SOLAR-Platypus-10.7B-v1

Model Details

Model Developers Kyujin Han (kyujinpy)

Input Models input text only.

Output Models generate text only.

Model Architecture
SOLAR-Platypus-10.7B-v1 is an auto-regressive language model based on the Llama2 architecture.

Base Model
upstage/SOLAR-10.7B-v1.0

Training Dataset
kyujinpy/Open-platypus-Commercial.

Notice

While training, I used LoRA.
The lora_r values is 16.

Q-LoRA config

• LoRA_r: 16
• LoRA_alpha: 16
• LoRA_dropout: 0.05
• LoRA_target_modules: [gate_proj, up_proj, down_proj]

Prompt

• Alpaca template.

Model Benchmark

Open leaderboard

• Follow up as link.

Model	Average	ARC	HellaSwag	MMLU	TruthfulQA	Winogrande	GSM8K
SOLAR-Platypus-10.7B-v1	58.62	61.69	84.23	60.37	51.58	82.79	11.07
SOLAR-Platypus-10.7B-v2	55.25	59.39	83.57	59.93	43.15	81.45	4.02
upstage/SOLAR-10.7B-v1.0	66.04	61.95	84.60	65.48	45.04	83.66	55.50

Implementation Code

### KO-Platypus
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

repo = "kyujinpy/SOLAR-Platypus-10.7B-v1"
OpenOrca = AutoModelForCausalLM.from_pretrained(
        repo,
        return_dict=True,
        torch_dtype=torch.float16,
        device_map='auto'
)
OpenOrca_tokenizer = AutoTokenizer.from_pretrained(repo)

---

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	58.62
AI2 Reasoning Challenge (25-Shot)	61.69
HellaSwag (10-Shot)	84.23
MMLU (5-Shot)	60.37
TruthfulQA (0-shot)	51.58
Winogrande (5-shot)	82.79
GSM8k (5-shot)	11.07