VAGO solutions Llama-3-SauerkrautLM-70b-Instruct
Introducing Llama-3-SauerkrautLM-70b-Instruct – our Sauerkraut version of the powerful meta-llama/Meta-Llama-3-70B-Instruct!
The model Llama-3-SauerkrautLM-70b-Instruct is a joint effort between VAGO Solutions and Hyperspace.ai.
- Aligned with DPO
Table of Contents
- Overview of all Llama-3-SauerkrautLM-70b-Instruct
- Model Details
- Evaluation
- Disclaimer
- Contact
- Collaborations
- Acknowledgement
All SauerkrautLM-llama-3-70b-Instruct
Model Details
SauerkrautLM-llama-3-70B-Instruct
- Model Type: Llama-3-SauerkrautLM-70b-Instruct is a finetuned Model based on meta-llama/Meta-Llama-3-70B-Instruct
- Language(s): German, English
- License: meta-llama
- Contact: VAGO solutions, Hyperspace.ai
Training procedure:
- We trained this model with DPO Fine-Tuning for 1 epoch with 70k data.
We improved the model's capabilities noticably by feeding it with curated German data.
Prompt Template:
English:
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
You are a helpful AI assistant.<|eot_id|><|start_header_id|>user<|end_header_id|>
Input<|eot_id|><|start_header_id|>assistant<|end_header_id|>
German:
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
Du bist ein freundlicher und hilfreicher deutscher KI-Assistent.<|eot_id|><|start_header_id|>user<|end_header_id|>
Input<|eot_id|><|start_header_id|>assistant<|end_header_id|>
Evaluation
Open LLM Leaderboard:
evaluated with lm-evaluation-benchmark-harness 0.4.2
Metric | Value |
---|---|
Avg. | 80.98 |
ARC (25-shot) | 74.31 |
HellaSwag (10-shot) | 87.56 |
MMLU (5-shot) | 81.09 |
TruthfulQA (0-shot) | 67.01 |
Winogrande (5-shot) | 84.69 |
GSM8K (5-shot) | 91.20 |
MT-Bench English
########## First turn ##########
score
model turn
Llama-3-SauerkrautLM-70b-Instruct 1 8.86875
########## Second turn ##########
score
model turn
Llama-3-SauerkrautLM-70b-Instruct 2 8.506329
########## Average ##########
score
model
Llama-3-SauerkrautLM-70b-Instruct 8.688679
MT-Bench German
########## First turn ##########
score
model turn
Llama-3-SauerkrautLM-70b-Instruct 1 8.725
########## Second turn ##########
score
model turn
Llama-3-SauerkrautLM-70b-Instruct 2 8.5
########## Average ##########
score
model
Llama-3-SauerkrautLM-70b-Instruct 8.6125
German RAG LLM Evaluation corrected result after FIX: https://github.com/huggingface/lighteval/pull/171
| Task |Version|Metric|Value| |Stderr|
|------------------------------------------------------|------:|------|----:|---|-----:|
|all | |acc |0.980|± |0.0034|
|community:german_rag_eval:_average:0 | |acc |0.980|± |0.0034|
|community:german_rag_eval:choose_context_by_question:0| 0|acc |0.998|± |0.0014|
|community:german_rag_eval:choose_question_by_context:0| 0|acc |1.000|± |0.0000|
|community:german_rag_eval:context_question_match:0 | 0|acc |0.973|± |0.0051|
|community:german_rag_eval:question_answer_match:0 | 0|acc |0.949|± |0.0070|
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 websites. We are also grateful for your feedback and suggestions.
Collaborations
We are also keenly seeking support and investment for our startups, VAGO solutions and Hyperspace 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 at VAGO solutions, Hyperspace.computer
Acknowledgement
Many thanks to Meta for providing such valuable model to the Open-Source community. Many thanks to redponike and cortecs for the quant. version
- Downloads last month
- 2,546