└── README.md


/README.md:
--------------------------------------------------------------------------------
  1 | # Language models on the command-line
  2 | 
  3 | Notes for a talk I gave at [Mastering LLMs: A Conference For Developers & Data Scientists](https://maven.com/parlance-labs/fine-tuning).
  4 | 
  5 | Here is the [full 67 minute talk video](https://twitter.com/hamelhusain/status/1800741994899394612).
  6 | 
  7 | - [Datasette](https://datasette.io/)
  8 | - [My blog](https://simonwillison.net/)
  9 | - [LLM](https://llm.datasette.io/en/stable/)
 10 | 
 11 | ## Getting started 
 12 | 
 13 | ```bash
 14 | brew install llm # or pipx or pip
 15 | llm keys set openai
 16 | # paste key here
 17 | llm "Say hello in Spanish"
 18 | ```
 19 | ## Installing Claude 3
 20 | ```bash
 21 | llm install llm-claude-3
 22 | llm keys set claude
 23 | # Paste key here
 24 | llm -m haiku 'Say hello from Claude Haiku'
 25 | ```
 26 | 
 27 | ## Local model with llm-gpt4all
 28 | 
 29 | ```bash
 30 | llm install llm-gpt4all
 31 | llm models
 32 | llm chat -m mistral-7b-instruct-v0
 33 | ```
 34 | ## Browsing logs with Datasette
 35 | 
 36 | https://datasette.io/
 37 | 
 38 | ```bash
 39 | pipx install datasette # or brew or pip
 40 | datasette "$(llm logs path)"
 41 | # Browse at http://127.0.0.1:8001/
 42 | ```
 43 | ### Templates
 44 | ```bash
 45 | llm --system 'You are a sentient cheesecake' -m gpt-4o --save cheesecake
 46 | ```
 47 | Now you can chat with a cheesecake:
 48 | ```bash
 49 | llm chat -t cheesecake
 50 | ```
 51 | 
 52 | More plugins: https://llm.datasette.io/en/stable/plugins/directory.html
 53 | 
 54 | ### llm-cmd
 55 | 
 56 | Help with shell commands. Blog entry is here: https://simonwillison.net/2024/Mar/26/llm-cmd/
 57 | 
 58 | ### files-to-prompt and shot-scraper
 59 | 
 60 | `files-to-prompt` is described here: 
 61 | https://simonwillison.net/2024/Apr/8/files-to-prompt/
 62 | 
 63 | `shot-scraper javascript` documentation: https://shot-scraper.datasette.io/en/stable/javascript.html
 64 | 
 65 | JSON output for Google search results:
 66 | 
 67 | ```bash
 68 | shot-scraper javascript 'https://www.google.com/search?q=nytimes+slop' '
 69 | Array.from(
 70 |   document.querySelectorAll("h3"),
 71 |   el => ({href: el.parentNode.href, title: el.innerText})
 72 | )'
 73 | ```
 74 | This version gets the HTML that includes the snippet summaries, then pipes it to LLM to answer a question:
 75 | ```bash
 76 | shot-scraper javascript 'https://www.google.com/search?q=nytimes+slop' '
 77 | () => {
 78 |     function findParentWithHveid(element) {
 79 |         while (element && !element.hasAttribute("data-hveid")) {
 80 |             element = element.parentElement;
 81 |         }
 82 |         return element;
 83 |     }
 84 |     return Array.from(
 85 |         document.querySelectorAll("h3"),
 86 |         el => findParentWithHveid(el).innerText
 87 |     );
 88 | }' | llm -s 'describe slop'
 89 | ```
 90 | ### Hacker news summary
 91 | 
 92 | https://til.simonwillison.net/llms/claude-hacker-news-themes describes my Hacker News summary script in detail.
 93 | 
 94 | ### Embeddings
 95 | 
 96 | Full documentation: https://llm.datasette.io/en/stable/embeddings/index.html
 97 | 
 98 | I ran this:
 99 | 
100 | ```bash
101 | curl -O https://datasette.simonwillison.net/simonwillisonblog.db
102 | llm embed-multi links \
103 |   -d simonwillisonblog.db \
104 |   --sql 'select id, link_url, link_title, commentary from blog_blogmark' \
105 |   -m 3-small --store
106 | ```
107 | Then looked for items most similar to a string like this:
108 | ```bash
109 | llm similar links \
110 |   -d simonwillisonblog.db \
111 |   -c 'things that make me angry'
112 | ```
113 | 
114 | ### More links
115 | 
116 | - [Coping strategies for the serial project hoarder](https://simonwillison.net/2022/Nov/26/productivity/) talk about personal productivity on different projects
117 | - [Figure out how to serve an AWS Lambda function with a Function URL from a custom subdomain](https://github.com/simonw/public-notes/issues/1) as an example of how I use GitHub Issues
118 | 


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