We have used our own OpenOrca dataset to fine-tune on top of Mistral 7B.
This dataset is our attempt to reproduce the dataset generated for Microsoft Research's Orca Paper.
We use OpenChat packing, trained with Axolotl.
This release is trained on a curated filtered subset of most of our GPT-4 augmented data.
It is the same subset of our data as was used in our OpenOrcaxOpenChat-Preview2-13B model.
HF Leaderboard evals place this model as #1 for all models smaller than 30B at release time, outperforming all other 7B and 13B models!
This release provides a first: a fully open model with class-breaking performance, capable of running fully accelerated on even moderate consumer GPUs.
Our thanks to the Mistral team for leading the way here.
We affectionately codename this model: "MistralOrca"
This means that, e.g., in oobabooga the "MPT-Chat" instruction template should work, as it also uses ChatML.
This formatting is also available via a pre-defined Transformers chat template,
which means that lists of messages can be formatted for you with the apply_chat_template() method:
python
1chat =[2{"role":"system","content":"You are MistralOrca, a large language model trained by Alignment Lab AI. Write out your reasoning step-by-step to be sure you get the right answers!"}3{"role":"user","content":"How are you?"},4{"role":"assistant","content":"I am doing well!"},5{"role":"user","content":"Please tell me about how mistral winds have attracted super-orcas."},6]7tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
which will yield:
<|im_start|>system
You are MistralOrca, a large language model trained by Alignment Lab AI. Write out your reasoning step-by-step to be sure you get the right answers!
<|im_end|>
<|im_start|>user
How are you?<|im_end|>
<|im_start|>assistant
I am doing well!<|im_end|>
<|im_start|>user
Please tell me about how mistral winds have attracted super-orcas.<|im_end|>
<|im_start|>assistant
If you use tokenize=True and return_tensors="pt" instead, then you will get a tokenized
and formatted conversation ready to pass to model.generate().
We have evaluated using the methodology and tools for the HuggingFace Leaderboard, and find that we have dramatically improved upon the base model.
We find 106% of the base model's performance on HF Leaderboard evals, averaging 65.84.
At release time, this beats all 7B and 13B models!
This is also 98.6% of Llama2-70b-chat's performance!
HF Leaderboard
Metric
Value
MMLU (5-shot)
62.24
ARC (25-shot)
64.08
HellaSwag (10-shot)
83.99
TruthfulQA (0-shot)
53.05
Avg.
65.84
We use Language Model Evaluation Harness to run the benchmark tests above, using the same version as the HuggingFace LLM Leaderboard.
AGIEval Performance
We compare our results to the base Mistral-7B model (using LM Evaluation Harness).
We find 129% of the base model's performance on AGI Eval, averaging 0.397.
As well, we significantly improve upon the official mistralai/Mistral-7B-Instruct-v0.1 finetuning, achieving 119% of their performance.
AGIEval Performance
BigBench-Hard Performance
We find 119% of the base model's performance on BigBench-Hard, averaging 0.416.
BigBench-Hard Performance
GPT4ALL Leaderboard Performance
We gain a slight edge over our previous releases, again topping the leaderboard, averaging 72.38.
GPT4ALL Performance
MT-Bench Performance
MT-Bench uses GPT-4 as a judge of model response quality, across a wide range of challenges.
We find our performance is on-par with Llama2-70b-chat, averaging 6.86.
MT-Bench Performance
Dataset
We used a curated, filtered selection of most of the GPT-4 augmented data from our OpenOrca dataset, which aims to reproduce the Orca Research Paper dataset.
Training
We trained with 8x A6000 GPUs for 62 hours, completing 4 epochs of full fine tuning on our dataset in one training run.
Commodity cost was ~$400.
Citation
bibtex
1@software{lian2023mistralorca1
2 title = {MistralOrca: Mistral-7B Model Instruct-tuned on Filtered OpenOrcaV1 GPT-4 Dataset},
3 author = {Wing Lian and Bleys Goodson and Guan Wang and Eugene Pentland and Austin Cook and Chanvichet Vong and "Teknium"},
4 year = {2023},
5 publisher = {HuggingFace},
6 journal = {HuggingFace repository},
7 howpublished = {\url{https://huggingface.co/Open-Orca/Mistral-7B-OpenOrca},
8}
9@misc{mukherjee2023orca,
10 title={Orca: Progressive Learning from Complex Explanation Traces of GPT-4},
11 author={Subhabrata Mukherjee and Arindam Mitra and Ganesh Jawahar and Sahaj Agarwal and Hamid Palangi and Ahmed Awadallah},
12 year={2023},
13 eprint={2306.02707},
14 archivePrefix={arXiv},
15 primaryClass={cs.CL}
16}
17@misc{longpre2023flan,
18 title={The Flan Collection: Designing Data and Methods for Effective Instruction Tuning},
19 author={Shayne Longpre and Le Hou and Tu Vu and Albert Webson and Hyung Won Chung and Yi Tay and Denny Zhou and Quoc V. Le and Barret Zoph and Jason Wei and Adam Roberts},
20 year={2023},
21 eprint={2301.13688},
22 archivePrefix={arXiv},
23 primaryClass={cs.AI}
24}
