killawhale2's picture
Update README.md
6625d6d
|
raw
history blame
4.71 kB
metadata
license: apache-2.0

Meet 10.7B Solar: Elevating Performance with Upstage Depth UP Scaling!

(This model is upstage/SOLAR-10.7B-v1.0 fine-tuned version for single-turn conversation. Detailed description to be added.)

Introduction

We introduce the first 10.7 billion (B) parameter model, SOLAR-10.7B. It's compact, yet remarkably powerful, and demonstrates unparalleled state-of-the-art performance in models with parameters under 30B.

We developed the Depth Up-Scaling technique. Built on the Llama2 architecture, SOLAR-10.7B incorporates the innovative Upstage Depth Up-Scaling. We then integrated Mistral 7B weights into the upscaled layers, and finally, continued pre-training for the entire model.

Depth-Upscaled SOLAR-10.7B has remarkable performance. It outperforms models with up to 30B parameters, even surpassing the recent Mixtral 8X7B model. For detailed information, please refer to the experimental table ([link to be updated soon]). Solar 10.7B is an ideal choice for fine-tuning. SOLAR-10.7B offers robustness and adaptability for your fine-tuning needs. Our simple instruction fine-tuning using the SOLAR-10.7B pre-trained model yields significant performance improvements. [[link to be updated soon]]

Instruction Fine-Tuning Strategy

We utilize state-of-the-art instruction fine-tuning methods including supervised fine-tuning (SFT) and direct preference optimization (DPO) [1]. Using open source datasets with Alpaca- and OpenOrca-style and generated synthetic datasets, we apply iterative DPO training, a proprietary alignment strategy, to maximize the performance of our resulting model.

Note: We were careful of data contamination during SFT and DPO, e.g., removing data created using TruthfulQA's prompts.

[1] Rafailov, R., Sharma, A., Mitchell, E., Ermon, S., Manning, C.D. and Finn, C., 2023. Direct preference optimization: Your language model is secretly a reward model. arXiv preprint arXiv:2305.18290.

Evaluation Results

Model H6 Model Size
SOLAR-10.7B-Instruct-v1.0 74.20 ~ 11B
mistralai/Mixtral-8x7B-Instruct-v0.1 72.62 ~ 46.7B
01-ai/Yi-34B-200K 70.81 ~ 34B
01-ai/Yi-34B 69.42 ~ 34B
mistralai/Mixtral-8x7B-v0.1 68.42 ~ 46.7B
meta-llama/Llama-2-70b-hf 67.87 ~ 70B
tiiuae/falcon-180B 67.85 ~ 180B
SOLAR-10.7B-v1.0 66.04 ~11B
mistralai/Mistral-7B-Instruct-v0.2 65.71 ~ 7B
Qwen/Qwen-14B 65.86 ~ 14B
01-ai/Yi-34B-Chat 65.32 ~34B
meta-llama/Llama-2-70b-chat-hf 62.4 ~ 70B
mistralai/Mistral-7B-v0.1 60.97 ~ 7B
mistralai/Mistral-7B-Instruct-v0.1 54.96 ~ 7B

Usage Instructions

This model has been fine-tuned primarily for single-turn conversation, making it less suitable for multi-turn conversations such as chat.

Version

Make sure you have the correct version of the transformers library installed:

pip install transformers==4.35.2

Loading the Model

Use the following Python code to load the model:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Upstage/SOLAR-10.7B-Instruct-v1.0")
model = AutoModelForCausalLM.from_pretrained(
    "Upstage/SOLAR-10.7B-Instruct-v1.0",
    device_map="auto",
    torch_dtype=torch.float16,
)

Conducting Single-Turn Conversation

conversation = [ {'role': 'user', 'content': 'Hello?'} ] 

prompt = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)

inputs = tokenizer(prompt, return_tensors="pt").to(model.device) 
outputs = model.generate(**inputs, use_cache=True, max_length=4096)
output_text = tokenizer.decode(outputs[0]) 
print(output_text)

Below is an example of the output.

<s> ### User:
Hello?

### Assistant:
Hello, how can I assist you today? Please feel free to ask any questions or request help with a specific task.</s>

The Upstage AI Team

Upstage is creating the best LLM and DocAI. Please find more information at https://upstage.ai

Contact Us

Any questions and suggestions, please use the discussion tab. If you want to contact us directly, drop an email to contact@upstage.ai