Qiskit/Qwen2.5-Coder-14B-Qiskit

Qwen-2.5-Coder-14B-Qiskit

Introduction

Qwen-2.5-Coder-14B-Qiskit is a model specialized in Qiskit coding and based on code-specific Qwen large language models. Particularly, this model is based on Qwen2.5-Coder 14 billion parameters model.

Main features compared to previous models specialized on Qiskit code:

• Significant improvements in code generation, code reasoning and code fixing.
• A more comprehensive foundation for real-world applications such as Code Agents. Not only enhancing coding capabilities but also maintaining its strengths in mathematics and general competencies.
• Long-context Support up to 128K tokens.

The model Qwen-2.5-Coder-14B-Qiskit has the following features:

• Type: Causal Language Models
• Training Stage: Pretraining & Post-training
• Architecture: transformers with RoPE, SwiGLU, RMSNorm, and Attention QKV bias
• Number of Parameters: 14.7B
• Number of Paramaters (Non-Embedding): 13.1B
• Number of Layers: 48
• Number of Attention Heads (GQA): 40 for Q and 8 for KV
• Context Length: Full 131,072 tokens

Requirements

Qwen-2.5-Coder-14B-Qiskit is compatible with the latest HuggingFace transformers and we advise you to use the latest version of transformers.

With transformers<4.37.0, you will encounter the following error:

KeyError: 'qwen2'

Quickstart

Here we provide a code snippet with apply_chat_template to show you how to load the tokenizer and model and how to generate content.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qiskit/Qwen-2.5-Coder-14B-Qiskit"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "Generate a random quantum circuit with 5 qubits."
messages = [
    {"role": "system", "content": "You are Qiskit Code Assistant. You are a helpful coding assistant."},
    {"role": "user", "content": prompt}
]
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=512
)
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]

Processing Long Texts

The current config.json is set for context length up to 32,768 tokens. To handle extensive inputs exceeding 32,768 tokens, we utilize YaRN, a technique for enhancing model length extrapolation, ensuring optimal performance on lengthy texts.

For supported frameworks, you could add the following to config.json to enable YaRN:

{
  ...,
  "rope_scaling": {
    "factor": 4.0,
    "original_max_position_embeddings": 32768,
    "type": "yarn"
  }
}

For deployment, we recommend using vLLM. Please refer to Qwen Documentation for usage if you are not familar with vLLM. Presently, vLLM only supports static YARN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts. We advise adding the rope_scaling configuration only when processing long contexts is required.

Training Data

• Data Collection and Filtering: Our code data is sourced from a combination of publicly available datasets (e.g., Code available on https://github.com), and additional synthetic data generated at IBM Quantum. We exclude code that is older than 2023.
• Exact and Fuzzy Deduplication: We use both exact and fuzzy deduplication to remove documents having (near) identical code content.
• PII: We redact Personally Identifiable Information (PII) in our datasets by replacing PII content (e.g., names, email addresses, keys, passwords) with corresponding tokens (e.g., ⟨NAME⟩, ⟨EMAIL⟩, ⟨KEY⟩, ⟨PASSWORD⟩).

Infrastructure

We train Qwen-2.5-Coder-14B-Qiskit using IBM's super computing cluster (Vela) using NVIDIA A100 GPUs. The cluster provides a scalable and efficient infrastructure for training.

Ethical Considerations and Limitations

The use of Large Language Models involves risks and ethical considerations people must be aware of. Regarding code generation, caution is urged against complete reliance on specific code models for crucial decisions or impactful information as the generated code is not guaranteed to work as intended. Qwen-2.5-Coder-14B-Qiskit model is not the exception in this regard. Even though this model is suited for multiple code-related tasks, it has not undergone any safety alignment, there it may produce problematic outputs. Additionally, it remains uncertain whether smaller models might exhibit increased susceptibility to hallucination in generation scenarios by copying source code verbatim from the training dataset due to their reduced sizes and memorization capacities. This aspect is currently an active area of research, and we anticipate more rigorous exploration, comprehension, and mitigations in this domain. Regarding ethics, a latent risk associated with all Large Language Models is their malicious utilization. We urge the community to use Qwen-2.5-Coder-14B-Qiskit model with ethical intentions and in a responsible way.