├── readme.md
└── prompt_settings.py


/readme.md:
--------------------------------------------------------------------------------
 1 | # Official code for Prompt-Engineering for Open-Source LLMs
 2 | 
 3 | This is a collaboration between [Lamini](https://lamini.ai/) and [Deeplearning.ai](https://www.deeplearning.ai/short-courses/finetuning-large-language-models/). Taught by [Sharon Zhou]().
 4 | 
 5 | ### Lamini API key (free)
 6 | Navigate [here](https://app.lamini.ai/) to copy your key. 
 7 | 
 8 | ### I want more
 9 | OK if you want more, [let us know](https://www.lamini.ai/contact). Particularly for enterprise usage.
10 | 
11 | ### Other resources
12 | 
13 | - [Simple RAG](https://github.com/lamini-ai/simple-rag): RAG in 80 lines of code. Fast, simple, easy. 80% of what you need. RAG is a form of prompt-engineering.
14 | 
15 | - [Lamini SDK](https://github.com/lamini-ai/lamini-sdk/tree/main)
16 | 
17 | - [Lamini docs](https://lamini-ai.github.io/)
18 | 
19 | 


--------------------------------------------------------------------------------
/prompt_settings.py:
--------------------------------------------------------------------------------
 1 | from lamini import Lamini
 2 | 
 3 | # 
 4 | # Mistral 7B
 5 | # 
 6 | 
 7 | llm = Lamini(model_name="mistralai/Mistral-7B-Instruct-v0.1")
 8 | 
 9 | prompts = [
10 |     "Given the fact that I'm drinking green juice, am I healthy?",
11 |     "Respond kindly to the child: i really hate zucchini. why should i eat it?",
12 |     "vscode - how to find code that has been checked in a long time ago, e.g. more than 6 months ago",
13 |     """what does an example jsonl file look like that is loaded with this function? def load_examples(self):
14 |         filename = self.saved_examples_path
15 |         if not os.path.exists(filename):
16 |             return {}
17 | 
18 |         # load the examples from the jsonl file using the jsonlines library
19 |         with jsonlines.open(filename) as reader:
20 |             examples = {}
21 |             for row in reader:
22 |                 class_name = row["class_name"]
23 |                 example = row["examples"]
24 |                 self.add_class(class_name)
25 |                 examples[class_name] = example
26 | 
27 |         return examples""",
28 | ]
29 | 
30 | for prompt in prompts:
31 |     print(f"=============Prompt: {prompt}=============")
32 |     print(llm.generate(prompt))
33 |     print(f"=============Prompt: <s>[INST] {prompt} [/INST]=============")
34 |     print(llm.generate(f"<s>[INST] {prompt} [/INST]"))
35 |     
36 | 
37 | # 
38 | # Llama 2
39 | # 
40 |     
41 | llm_compare = Lamini(model_name="meta-llama/Llama-2-7b-chat-hf")
42 | 
43 | prompts = [
44 |     {
45 |         "system": "You are a healths food nut.",
46 |         "user": "I'm drinking green juice",
47 |     },
48 | ]
49 | 
50 | for prompt in prompts:
51 |     concat_prompt = f"{prompt['system']} {prompt['user']}"
52 |     hydrated_prompt = f"<s>[INST] <<SYS>>\n{prompt['system']}\n<</SYS>>\n{prompt['user']} [/INST]"
53 |     print(f"=============Prompt: {concat_prompt}=============")
54 |     print(llm_compare.generate(concat_prompt))
55 |     print(f"=============Prompt: {hydrated_prompt}=============")
56 |     print(llm_compare.generate(hydrated_prompt))
57 | 
58 | 


--------------------------------------------------------------------------------