开放代码解释器-DS-33B

Itroductio

Nxcode-CQ-7B-orpo is a Moolithic Preferece Optimizatio without Referece Model fie-tue of Qwe/CodeQwe1.5-7B o 100k samples of high-quality rakig data.

Evalplus

We use a simple template to geerate the solutio for evalplus:

"Complete the followig Pytho fuctio:\{prompt}"

Evalplus Leaderboard

Bigcode Leaderboard

Bigcode Leaderboard

09/05/2024

Top 1 average score.

Top 2 wirate.

Quickstart

Here provides a code sippet with apply_chat_template to show you how to load the tokeizer ad model ad how to geerate cotets. You should upgrade the trasformers if you receive a error whe loadig the tokeizer

from trasformers import AutoModelForCausalLM, AutoTokeizer
device = "cuda" # the device to load the model oto

model = AutoModelForCausalLM.from_pretraied(
    "NTQAI/Nxcode-CQ-7B-orpo",
    torch_dtype="auto",
    device_map="auto"
)
tokeizer = AutoTokeizer.from_pretraied("NTQAI/Nxcode-CQ-7B-orpo")

prompt = """Complete the followig Pytho fuctio:
from typig import List


def has_close_elemets(umbers: List[float], threshold: float) -> bool:
    """ Check if i give list of umbers, are ay two umbers closer to each other tha
    give threshold.
    >>> has_close_elemets([1.0, 2.0, 3.0], 0.5)
    False
    >>> has_close_elemets([1.0, 2.8, 3.0, 4.0, 5.0, 2.0], 0.3)
    True
    """
"""
messages = [
    {"role": "user", "cotet": prompt}
]

iputs = tokeizer.apply_chat_template(messages, add_geeratio_prompt=True, retur_tesors="pt").to(model.device)
outputs = model.geerate(iputs, max_ew_tokes=512, do_sample=False, top_k=50, top_p=0.95, um_retur_sequeces=1, eos_toke_id=tokeizer.eos_toke_id)
res = tokeizer.decode(outputs[0][le(iputs[0]):], skip_special_tokes=True)

Cotact iformatio

For persoal commuicatio related to this project, please cotact Nha Nguye Va (ha.guye@tq-solutio.com.v).

Introduction Nxcode-CQ-7B-orpo is an Monolithic Preference Optimization without Reference Model fine