Shell-AI: a simple CLI to write your bash commands & more LLM news

Week 35 of Coding with Intelligence

📰 News

• OpenAI on track to generate more than $1 bln revenue over 12 months

📦 Repos

• LIDA: Automatic Generation of Visualizations and Infographics using Large Language Models

• Shell-AI: LLM autocomplete CLI for your terminal

  I created a wrapper for my own use, I use it daily! Even on my Raspberry Pi :) tldr; pip install shell-ai then run something like shai terraform dry run thingy and it suggests three commands you can choose to execute or dismiss.

• WizardCoder-Python-34B-V1.0 reaches 73.2 pass@1 for HumanEval

  Here's the HF checkpoint: https://huggingface.co/WizardLM/WizardCoder-Python-34B-V1.0

📄 Papers

• Unnatural Instructions: Tuning Language Models with (Almost) No Human Labor

  This paper describes a fine-tuning technique that was used by Code Llama to create the highly performant "Unnatural Llama" variant that showed GPT-4 level coding performance on HumanEval.

• PMET: Precise Model Editing in a Transformer

🛠️ Products

• Anyscale serverless LLM endpoint/fine-tuning

• Fireworks: A new LLM inference platform

  Impressive inference latency. The following numbers are achieved for Llama 2 70B (non-quantized). Time to first token: 553 ms, Successive tokens: 27.33 ms.

📚 Resources

• A minimal CLI dashboard for OpenAI fine-tuning jobs

---

Want more? Follow me on Twitter! @ricklamers