Hugging Face's logo

eachadea
/
ggml-vicuna-13b-1.1

Document Question Answering
Model card Files Community
25
ggml-vicuna-13b-1.1 / README.md
eachadea's picture
eachadea
Update README.md
9080c87
preview code
|
raw
history blame
2.6 kB
metadata
license: apache-2.0
inference: false

perplexity stats

NOTE: This GGML conversion is primarily for use with llama.cpp.

  • 13B parameters

  • 4-bit quantized

  • Based on version 1.1

  • Used PR "More accurate Q4_0 and Q4_1 quantizations #896" (should be closer in quality to unquantized)

  • Choosing between q4_0 and q4_1, the logic of higher number = better does not apply. If you are confused, stick with q4_0.

    • If you have performance to spare, it might be worth getting the q4_1. It's ~20% slower and requires 1GB more RAM, but has a ~5% lower perplexity, which is good for generation quality. You're not gonna notice it though.
    • If you have lots of performance to spare, TheBloke's conversion is maybe ~7% better in perplexity but ~50% slower and requires 2GB more RAM.

  • 7B version of this can be found here: https://huggingface.co/eachadea/ggml-vicuna-7b-1.1



Vicuna Model Card

Model details

Model type: Vicuna is an open-source chatbot trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT. It is an auto-regressive language model, based on the transformer architecture.

Model date: Vicuna was trained between March 2023 and April 2023.

Organizations developing the model: The Vicuna team with members from UC Berkeley, CMU, Stanford, and UC San Diego.

Paper or resources for more information: https://vicuna.lmsys.org/

License: Apache License 2.0

Where to send questions or comments about the model: https://github.com/lm-sys/FastChat/issues

Intended use

Primary intended uses: The primary use of Vicuna is research on large language models and chatbots.

Primary intended users: The primary intended users of the model are researchers and hobbyists in natural language processing, machine learning, and artificial intelligence.

Training dataset

70K conversations collected from ShareGPT.com.

Evaluation dataset

A preliminary evaluation of the model quality is conducted by creating a set of 80 diverse questions and utilizing GPT-4 to judge the model outputs. See https://vicuna.lmsys.org/ for more details.

Major updates of weights v1.1

  • Refactor the tokenization and separator. In Vicuna v1.1, the separator has been changed from "###" to the EOS token "</s>". This change makes it easier to determine the generation stop criteria and enables better compatibility with other libraries.
  • Fix the supervised fine-tuning loss computation for better model quality.