language:
- en
- de
license: cc-by-nc-4.0
library_name: transformers
tags:
- finetune
- dpo
- Instruct
- augmentation
- german
datasets:
- argilla/distilabel-math-preference-dpo
pipeline_tag: text-generation
model-index:
- name: LUNA-SOLARkrautLM-Instruct
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: AI2 Reasoning Challenge (25-Shot)
type: ai2_arc
config: ARC-Challenge
split: test
args:
num_few_shot: 25
metrics:
- type: acc_norm
value: 71.16
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/LUNA-SOLARkrautLM-Instruct
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: HellaSwag (10-Shot)
type: hellaswag
split: validation
args:
num_few_shot: 10
metrics:
- type: acc_norm
value: 88.28
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/LUNA-SOLARkrautLM-Instruct
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU (5-Shot)
type: cais/mmlu
config: all
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 66.11
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/LUNA-SOLARkrautLM-Instruct
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: TruthfulQA (0-shot)
type: truthful_qa
config: multiple_choice
split: validation
args:
num_few_shot: 0
metrics:
- type: mc2
value: 73.37
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/LUNA-SOLARkrautLM-Instruct
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Winogrande (5-shot)
type: winogrande
config: winogrande_xl
split: validation
args:
num_few_shot: 5
metrics:
- type: acc
value: 82.95
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/LUNA-SOLARkrautLM-Instruct
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GSM8k (5-shot)
type: gsm8k
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 60.88
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fblgit/LUNA-SOLARkrautLM-Instruct
name: Open LLM Leaderboard
VAGO solutions LUNA-SOLARkrautLM-Instruct
Introducing LUNA-SOLARkrautLM-Instruct – a UNA-Sauerkraut version of the powerful upstage/SOLAR-10.7B-Instruct-v1.0 ! Aligned with DPO and tamed with UNA.
Table of Contents
- Overview of all LUNA-SOLARkrautLM-Instruct models
- Model Details
- Evaluation
- Disclaimer
- Contact
- Collaborations
- Acknowledgement
Model Details
LUNA-SOLARkrautLM-Instruct
- Model Type: LUNA-SOLARkrautLM-Instruct is a UNA Model based on fblgit/UNA-SOLAR-10.7B-Instruct-v1.0 and the powerful set of SauerkrautLM-SOLAR-Instruct
- Language(s): English, German
- License: cc-by-nc-4.0
- Contact: Website David Golchinfar Juanako.AI - UNA
Training Dataset:
LUNA-SOLARkrautLM-Instruct was trained with mix of German data augmentation and translated data.
Aligned through DPO with our new German SauerkrautLM-DPO dataset based on parts of the SFT SauerkrautLM dataset
as chosen answers and Sauerkraut-7b-HerO as rejected answers. Added with additional translated Parts of the HuggingFaceH4/ultrafeedback_binarized (Our dataset do not contain any TruthfulQA prompts - check Data Contamination Test Results) and argilla/distilabel-math-preference-dpo.
We found, that only a simple translation of training data can lead to unnatural German phrasings.
Data augmentation techniques were used to grant grammatical, syntactical correctness and a more natural German wording in our training data.
We improved the German language skills on this model. Nevertheless, certain formulations may occur that are not entirely correct.
Data Contamination Test Results
Some models on the HuggingFace leaderboard had problems with wrong data getting mixed in. We checked our SauerkrautLM-DPO dataset with a special test [1] on this model as target model and upstage/SOLAR-10.7B-Instruct-v1.0 as reference model. The HuggingFace team used the same methods [2, 3].
Our results, with result < 0.1, %: being well below 0.9, indicate that our dataset is free from contamination.
The data contamination test results of HellaSwag and Winograde will be added once [1] supports them.
| Dataset | ARC | MMLU | TruthfulQA | GSM8K |
|---|---|---|---|---|
| SauerkrautLM-DPO | result < 0.1, %: 0.0 | result < 0.1, %: 0.09 | result < 0.1, %: 0.13 | result < 0.1, %: 0.16 |
[1] https://github.com/swj0419/detect-pretrain-code-contamination
Prompt Template:
<|im_start|>system
Du bist LUNA-SOLARkrautLM, ein großes Sprachmodell, das höflich und kompetent antwortet.<|im_end|>
<|im_start|>user
Wie geht es dir?<|im_end|>
<|im_start|>assistant
### User:
Hello, how are you?
### Assistant:
Hi there! I am an AI language model, so I don't have personal feelings or emotions in the traditional sense. However, I can assure you that my systems and processes are functioning well at this moment, allowing me to provide helpful responses for your queries.
How may I assist you today?
Evaluation
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct), gen_kwargs: (), limit: None, num_fewshot: 5, batch_size: auto
|Tasks|Version| Filter |n-shot| Metric |Value | |Stderr|
|-----|-------|----------|-----:|-----------|-----:|---|-----:|
|gsm8k|Yaml |get-answer| 5|exact_match|0.6467|± |0.0132|
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct), gen_kwargs: (), limit: None, num_fewshot: 0, batch_size: auto (64)
| Tasks |Version|Filter|n-shot|Metric|Value | |Stderr|
|--------------|-------|------|-----:|------|-----:|---|-----:|
|truthfulqa_mc2|Yaml |none | 0|acc |0.7368|± |0.0149|
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct), gen_kwargs: (), limit: None, num_fewshot: 25, batch_size: auto (32)
| Tasks |Version|Filter|n-shot| Metric |Value| |Stderr|
|-------------|-------|------|-----:|--------|----:|---|-----:|
|arc_challenge|Yaml |none | 25|acc |0.692|± |0.0135|
| | |none | 25|acc_norm|0.715|± |0.0132|
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct), gen_kwargs: (), limit: None, num_fewshot: 0, batch_size: auto (64)
| Tasks |Version|Filter|n-shot|Metric| Value | |Stderr|
|-----------|-------|------|-----:|------|------:|---|-----:|
|paws_de |Yaml |none | 0|acc | 0.3965|± |0.0109|
|wmt16-en-de|Yaml |none | 0|bleu | 3.5784|± |0.1325|
| | |none | 0|ter |64.5707|± |0.4514|
| | |none | 0|chrf |45.7068|± |0.3861|
|xnli_de |Yaml |none | 0|acc | 0.4129|± |0.0099|
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct), gen_kwargs: (), limit: None, num_fewshot: 10, batch_size: auto (32)
| Tasks |Version|Filter|n-shot| Metric |Value | |Stderr|
|---------|-------|------|-----:|--------|-----:|---|-----:|
|hellaswag|Yaml |none | 10|acc |0.7131|± |0.0045|
| | |none | 10|acc_norm|0.8815|± |0.0032|
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct), gen_kwargs: (), limit: None, num_fewshot: 5, batch_size: auto (64)
| Tasks |Version|Filter|n-shot|Metric| Value | |Stderr|
|-----------|-------|------|-----:|------|------:|---|-----:|
|wmt16-de-en|Yaml |none | 5|bleu |14.9310|± |0.8014|
| | |none | 5|ter |46.3206|± |0.4087|
| | |none | 5|chrf |60.8637|± |0.4436|
|wmt16-en-de|Yaml |none | 5|bleu | 6.2016|± |0.2918|
| | |none | 5|ter |63.9997|± |0.4591|
| | |none | 5|chrf |51.1399|± |0.3978|
|xnli_de |Yaml |none | 5|acc | 0.4703|± |0.0100|
hf (pretrained=fblgit/LUNA-SOLARkrautLM-Instruct,dtype=float16), gen_kwargs: (), limit: None, num_fewshot: 5, batch_size: auto (16)
| Tasks |Version|Filter|n-shot|Metric|Value | |Stderr|
|---------------------------------------|-------|------|-----:|------|-----:|---|-----:|
|mmlu |N/A |none | 0|acc |0.6461|± |0.1215|
| - humanities |N/A |none | 5|acc |0.5960|± |0.1200|
| - formal_logic |Yaml |none | 5|acc |0.4683|± |0.0446|
| - high_school_european_history |Yaml |none | 5|acc |0.8121|± |0.0305|
| - high_school_us_history |Yaml |none | 5|acc |0.8480|± |0.0252|
| - high_school_world_history |Yaml |none | 5|acc |0.8312|± |0.0244|
| - international_law |Yaml |none | 5|acc |0.7851|± |0.0375|
| - jurisprudence |Yaml |none | 5|acc |0.7685|± |0.0408|
| - logical_fallacies |Yaml |none | 5|acc |0.7423|± |0.0344|
| - moral_disputes |Yaml |none | 5|acc |0.7283|± |0.0239|
| - moral_scenarios |Yaml |none | 5|acc |0.3899|± |0.0163|
| - philosophy |Yaml |none | 5|acc |0.7074|± |0.0258|
| - prehistory |Yaml |none | 5|acc |0.7716|± |0.0234|
| - professional_law |Yaml |none | 5|acc |0.4824|± |0.0128|
| - world_religions |Yaml |none | 5|acc |0.7661|± |0.0325|
| - other |N/A |none | 5|acc |0.7097|± |0.0900|
| - business_ethics |Yaml |none | 5|acc |0.7700|± |0.0423|
| - clinical_knowledge |Yaml |none | 5|acc |0.6792|± |0.0287|
| - college_medicine |Yaml |none | 5|acc |0.6647|± |0.0360|
| - global_facts |Yaml |none | 5|acc |0.3600|± |0.0482|
| - human_aging |Yaml |none | 5|acc |0.6861|± |0.0311|
| - management |Yaml |none | 5|acc |0.8350|± |0.0368|
| - marketing |Yaml |none | 5|acc |0.8504|± |0.0234|
| - medical_genetics |Yaml |none | 5|acc |0.6700|± |0.0473|
| - miscellaneous |Yaml |none | 5|acc |0.7893|± |0.0146|
| - nutrition |Yaml |none | 5|acc |0.7549|± |0.0246|
| - professional_accounting |Yaml |none | 5|acc |0.5213|± |0.0298|
| - professional_medicine |Yaml |none | 5|acc |0.7353|± |0.0268|
| - virology |Yaml |none | 5|acc |0.5783|± |0.0384|
| - social_sciences |N/A |none | 5|acc |0.7501|± |0.0684|
| - econometrics |Yaml |none | 5|acc |0.5175|± |0.0470|
| - high_school_geography |Yaml |none | 5|acc |0.8485|± |0.0255|
| - high_school_government_and_politics|Yaml |none | 5|acc |0.8912|± |0.0225|
| - high_school_macroeconomics |Yaml |none | 5|acc |0.6615|± |0.0240|
| - high_school_microeconomics |Yaml |none | 5|acc |0.7311|± |0.0288|
| - high_school_psychology |Yaml |none | 5|acc |0.8385|± |0.0158|
| - human_sexuality |Yaml |none | 5|acc |0.7023|± |0.0401|
| - professional_psychology |Yaml |none | 5|acc |0.6683|± |0.0190|
| - public_relations |Yaml |none | 5|acc |0.6909|± |0.0443|
| - security_studies |Yaml |none | 5|acc |0.7633|± |0.0272|
| - sociology |Yaml |none | 5|acc |0.8358|± |0.0262|
| - us_foreign_policy |Yaml |none | 5|acc |0.8800|± |0.0327|
| - stem |N/A |none | 5|acc |0.5569|± |0.1360|
| - abstract_algebra |Yaml |none | 5|acc |0.3800|± |0.0488|
| - anatomy |Yaml |none | 5|acc |0.6148|± |0.0420|
| - astronomy |Yaml |none | 5|acc |0.7237|± |0.0364|
| - college_biology |Yaml |none | 5|acc |0.7708|± |0.0351|
| - college_chemistry |Yaml |none | 5|acc |0.4600|± |0.0501|
| - college_computer_science |Yaml |none | 5|acc |0.5400|± |0.0501|
| - college_mathematics |Yaml |none | 5|acc |0.2700|± |0.0446|
| - college_physics |Yaml |none | 5|acc |0.3333|± |0.0469|
| - computer_security |Yaml |none | 5|acc |0.7300|± |0.0446|
| - conceptual_physics |Yaml |none | 5|acc |0.6213|± |0.0317|
| - electrical_engineering |Yaml |none | 5|acc |0.6276|± |0.0403|
| - elementary_mathematics |Yaml |none | 5|acc |0.4788|± |0.0257|
| - high_school_biology |Yaml |none | 5|acc |0.8065|± |0.0225|
| - high_school_chemistry |Yaml |none | 5|acc |0.5123|± |0.0352|
| - high_school_computer_science |Yaml |none | 5|acc |0.7000|± |0.0461|
| - high_school_mathematics |Yaml |none | 5|acc |0.3889|± |0.0297|
| - high_school_physics |Yaml |none | 5|acc |0.3576|± |0.0391|
| - high_school_statistics |Yaml |none | 5|acc |0.5926|± |0.0335|
| - machine_learning |Yaml |none | 5|acc |0.4554|± |0.0473|
| Groups |Version|Filter|n-shot|Metric|Value | |Stderr|
|------------------|-------|------|-----:|------|-----:|---|-----:|
|mmlu |N/A |none | 0|acc |0.6461|± |0.1215|
| - humanities |N/A |none | 5|acc |0.5960|± |0.1200|
| - other |N/A |none | 5|acc |0.7097|± |0.0900|
| - social_sciences|N/A |none | 5|acc |0.7501|± |0.0684|
| - stem |N/A |none | 5|acc |0.5569|± |0.1360|
MT-Bench
########## Average ##########
score
model
gpt-4 8.990625
gpt-3.5-turbo 7.943750
claude-instant-v1 7.905660
claude-v1 7.900000
UNA-SOLAR-10.7B-Instruct-v1.0 7.521875
LUNA-SOLARkrautLM-Instruct 7.462500
vicuna-33b-v1.3 7.121875
wizardlm-30b 7.009375
Llama-2-70b-chat 6.856250
Llama-2-13b-chat 6.650000
guanaco-33b 6.528125
tulu-30b 6.434375
guanaco-65b 6.409375
oasst-sft-7-llama-30b 6.409375
palm-2-chat-bison-001 6.400000
mpt-30b-chat 6.393750
vicuna-13b-v1.3 6.387500
wizardlm-13b 6.353125
Llama-2-7b-chat 6.268750
vicuna-7b-v1.3 5.996875
baize-v2-13b 5.750000
nous-hermes-13b 5.553459
mpt-7b-chat 5.459119
gpt4all-13b-snoozy 5.452830
koala-13b 5.350000
mpt-30b-instruct 5.218750
falcon-40b-instruct 5.168750
h2ogpt-oasst-open-llama-13b 4.625000
alpaca-13b 4.531250
chatglm-6b 4.500000
oasst-sft-4-pythia-12b 4.318750
rwkv-4-raven-14b 3.984375
dolly-v2-12b 3.275000
fastchat-t5-3b 3.040625
stablelm-tuned-alpha-7b 2.753125
llama-13b 2.606250
Disclaimer
We must inform users that despite our best efforts in data cleansing, the possibility of uncensored content slipping through cannot be entirely ruled out. However, we cannot guarantee consistently appropriate behavior. Therefore, if you encounter any issues or come across inappropriate content, we kindly request that you inform us through the contact information provided. Additionally, it is essential to understand that the licensing of these models does not constitute legal advice. We are not held responsible for the actions of third parties who utilize our models.
Contact
If you are interested in customized LLMs for business applications, please get in contact with us via our website or contact us at Dr. Daryoush Vaziri. We are also grateful for your feedback and suggestions.
Collaborations
We are also keenly seeking support and investment for our startup, VAGO Solutions, where we continuously advance the development of robust language models designed to address a diverse range of purposes and requirements. If the prospect of collaboratively navigating future challenges excites you, we warmly invite you to reach out to us.
Juanako.AI is also seeking support and investment for our startup, we also are open for collaborating with other labs to make awesome models like this one.
Acknowledgement
Big Hug to VAGO Solutions, we merely used our UNA transformers library on their code and dataset, nothing else. This won't be possible without them, thanks!
Many thanks to argilla and Huggingface for providing such valuable datasets to the Open-Source community. And of course a big thanks to upstage for providing the open source community with their latest technology!
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 73.79 |
| AI2 Reasoning Challenge (25-Shot) | 71.16 |
| HellaSwag (10-Shot) | 88.28 |
| MMLU (5-Shot) | 66.11 |
| TruthfulQA (0-shot) | 73.37 |
| Winogrande (5-shot) | 82.95 |
| GSM8k (5-shot) | 60.88 |