flock-io/Flock_Web3_Agent_Model

Introduction

Flock Web3 Agent Model is aimed at helping process function call queries in the specific web3 domain.

Requirements

We advise you to use the latest version of transformers.

Quickstart

Given a query and a list of available tools. The model generate function calls using the provided tools to respond the query correctly.

Example query and tools format

input_example=
{
  "query": "Track crosschain message verification, implement timeout recovery procedures.",
  "tools": [
    {"type": "function", "function": {"name": "track_crosschain_message", "description": "Track the status of a crosschain message", "parameters": {"type": "object", "properties": {"message_id": {"type": "string"}}}}},
    {"type": "function", "function": {"name": "schedule_timeout_check", "description": "Schedule a timeout check for a message", "parameters": {"type": "object", "properties": {"message_id": {"type": "string"}, "timeout": {"type": "integer"}}}}}
  ]
}

Function calling generation

from transformers import AutoModelForCausalLM, AutoTokenizer
import json

model_name = "flock-io/Flock_Web3_Agent_Model"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [
    {"role": "system", "content": "You are a helpful assistant with access to the following functions. Use them if required -"
                                   + json.dumps(input_example["tools"], ensure_ascii=False)},
    {"role": "user", "content": input_example["query"]}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=3000
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

The output text is in the string format

[
    {"name": "track_crosschain_message", "arguments": {"message_id": "msg12345"}},
    {"name": "schedule_timeout_check", "arguments": {"message_id": "msg12345", "timeout": "30"}}
  ]