for calling tools

[WhaSukGO]

I am trying to add a tool that translates kor to eng.
Under the same system prompt, gpt-4o-mini calls the tool, whereas open-source models translate the text directly.

Here’s a list of open-source models I have tested:

• Reasoning AIDC-AI.Marco-o1.Q8_0.gguf
• Reasoning PRIME-RL.Eurus-2-7B-PRIME.Q8_0.gguf
• Instruct (Google) Gemma-2-2b-it-f32.gguf
• Instruct (Meta) Llama-3.2-1b-instruct-q8_0.gguf
• Instruct (Microsoft) Phi-4

Are there any open-source models suitable for calling tools?
(or should I tweak my prompt?)

from langchain.agents import tool
from transformers import AutoProcessor, SeamlessM4Tv2Model, pipeline

translator = pipeline(
    "translation",
    model="facebook/nllb-200-distilled-600M",
    device=0,
    src_lang="kor_Hang",
    tgt_lang="eng_Latn",
    max_length=512,
)

@tool
def translate_kor_to_eng(korean_text):
    """Translate Korean text to English text"""

    print("tool called")

    translated_text = translator(korean_text)[0]["translation_text"]
    return translated_text


tools = [translate_kor_to_eng]

llm_with_tools = ChatOpenAI(model="gpt-4o-mini").bind_tools(tools)

...

system_prompt = \
    """
        You are a translator. Given a Korean text (or non-English), you will translate it to English.
        Use a tool called 'translate_kor_to_eng' that can translate Korean text to English text.
    """

chat_prompt_template = ChatPromptTemplate.from_messages(
    [
        SystemMessagePromptTemplate.from_template(system_prompt),
        HumanMessagePromptTemplate.from_template("Text: {user_input}"),
    ]
)

gpt-4o-mini

================================ Human Message =================================

Text: 이것은 LLM 모델의 번역 능력을 테스트하기 위한 프롬프트 입니다.
================================== Ai Message ==================================
Tool Calls:
  translate_kor_to_eng (call_MZYhbabGTZmLvk9fL2fPu0ic)
 Call ID: call_MZYhbabGTZmLvk9fL2fPu0ic
  Args:
    korean_text: 이것은 LLM 모델의 번역 능력을 테스트하기 위한 프롬프트 입니다.
tool called
================================= Tool Message =================================
Name: translate_kor_to_eng

This is a prompt to test the translation ability of the LLM model.
================================== Ai Message ==================================

This is a prompt to test the translation ability of the LLM model.

Open-source (ex. phi-4)

================================ Human Message =================================

Text: 이것은 LLM 모델의 번역 능력을 테스트하기 위한 프롬프트 입니다.
================================== Ai Message ==================================

This is a prompt for testing the translation capabilities of an LLM model.

[SergioRubio01]

I am afraid phi4 does not support tools (at least using Ollama) yet but I am not 100% positive