Sorjin1-7B-Instruct / README.md
muzaffercky's picture
add licence
a77e100 verified
|
Raw
History Blame Contribute Delete
2.15 kB
metadata
language:
  - ku
  - kmr
base_model:
  - Qwen/Qwen2.5-7B-Instruct
  - muzaffercky/Sorjin1-7B
tags:
  - kurdish
  - kurmanji
  - instruct
  - merge
  - mergekit
  - qwen2.5
license: apache-2.0
datasets:
  - muzaffercky/kurdish-kurmanji-theses

Sorjîn1-7B-Instruct

Sorjîn1-7B-Instruct is a Kurdish Kurmanji instruction-following model produced by merging Sorjîn1-7B with Qwen/Qwen2.5-7B-Instruct using mergekit.

It combines Kurdish language knowledge from continual pretraining with the instruction-following capability of Qwen2.5-7B-Instruct.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "muzaffercky/Sorjin1-7B-Instruct",
    torch_dtype="bfloat16",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("muzaffercky/Sorjin1-7B-Instruct")

messages = [{"role": "user", "content": "Tu çawa yî?"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Merge Details

  • Method: TIES merge
  • Models: Sorjîn1-7B (Kurdish base) + Qwen/Qwen2.5-7B-Instruct
  • Interpolation factor: 0.5 (equal weight)
  • Density: 0.7
  • Tool: mergekit

Limitations

  • Kurdish language coverage is limited to formal academic style — trained on university theses
  • May mix languages (Kurdish/English/Turkish) on topics not covered in training data
  • Not aligned with RLHF — use with appropriate safety considerations

Model Family

Model Description
Sorjîn1-LoRA Raw LoRA adapter weights
Sorjîn1-7B Continually pretrained base model
Sorjîn1-7B-Instruct (this model) Merged with Qwen2.5-7B-Instruct