dddsaty/SOLAR-Instruct-ko-Adapter-Attach
Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:10.7BQuant:FP8Ctx Length:4kPublished:Jan 29, 2024License:cc-by-nc-4.0Architecture:Transformer Open Weights Warm

The dddsaty/SOLAR-Instruct-ko-Adapter-Attach is a 10.7 billion parameter causal language model, built upon the Upstage SOLAR-10.7B-Instruct-v1.0 base model. This model integrates a DPO-applied adapter, specifically fine-tuned for Korean language processing using the beomi/OPEN-SOLAR-KO-10.7B adapter and the maywell/ko_Ultrafeedback_binarized corpus. It is optimized for instruction-following tasks in Korean, demonstrating strong performance across various benchmarks including ARC, HellaSwag, MMLU, and TruthfulQA.

Loading preview...

dddsaty/SOLAR-Instruct-ko-Adapter-Attach: Korean-Optimized Instruction Model

This model, developed by dddsaty, is a 10.7 billion parameter instruction-tuned language model derived from the Upstage SOLAR-10.7B-Instruct-v1.0 base. Its key differentiator is the integration of a DPO (Direct Preference Optimization) applied adapter, specifically designed to enhance its performance and instruction-following capabilities in the Korean language.

Key Capabilities & Features

  • Korean Language Specialization: Fine-tuned using the beomi/OPEN-SOLAR-KO-10.7B adapter and the maywell/ko_Ultrafeedback_binarized corpus, making it highly proficient in understanding and generating Korean text.
  • Instruction Following: Benefits from DPO application, improving its ability to adhere to user instructions and generate relevant responses.
  • Solid Benchmark Performance: Achieves a competitive average score of 74.11 across various benchmarks, including:
    • ARC: 71.08
    • HellaSwag: 88.2
    • MMLU: 66.09
    • TruthfulQA: 71.51
    • Winogrande: 83.5
    • GSM8K: 64.29
  • Context Length: Supports a context length of 4096 tokens.

Good For

  • Applications requiring high-quality, instruction-tuned Korean language generation.
  • Developers looking for a robust 10.7B parameter model with strong performance in Korean NLP tasks.
  • Research and development in Korean-centric large language models.
Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p