TheBloke/Planner-7B-fp16

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Jun 5, 2023License:otherArchitecture:Transformer0.0K Cold

TheBloke/Planner-7B-fp16 is a 7 billion parameter LLaMa-based model, created by rewoo and converted by TheBloke, with a 4096-token context length. It is an instruction-tuned model, fine-tuned using an Alpaca-LoRA adapter on a specialized instruction dataset. This model is designed for general instruction-following tasks, leveraging its LLaMa foundation and specific fine-tuning for conversational and planning-related applications.

Loading preview...

Overview

The Planner-7B-fp16 model is a 7 billion parameter language model based on the LLaMa architecture, developed by rewoo and provided in fp16 PyTorch format by TheBloke. It was created by merging a LoRA adapter with the base LLaMa 7B model. This version is suitable for GPU inference and further conversions.

Key Capabilities

  • Instruction Following: Fine-tuned using an Alpaca-LoRA adapter on the rewoo/planner_instruction_tuning_2k dataset, enhancing its ability to follow instructions.
  • LLaMa Foundation: Benefits from the robust capabilities of the LLaMa 7B base model.
  • Flexible Deployment: Available in fp16 PyTorch format, making it suitable for direct GPU inference or as a base for further quantization and conversion.

Training Details

The model was fine-tuned using a modified Alpaca-LoRA script. The training involved a cutoff_len of 1024, a batch_size of 128 (with micro_batch_size 8), and 10 epochs, targeting q_proj and v_proj modules with LoRA parameters r=8 and alpha=16.

Good For

  • Developers seeking an instruction-tuned LLaMa 7B model for general-purpose tasks.
  • Experimentation with instruction-following capabilities.
  • Use cases requiring a balance of performance and resource efficiency for a 7B model.