nbeerbower/Dumpling-Qwen2.5-1.5B

Hugging Face
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jan 29, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

Dumpling-Qwen2.5-1.5B by nbeerbower is a 1.5 billion parameter language model based on the Qwen2.5 architecture, fine-tuned using the QLoRA ORPO method. It was trained on a diverse collection of DPO datasets, including those focused on truthfulness, physical reasoning, and various literary styles. This model is primarily designed for nuanced conversational tasks and instruction following, leveraging its specialized DPO training to generate helpful and contextually appropriate responses.

Loading preview...

Dumpling-Qwen2.5-1.5B Overview

Dumpling-Qwen2.5-1.5B is a 1.5 billion parameter language model developed by nbeerbower, built upon the Qwen2.5 architecture. This model has undergone a specialized fine-tuning process using the QLoRA ORPO method, which involved training on a diverse array of Direct Preference Optimization (DPO) datasets. The training utilized two RTX 3090 GPUs over two epochs, focusing on enhancing its ability to follow instructions and generate contextually relevant text.

Key Capabilities

  • Instruction Following: Optimized through DPO training to better understand and execute user instructions.
  • Nuanced Response Generation: Benefits from datasets like GreatFirewall-DPO, Schule-DPO, and Purpura-DPO to produce varied and context-aware outputs.
  • Truthfulness and Reasoning: Incorporates datasets such as jondurbin/truthy-dpo-v0.1 and antiven0m/physical-reasoning-dpo to improve factual accuracy and logical coherence.
  • Literary Style Adaptation: Fine-tuned with gutenberg-dpo-v0.1, gutenberg2-dpo, and gutenberg-moderne-dpo for potential versatility in generating text with different stylistic nuances.

Good For

  • Applications requiring a compact model with strong instruction-following capabilities.
  • Generating responses that are both helpful and contextually appropriate.
  • Tasks benefiting from improved truthfulness and physical reasoning in generated text.
  • Exploratory use cases where a model trained on diverse DPO datasets might offer unique conversational characteristics.

Finetunes
2 models