Qwen2.5 7B Kurdish
RegaLabs/Qwen2.5-7B-Kurdish is a Kurdish-focused adaptation of
Qwen/Qwen2.5-7B. This repository
contains the merged full model weights in BF16 safetensors format together
with the tokenizer and generation configuration.
Model details
- Architecture: Qwen2 causal language model
- Size: 7B parameters
- Weight format: BF16 safetensors, sharded across four files
- Context length: up to 32,768 tokens, subject to available memory
- Language: Kurdish (
ku) - Base model:
Qwen/Qwen2.5-7B
The exact fine-tuning dataset, hyperparameters, and evaluation scores are not included in the local release artifacts. They should be added to this card when those details are available.
Usage
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "RegaLabs/Qwen2.5-7B-Kurdish"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
)
messages = [
{"role": "user", "content": "سڵاو، چۆنی؟"},
]
prompt = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
output = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(output[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True))
Limitations
This model may produce incorrect, unsafe, or culturally inappropriate output. It should be evaluated for the intended Kurdish dialects and use cases before deployment. Do not rely on it as an authoritative source or use it for high-impact decisions without appropriate human review.
License
This model is based on Qwen2.5-7B, which is released under the Apache-2.0 license. Users must comply with the upstream license and with all applicable terms for the fine-tuning data and any additional modifications.
- Upstream model: https://huggingface.co/Qwen/Qwen2.5-7B
- Upstream license: https://huggingface.co/Qwen/Qwen2.5-7B/blob/main/LICENSE