Mel-Iza0/Llama2-7B_ZeroShot-20K_classe_other_port

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

Mel-Iza0/Llama2-7B_ZeroShot-20K_classe_other_port is a Llama 2-based language model developed by Mel-Iza0. This model was trained using 4-bit quantization with the nf4 quantization type and double quantization enabled, leveraging bfloat16 for computation. It is specifically fine-tuned for zero-shot classification tasks on a 20K dataset, focusing on the 'other' class. Its primary application is efficient classification in resource-constrained environments due to its optimized training configuration.

Loading preview...

Model Overview

Mel-Iza0/Llama2-7B_ZeroShot-20K_classe_other_port is a Llama 2-based model developed by Mel-Iza0, specifically fine-tuned for zero-shot classification. The training process utilized advanced quantization techniques to optimize for efficiency.

Key Training Details

  • Quantization: The model was trained with bitsandbytes 4-bit quantization (load_in_4bit: True).
  • Quantization Type: It employs nf4 (NormalFloat4) quantization, known for its efficiency in deep learning.
  • Double Quantization: bnb_4bit_use_double_quant was enabled, further reducing memory footprint during training.
  • Compute Data Type: Training computations were performed using bfloat16 (bnb_4bit_compute_dtype: bfloat16).
  • Framework: PEFT (Parameter-Efficient Fine-Tuning) version 0.4.0 was used, indicating an efficient fine-tuning approach.

Intended Use

This model is designed for zero-shot classification tasks, particularly for identifying the 'other' class within a 20,000-sample dataset. Its optimized training with 4-bit quantization makes it suitable for applications where computational resources or memory are a concern, offering efficient inference for classification tasks.