SOLAR-tail-10.7B-instruct-v1.0

Model Details

Model Developers Kyujin Han (kyujinpy)

Method
Instruction-tuning with PracticeLLM/SOLAR-tail-10.7B-Merge-v1.0.

Datasets
datasets: kyujinpy/KOR-OpenOrca-Platypus-v3.

Hyperparameters

python finetune.py \
    --base_model PracticeLLM/SOLAR-tail-10.7B-Merge-v1.0 \
    --data-path  kyujinpy/KOR-OpenOrca-Platypus-v3 \
    --output_dir ./SOLAR-tail-10.7B-instruct \
    --batch_size 64 \
    --micro_batch_size 1 \
    --num_epochs 1 \
    --learning_rate 3e-5 \
    --cutoff_len 4096 \
    --val_set_size 0 \
    --lora_r 16 \
    --lora_alpha 16 \
    --lora_dropout 0.05 \
    --lora_target_modules '[q_proj, k_proj, v_proj, o_proj, gate_proj, down_proj, up_proj, lm_head]' \
    --train_on_inputs False \
    --add_eos_token False \
    --group_by_length False \
    --prompt_template_name user_prompt \
    --lr_scheduler 'cosine' \

Platypus repo.

Model Benchmark

Open leaderboard

• Follow up as link.

Model	Average	ARC	HellaSwag	MMLU	TruthfulQA	Ko-CommonGenV2
PracticeLLM/SOLAR-tail-10.7B-instruct-v1.0	51.70	46.93	58.19	53.15	46.52	53.72
PracticeLLM/SOLAR-tail-10.7B-Merge-v1.0	48.32	45.73	56.97	38.77	38.75	61.16
jjourney1125/M-SOLAR-10.7B-v1.0	55.15	49.57	60.12	54.60	49.23	62.22

Implementation Code

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

repo = "PracticeLLM/SOLAR-tail-10.7B-instruct-v1.0"
OpenOrca = AutoModelForCausalLM.from_pretrained(
        repo,
        return_dict=True,
        torch_dtype=torch.float16,
        device_map='auto'
)
OpenOrca_tokenizer = AutoTokenizer.from_pretrained(repo)

---