macadeliccc/SOLAR-math-2x10.7b

🌞🚀 SOLAR-math-2x10.7_19B

• This model is part of MoE experimentation. The other solar models in the collection are available here

• If you like this model, version 2 is even better! It is competitve with GPT-3.5 Turbo and Gemini Pro. It exceeds the scores of Mixtral8x7b macadeliccc/SOLAR-math-2x10.7b-v0.2

🌅 Code Example

Example also available in colab

from transformers import AutoModelForCausalLM, AutoTokenizer

def generate_response(prompt):
    """
    Generate a response from the model based on the input prompt.

    Args:
    prompt (str): Prompt for the model.

    Returns:
    str: The generated response from the model.
    """
    # Tokenize the input prompt
    inputs = tokenizer(prompt, return_tensors="pt")
    
    # Generate output tokens
    outputs = model.generate(**inputs, max_new_tokens=512, eos_token_id=tokenizer.eos_token_id, pad_token_id=tokenizer.pad_token_id)

    # Decode the generated tokens to a string
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    return response


# Load the model and tokenizer
model_id = "macadeliccc/SOLAR-math-2x10.7b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, load_in_4bit=True)

prompt = "Explain the proof of Fermat's Last Theorem and its implications in number theory."


print("Response:")
print(generate_response(prompt), "\n")

Evaluations

Model	AGIEval	GPT4All	TruthfulQA	Bigbench	Average
SOLAR-math-2x10.7b	47.2	75.18	64.73	45.15	58.07

AGIEval

Task	Version	Metric	Value		Stderr
agieval_aqua_rat	0	acc	30.31	±	2.89
		acc_norm	30.31	±	2.89
agieval_logiqa_en	0	acc	43.78	±	1.95
		acc_norm	43.93	±	1.95
agieval_lsat_ar	0	acc	21.74	±	2.73
		acc_norm	19.13	±	2.60
agieval_lsat_lr	0	acc	57.25	±	2.19
		acc_norm	56.47	±	2.20
agieval_lsat_rc	0	acc	68.77	±	2.83
		acc_norm	68.03	±	2.85
agieval_sat_en	0	acc	78.16	±	2.89
		acc_norm	79.13	±	2.84
agieval_sat_en_without_passage	0	acc	47.57	±	3.49
		acc_norm	44.66	±	3.47
agieval_sat_math	0	acc	41.36	±	3.33
		acc_norm	35.91	±	3.24

Average: 47.2%

GPT4All

Task	Version	Metric	Value		Stderr
arc_challenge	0	acc	59.22	±	1.44
		acc_norm	61.43	±	1.42
arc_easy	0	acc	84.26	±	0.75
		acc_norm	83.63	±	0.76
boolq	1	acc	88.69	±	0.55
hellaswag	0	acc	65.98	±	0.47
		acc_norm	84.29	±	0.36
openbookqa	0	acc	34.20	±	2.12
		acc_norm	47.20	±	2.23
piqa	0	acc	81.83	±	0.90
		acc_norm	82.59	±	0.88
winogrande	0	acc	78.45	±	1.16

Average: 75.18%

TruthfulQA

Task	Version	Metric	Value		Stderr
truthfulqa_mc	1	mc1	48.47	±	1.75
		mc2	64.73	±	1.53

Average: 64.73%

Bigbench

Task	Version	Metric	Value		Stderr
bigbench_causal_judgement	0	multiple_choice_grade	61.05	±	3.55
bigbench_date_understanding	0	multiple_choice_grade	68.56	±	2.42
bigbench_disambiguation_qa	0	multiple_choice_grade	35.27	±	2.98
bigbench_geometric_shapes	0	multiple_choice_grade	31.20	±	2.45
		exact_str_match	0.00	±	0.00
bigbench_logical_deduction_five_objects	0	multiple_choice_grade	30.00	±	2.05
bigbench_logical_deduction_seven_objects	0	multiple_choice_grade	23.43	±	1.60
bigbench_logical_deduction_three_objects	0	multiple_choice_grade	46.00	±	2.88
bigbench_movie_recommendation	0	multiple_choice_grade	35.60	±	2.14
bigbench_navigate	0	multiple_choice_grade	57.50	±	1.56
bigbench_reasoning_about_colored_objects	0	multiple_choice_grade	55.80	±	1.11
bigbench_ruin_names	0	multiple_choice_grade	45.98	±	2.36
bigbench_salient_translation_error_detection	0	multiple_choice_grade	40.58	±	1.56
bigbench_snarks	0	multiple_choice_grade	66.85	±	3.51
bigbench_sports_understanding	0	multiple_choice_grade	71.40	±	1.44
bigbench_temporal_sequences	0	multiple_choice_grade	56.40	±	1.57
bigbench_tracking_shuffled_objects_five_objects	0	multiple_choice_grade	24.00	±	1.21
bigbench_tracking_shuffled_objects_seven_objects	0	multiple_choice_grade	17.09	±	0.90
bigbench_tracking_shuffled_objects_three_objects	0	multiple_choice_grade	46.00	±	2.88

Average: 45.15%

Average score: 58.07%

Elapsed time: 04:05:27

📚 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}
}

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	73.37
AI2 Reasoning Challenge (25-Shot)	68.43
HellaSwag (10-Shot)	86.31
MMLU (5-Shot)	66.90
TruthfulQA (0-shot)	64.21
Winogrande (5-shot)	83.35
GSM8k (5-shot)	71.04