hemlang/Hembot-7B-v1
Hembot-7B-v1 is a 7.6 billion parameter language model developed by hemlang, fine-tuned using the ORPO method on the Hemlock-Apothecary-7B base model. It features a maximum sequence length of 2048 tokens and was trained with 4-bit quantization. This model is optimized for tasks benefiting from ORPO-based alignment, offering a compact yet capable solution for various language generation needs.
Loading preview...
Hembot-7B-v1 Overview
Hembot-7B-v1 is a 7.6 billion parameter language model developed by hemlang, built upon the hemlang/Hemlock-Apothecary-7B base model. It was fine-tuned using the ORPO (Odds Ratio Preference Optimization) training method, which is designed to align the model's outputs with human preferences more effectively.
Key Technical Specifications
- Base Model:
hemlang/Hemlock-Apothecary-7B - Training Method: ORPO
- Parameter Count: 7.6 billion
- Max Sequence Length: 2048 tokens
- Quantization: 4-bit (NF4)
- Optimizer: paged_adamw_8bit
- LoRA Configuration: Rank 32, Alpha 64, Dropout 0.05, targeting
up_proj,down_proj,gate_proj,k_proj,q_proj,v_proj,o_projmodules.
Training Details
The model underwent 8 epochs of training with a batch size of 4 and a gradient accumulation of 4, resulting in an effective batch size of 16. A cosine learning rate scheduler was used with a warmup ratio of 0.05. The training process also incorporated a maximum gradient norm of 0.1 and a beta value of 0.1 for the ORPO objective.
Potential Use Cases
Given its ORPO-based fine-tuning, Hembot-7B-v1 is well-suited for applications requiring:
- Preference-aligned text generation: Tasks where outputs need to conform to specific stylistic or content preferences.
- Instruction following: Generating responses that accurately follow given instructions.
- General language understanding and generation: A versatile model for various NLP tasks within its context window and parameter size.