Qwen2.5 CoderX 7B V0.5

🚀 Qwen2.5-CoderX-7B-v0.5

Qwen2.5-CoderX-7B-v0.5 is a highly specialized code generation model. It can generate complex, well - structured code with detailed explanations, acting as an AI coding assistant.

🚀 Quick Start

Prompting Strategies

Be specific and comprehensive when prompting the model. Clearly define the problem, desired functionalities, programming language, and constraints. Encourage structured output by using phrases like "Outline your plan before providing the code." For complex tasks, use iterative refinement.

Setting up the Environment (with Unsloth)

Install Unsloth and its dependencies in your Python environment. If using a CUDA - enabled GPU (recommended for performance), run pip install unsloth. Refer to the Unsloth GitHub repository for the latest installation instructions and compatibility.

Running Inference

Here is a Python snippet to load and use the model with Unsloth for inference:

from unsloth import FastLanguageModel
import torch

model_name = "oscar128372/Qwen2.5-CoderX-7B-v0.5"
max_seq_length = 4096
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_name,
    max_seq_length=max_seq_length,
    dtype=dtype,
    load_in_4bit=load_in_4bit,
)

FastLanguageModel.for_inference(model)

tokenizer = get_chat_template(
    tokenizer,
    chat_template = "qwen-2.5",
)

messages = [
    {"role": "user", "content": "Continue the fibonnaci sequence: 1, 1, 2, 3, 5, 8,"},
]
inputs = tokenizer.apply_chat_template(
    messages,
    tokenize = True,
    add_generation_prompt = True, # Must add for generation
    return_tensors = "pt",
).to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer, skip_prompt = True)
_ = model.generate(input_ids = inputs, streamer = text_streamer, max_new_tokens = 128,
                   use_cache = True, temperature = 1.5, min_p = 0.1)

✨ Features

• Complex Code Generation: Capable of handling intricate programming tasks like implementing physics engines or complex algorithms.
• Design Rationale and Planning: Often provides a "Plan of Implementation" before code generation.
• In - depth Explanations: Offers detailed explanations of the generated code.
• Adherence to Best Practices: Demonstrates good software engineering practices.
• Instruction Following: Accurately interprets and addresses complex prompts.

📦 Installation

Install Unsloth and Dependencies

If you're using a CUDA - enabled GPU (recommended for performance), run the following command in your Python environment:

pip install unsloth

Refer to the Unsloth GitHub repository for the latest installation instructions and compatibility.

💻 Usage Examples

Basic Usage

# The following is an example of using the model to continue the Fibonacci sequence
from unsloth import FastLanguageModel
import torch

model_name = "oscar128372/Qwen2.5-CoderX-7B-v0.5"
max_seq_length = 4096
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_name,
    max_seq_length=max_seq_length,
    dtype=dtype,
    load_in_4bit=load_in_4bit,
)

FastLanguageModel.for_inference(model)

tokenizer = get_chat_template(
    tokenizer,
    chat_template = "qwen-2.5",
)

messages = [
    {"role": "user", "content": "Continue the fibonnaci sequence: 1, 1, 2, 3, 5, 8,"},
]
inputs = tokenizer.apply_chat_template(
    messages,
    tokenize = True,
    add_generation_prompt = True, # Must add for generation
    return_tensors = "pt",
).to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer, skip_prompt = True)
_ = model.generate(input_ids = inputs, streamer = text_streamer, max_new_tokens = 128,
                   use_cache = True, temperature = 1.5, min_p = 0.1)

📚 Documentation

Model Details

Model Description

Qwen2.5-CoderX-7B-v0.5 is fine - tuned from unsloth/qwen2.5-coder-7b-instruct-bnb-4bit. It is trained on a high - quality dataset of 46 examples, focusing on complex coding tasks.

Training Data

The model was fine - tuned on a proprietary dataset of 46 examples. These examples are of exceptional quality, high information density, and focus on complex tasks.

Evaluation

Qualitative Assessment: The model performs well on tasks similar to its training data, generating correct code, robust implementations, clear explanations, and well - designed plans.

Performance: It leverages the coding foundation of Qwen2.5 and the efficiency of Unsloth's 4 - bit training. Fine - tuning enhances its ability to solve complex coding problems.

Limitations and Bias

• Limited dataset size may affect generalization.
• The model may overfit to the training data style.
• It may inherit limitations from the base model.
• Human oversight is required for generated code.
• Generating long and complex responses may need significant computational resources.

Ethical Considerations

• The model is for responsible AI development.
• Users should be aware of potential misuse.
• The small dataset may have bias in reflecting diverse coding styles.

Acknowledgements

• Thanks to the Unsloth AI team for their library and base models.
• Thanks to the creators of the Qwen models.

📄 License

This model is licensed under the Apache 2.0 license.