---
license: mit
tags:
- code
pipeline_tag: text-generation
---
# Exllama v2 NTQAI/Nxcode-CQ-7B-orpo
Using turboderp's ExLlamaV2 v0.0.21 for quantization.
The "main" branch only contains the measurement.json, download one of the other branches for the model
Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
Original model: NTQAI/Nxcode-CQ-7B-orpo
Calibration dataset is based on: m-a-p/CodeFeedback-Filtered-Instruction
## Available sizes
| Branch | Bits | lm_head bits | Description |
| ----- | ---- | ------- | ------------ |
| [8_0](https://huggingface.co/Apel-sin/nxcode-CQ-7B-orpo-exl2/tree/8_0) | 8.0 | 8.0 | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
| [6_5](https://huggingface.co/Apel-sin/nxcode-CQ-7B-orpo-exl2/tree/6_5) | 6.5 | 8.0 | Very similar to 8.0, good tradeoff of size vs performance, **recommended**. |
| [5_5](https://huggingface.co/Apel-sin/nxcode-CQ-7B-orpo-exl2/tree/5_5) | 5.5 | 8.0 | Slightly lower quality vs 6.5, but usable on 8GB cards. |
## Introduction
Nxcode-CQ-7B-orpo is an [Monolithic Preference Optimization without Reference Model](https://arxiv.org/abs/2403.07691) fine-tune of Qwen/CodeQwen1.5-7B on 100k samples of high-quality ranking data.
## [Evalplus](https://github.com/evalplus/evalplus)
| EvalPlus | pass@1 |
| --- | --- |
| HumanEval | 86.6 |
| HumanEval+ | 83.5 |
| MBPP(v0.2.0) | 82.3 |
| MBPP+(v0.2.0) | 70.4 |
We use a simple template to generate the solution for evalplus:
```python
"Complete the following Python function:\n{prompt}"
```
[Evalplus Leaderboard](https://evalplus.github.io/leaderboard.html)
| Models | HumanEval | HumanEval+|
|------ | ------ | ------ |
| GPT-4-Turbo (April 2024)| 90.2| 86.6|
| GPT-4 (May 2023)| 88.4| 81.17|
| GPT-4-Turbo (Nov 2023)| 85.4| 79.3|
| CodeQwen1.5-7B-Chat| 83.5| 78.7|
| claude-3-opus (Mar 2024)| 82.9| 76.8|
| DeepSeek-Coder-33B-instruct| 81.1| 75.0|
| WizardCoder-33B-V1.1| 79.9| 73.2|
| OpenCodeInterpreter-DS-33B| 79.3| 73.8|
| speechless-codellama-34B-v2.0| 77.4| 72|
| GPT-3.5-Turbo (Nov 2023)| 76.8| 70.7|
| Llama3-70B-instruct| 76.2| 70.7|
## Bigcode Leaderboard
[Bigcode Leaderboard](https://huggingface.co/spaces/bigcode/bigcode-models-leaderboard)
**09/05/2024**
Top 1 average score.
Top 2 winrate.
## Quickstart
Here provides a code snippet with `apply_chat_template` to show you how to load the tokenizer and model and how to generate contents. You should upgrade the transformers if you receive an error when loading the tokenizer
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained(
"NTQAI/Nxcode-CQ-7B-orpo",
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("NTQAI/Nxcode-CQ-7B-orpo")
prompt = """Complete the following Python function:
from typing import List
def has_close_elements(numbers: List[float], threshold: float) -> bool:
""" Check if in given list of numbers, are any two numbers closer to each other than
given threshold.
>>> has_close_elements([1.0, 2.0, 3.0], 0.5)
False
>>> has_close_elements([1.0, 2.8, 3.0, 4.0, 5.0, 2.0], 0.3)
True
"""
"""
messages = [
{"role": "user", "content": prompt}
]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_new_tokens=512, do_sample=False, top_k=50, top_p=0.95, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
res = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
```
### Contact information
For personal communication related to this project, please contact Nha Nguyen Van (nha.nguyen@ntq-solution.com.vn).