Mel-Iza0/Llama2-7B_ZeroShot-20K_classe_bias_port

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kArchitecture:Transformer Cold

Mel-Iza0/Llama2-7B_ZeroShot-20K_classe_bias_port is a 7 billion parameter Llama 2 model developed by Mel-Iza0. This model was trained using bitsandbytes 4-bit quantization with nf4 quantization type and bfloat16 compute dtype. Its primary focus is on zero-shot classification tasks, specifically addressing class bias, as indicated by its training methodology.

Loading preview...

Model Overview

Mel-Iza0/Llama2-7B_ZeroShot-20K_classe_bias_port is a 7 billion parameter Llama 2 model. It was developed by Mel-Iza0 with a specific focus on zero-shot classification, particularly in the context of class bias.

Training Details

The model underwent training utilizing bitsandbytes 4-bit quantization. Key configurations for this process included:

  • Quantization Type: nf4
  • Double Quantization: Enabled (bnb_4bit_use_double_quant: True)
  • Compute Data Type: bfloat16 (bnb_4bit_compute_dtype: bfloat16)

This training approach suggests an optimization for efficient deployment while maintaining performance for its intended zero-shot classification tasks.

Key Characteristics

  • Architecture: Llama 2 (7B parameters)
  • Training Method: Quantized using bitsandbytes (4-bit, nf4, bfloat16)
  • Focus: Zero-shot classification with an emphasis on class bias mitigation or analysis.

Potential Use Cases

This model is likely suitable for applications requiring:

  • Zero-shot text classification: Categorizing text without explicit prior examples for each category.
  • Bias analysis: Investigating and potentially mitigating class-related biases in classification outcomes.
  • Resource-efficient deployment: The 4-bit quantization makes it more amenable to environments with limited computational resources.