bhuvana-ak7/OrpoLlama-3.2-1B-V1_q4_k_m

Hugging Face
TEXT GENERATIONPricing:Input $0.108 / Output $0.804Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Nov 3, 2024Architecture:Transformer Featherless Exclusive Warm

The bhuvana-ak7/OrpoLlama-3.2-1B-V1_q4_k_m is a 1 billion parameter quantized language model, fine-tuned by bhuvana-ak7 using the ORPO (Optimized Regularization for Prompt Optimization) Trainer. It is based on the meta-llama/Llama-3.2-1B architecture and was trained on a subset of the mlabonne/orpo-dpo-mix-40k dataset. This model maintains the performance of its base version, as indicated by hellaswag benchmarks, making it suitable for efficient deployment in applications requiring a compact yet capable LLM.

Loading preview...

Model Overview

The bhuvana-ak7/OrpoLlama-3.2-1B-V1_q4_k_m is a 1 billion parameter language model, representing a quantized version of the bhuvana-ak7/OrpoLlama-3.2-1B-V1 model. It is built upon the meta-llama/Llama-3.2-1B architecture and has been fine-tuned using the ORPO (Optimized Regularization for Prompt Optimization) Trainer.

Key Characteristics

  • Quantized Version: This model is a quantized variant, designed for more efficient inference and reduced memory footprint.
  • ORPO Fine-tuning: It leverages the ORPO training method, specifically fine-tuned on 1000 samples from the mlabonne/orpo-dpo-mix-40k dataset.
  • Performance Preservation: Despite quantization, the model demonstrates identical performance to its non-quantized base model on the hellaswag benchmark, with an acc of 0.4772 and acc_norm of 0.6366.
  • Context Length: Supports a context length of 32768 tokens.

Use Cases

This model is well-suited for applications where:

  • Resource Efficiency is critical, due to its quantized nature.
  • Compact Deployment is required, offering a capable LLM in a smaller package.
  • General Language Understanding tasks are needed, maintaining performance on benchmarks like hellaswag.