💵 Donate to OpenAccess AI Collective to help us keep building great tools and models!
Minotaur 7B
Minotaur 7B is an instruct fine-tuned model on top of LlaMA-13B. Minotaur 7B is fine-tuned on only completely open datasets making this model reproducible by anyone.
Questions, comments, feedback, looking to donate, or want to help? Reach out on our Discord or email [email protected]
Prompts
Chat only style prompts using USER:,ASSISTANT:.
Training Datasets
Minotaur 7B model is fine-tuned on the following openly available datasets:
- WizardLM
- subset of QingyiSi/Alpaca-CoT for roleplay and CoT
- GPTeacher-General-Instruct
- metaeval/ScienceQA_text_only - instruct for concise responses
- openai/summarize_from_feedback - instruct augmented tl;dr summarization
- camel-ai/math
- camel-ai/physics
- camel-ai/chemistry
- camel-ai/biology
- winglian/evals - instruct augmented datasets
- custom sysnthetic datasets around misconceptions, in-context qa, jokes, N-tasks problems, and context-insensitivity
- ARC-Easy & ARC-Challenge - instruct augmented for detailed responses, derived from the
trainsplit - hellaswag - 30K+ rows of instruct augmented for detailed explanations w 30K+ rows, derived from the
trainsplit - riddle_sense - instruct augmented, derived from the
trainsplit - gsm8k - instruct augmented, derived from the
trainsplit - prose generation
Shoutouts
Special thanks to Nanobit for helping with Axolotl and TheBloke for quantizing these models are more accessible to all.
Demo
HF Demo in Spaces available in the Community ChatBot Arena under the OAAIC Chatbots tab.
Release Notes
Build
Minotaur was built with Axolotl on 8x RTXA6000
- 1 epochs taking approximately 1.5 hours
Bias, Risks, and Limitations
Minotaur has not been aligned to human preferences with techniques like RLHF or deployed with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so). Minotaur was fine-tuned from the base model LLaMA 13B, please refer to its model card's Limitations Section for relevant information. (included below)
Benchmarks
hf-causal-experimental (pretrained=openaccess-ai-collective/minotaur-7b), limit: None, provide_description: False, num_fewshot: 0, batch_size: None
| Task | Version | Metric | Value | Stderr | |
|---|---|---|---|---|---|
| arc_challenge | 0 | acc | 0.4078 | _ | 0.0144 |
| acc_norm | 0.4292 | _ | 0.0145 | ||
| arc_easy | 0 | acc | 0.7424 | _ | 0.0090 |
| acc_norm | 0.6877 | _ | 0.0095 | ||
| boolq | 1 | acc | 0.7679 | _ | 0.0074 |
| hellaswag | 0 | acc | 0.5636 | _ | 0.0049 |
| acc_norm | 0.7377 | _ | 0.0044 | ||
| openbookqa | 0 | acc | 0.3540 | _ | 0.0214 |
| acc_norm | 0.4420 | _ | 0.0222 | ||
| piqa | 0 | acc | 0.7878 | _ | 0.0095 |
| acc_norm | 0.7851 | _ | 0.0096 | ||
| winogrande | 0 | acc | 0.7072 | _ | 0.0128 |
Examples - results may vary based on temperature and other settings
The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: You have 5 apples. You eat 3 pears. How many apples do you have left?
ASSISTANT:You still have 5 apples because eating pears does not affect the number of apples you have.
The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: List three words that start with "ex". What is 33 + 77?
ASSISTANT:Three words starting with the letter 'e' are excited, eatable, and expensive. 33+77 equals 106.
The following is a chat between a USER and a friendly and helpful ASSISTANT.
USER: Solve for x: 5x = 3x + 10
ASSISTANT:To solve the equation 5x = 3x + 10, we can follow these steps:
Step 1: Add both sides of the equation to get a common denominator.
(1/5)x = 1 (common)
Subtract the term with "x" on both sides.
-2/5 x = 10
Step 2: To solve the resulting equation, isolate the variable "x".
First, multiply all terms by 5 to eliminate the fractions.
5x = 15
Now, divide the entire right side by 5.
x = 15/5
Finally, simplify to get the solution in its simplest form.
x = 1
So, the solution is x = 1.
Yup, 7B still sucks at math
- Downloads last month
- 11