🌞🚀 SOLAR-10.7x2_19B

Merge of two Solar-10.7B instruct finetunes.

Performs higher than mistralai/mixtral-8x7b-Instruct-v0.1

🌅 Code Example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("macadeliccc/SOLAR-math-2x10.7b",load_in_4bit=True)
model = AutoModelForCausalLM.from_pretrained(
    "macadeliccc/SOLAR-math-2x10.7b",
    device_map="auto",
    torch_dtype=torch.float16,
)

conversation = [ {'role': 'user', 'content': 'A rectangle has a length that is twice its width and its area is 50 square meters. Find the dimensions of the rectangle.'} ] 

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)

Evaluations

model is currently experimental and was evaluated in 4-bit

Tasks	Version	Filter	n-shot	Metric	Value		Stderr
arc_challenge	Yaml	none	0	acc	0.5691	±	0.0145
		none	0	acc_norm	0.5998	±	0.0143
arc_easy	Yaml	none	0	acc	0.8354	±	0.0076
		none	0	acc_norm	0.8258	±	0.0078
boolq	Yaml	none	0	acc	0.8716	±	0.0059
hellaswag	Yaml	none	0	acc	0.6397	±	0.0048
		none	0	acc_norm	0.8268	±	0.0038
openbookqa	Yaml	none	0	acc	0.3380	±	0.0212
		none	0	acc_norm	0.4660	±	0.0223
piqa	Yaml	none	0	acc	0.8139	±	0.0091
		none	0	acc_norm	0.8205	±	0.0090
winogrande	Yaml	none	0	acc	0.7609	±	0.0120

📚 Citations

@misc{kim2023solar,
      title={SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling}, 
      author={Dahyun Kim and Chanjun Park and Sanghoon Kim and Wonsung Lee and Wonho Song and Yunsu Kim and Hyeonwoo Kim and Yungi Kim and Hyeonju Lee and Jihoo Kim and Changbae Ahn and Seonghoon Yang and Sukyung Lee and Hyunbyung Park and Gyoungjin Gim and Mikyoung Cha and Hwalsuk Lee and Sunghun Kim},
      year={2023},
      eprint={2312.15166},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}