from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "hiieu/Meta-Llama-3-8B-Instruct-function-calling-json-mode"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are a helpful assistant, answer in JSON with key \"message\""},
    {"role": "user", "content": "Who are you?"},
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
# >> {"message": "I am a helpful assistant, with access to a vast amount of information. I can help you with tasks such as answering questions, providing definitions, translating text, and more. Feel free to ask me anything!"}

Function Calling

Function calling requires two step inferences, below is the example:

Step 1:

python

functions_metadata = [
    {
      "type": "function",
      "function": {
        "name": "get_temperature",
        "description": "get temperature of a city",
        "parameters": {
          "type": "object",
          "properties": {
            "city": {
              "type": "string",
              "description": "name"
            }
          },
          "required": [
            "city"
          ]
        }
      }
    }
]

messages = [
    { "role": "system", "content": f"""You are a helpful assistant with access to the following functions: \n {str(functions_metadata)}\n\nTo use these functions respond with:\n<functioncall> {{ "name": "function_name", "arguments": {{ "arg_1": "value_1", "arg_1": "value_1", ... }} }} </functioncall>\n\nEdge cases you must handle:\n - If there are no functions that match the user request, you will respond politely that you cannot help."""},
    { "role": "user", "content": "What is the temperature in Tokyo right now?"}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
# >> <functioncall> {"name": "get_temperature", "arguments": '{"city": "Tokyo"}'} </functioncall>"""}

Step 2:

python

messages = [
    { "role": "system", "content": f"""You are a helpful assistant with access to the following functions: \n {str(functions_metadata)}\n\nTo use these functions respond with:\n<functioncall> {{ "name": "function_name", "arguments": {{ "arg_1": "value_1", "arg_1": "value_1", ... }} }} </functioncall>\n\nEdge cases you must handle:\n - If there are no functions that match the user request, you will respond politely that you cannot help."""},
    { "role": "user", "content": "What is the temperature in Tokyo right now?"},
    # You will get the previous prediction, extract it will the tag <functioncall>
    # execute the function and append it to the messages like below:
    { "role": "assistant", "content": """<functioncall> {"name": "get_temperature", "arguments": '{"city": "Tokyo"}'} </functioncall>"""},    
    { "role": "user", "content": """<function_response> {"temperature":30 C} </function_response>"""}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
# >> The current temperature in Tokyo is 30 degrees Celsius.

Uploaded model

Developed by: hiieu

This model was trained 2x faster with Unsloth and Huggingface's TRL library.