cockroach54/solar-sft-qlora

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:10.7BQuant:FP8Ctx Length:4kLicense:apache-2.0Architecture:Transformer0.0K Open Weights Warm

The cockroach54/solar-sft-qlora is a 10.7 billion parameter language model, fine-tuned using QLoRA. This model is trained with transformers, PEFT, and bitsandbytes, and evaluated on various Korean language benchmarks. It demonstrates performance across tasks like sentiment analysis, commonsense reasoning, and question answering, making it suitable for Korean NLP applications.

Loading preview...

Model Overview

The cockroach54/solar-sft-qlora is a 10.7 billion parameter language model, fine-tuned using QLoRA. It was developed as a practice model, leveraging popular libraries such as transformers, PEFT, and bitsandbytes for efficient training.

Key Capabilities & Performance

This model has been evaluated on a suite of Korean language benchmarks using the kollm_evaluation framework. Its performance highlights include:

  • kobest_sentineg: Achieved an accuracy of 0.8791 and an F1 score of 0.8779 for sentiment negation tasks.
  • kobest_copa: Demonstrated an accuracy of 0.6540 and an F1 score of 0.6535 on commonsense reasoning.
  • ko_common_gen: Scored an accuracy of 0.8095 for common generation tasks.
  • ko_truthfulqa: Recorded an accuracy of 0.2938.
  • ko_hellaswag: Achieved an accuracy of 0.3816 and a normalized accuracy of 0.5171.

Use Cases

Given its evaluation on Korean-specific tasks, this model is primarily suited for applications requiring Korean natural language understanding and generation, including sentiment analysis, commonsense reasoning, and general text processing in Korean.

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