Running a 26B parameter model with only 6GB RAM using mmap
This guide covers running Gemma 4 26B MOE (Mixture of Experts) locally on an Apple Silicon Mac using llama.cpp with memory-mapped files. The MOE architecture activates only 8 of 128 experts per token, making it incredibly efficient.
| Spec | Value |
|---|
A synthesis of Andrej Karpathy's No Priors interview (March 2026), the auto research ecosystem, and what it all means for how we build things.
Karpathy says he hasn't typed a line of code since December 2025. Not because he stopped working — he works 16 hours a day — but because the verb changed:
Launch a new agent that has access to the following tools: Bash, Glob, Grep, LS, exit_plan_mode, Read, Edit, MultiEdit, Write, NotebookRead, NotebookEdit, WebFetch, TodoRead, TodoWrite, WebSearch. When you are searching for a keyword or file and are not confident that you will find the right match in the first few tries, use the Agent tool to perform the search for you.
When to use the Agent tool:
When NOT to use the Agent tool:
How would you improve the prompt based on the below document? Be explicit, provide detailed analysis of what's missing, what could be better, what would be tweaked. And then after all your analysis and bullet points, give me the new prompt.
https://github.com/steipete/agent-scripts/blob/main/skills/create-cli/SKILL.md
This note documents how we determined the lowest Pydantic 2.x version that
works across our supported Python versions (3.10–3.14) using uv and pytest.
It also records the final requirement change made to pyproject.toml.
| # File: run.py | |
| #!/usr/bin/env python3 | |
| """ | |
| Working demonstration of the BioMCP API with actual function calls. | |
| This shows the corrected API in action. | |
| """ | |
| import asyncio | |
| from biomcp.variants.search import ( |
GitHub Issue: Enable Schema-Formatted Output for Tool-Using Chains
1. Problem:
Currently, the llm CLI's --schema option applies to the direct output of a single Language Model (LLM) call. When tools (via --tool or --functions) are used, the LLM engages in a multi-step chain (e.g., ReAct pattern) where intermediate outputs are tool call requests or textual reasoning. There's no direct way to specify that the final, user-visible result of such a multi-step, tool-using chain should conform to a user-defined schema. The existing --schema option doesn't automatically apply to the culmination of this chain.
2. Alternatives Considered:
--final-schema or --output-schema) specifically for specifying the schema of the final output after a tool chain. This would keep the existing --schema behavior for direct, single-turn schema output and make the post-chain formatting explicit.--schema (Implic
| # Demo output - assume llm 0.26 installed and in path | |
| # | |
| # llm-mcp repo: | |
| # https://github.com/genomoncology/llm-mcp | |
| # | |
| # includes: | |
| # - Desktop Command (local MCP): https://desktopcommander.app/ | |
| # - Git MCP for simonw/llm: https://gitmcp.io/simonw/llm (auth-less remote example) | |
| # | |
| # version 0.0.2 |
| #!/usr/bin/env -S uv --quiet run --script | |
| # /// script | |
| # requires-python = ">=3.11" | |
| # dependencies = [ | |
| # "typer", | |
| # ] | |
| # /// | |
| """ | |
| Clean Text: Unicode Hyphen and Space Normalizer |
What does this change https://grants.nih.gov/grants/guide/notice-files/NOT-OD-25-047.html
Bottom-line: NIH’s new Public Access Policy (NOT-OD-25-047) scraps the 12-month waiting period and makes every NIH-funded paper publicly available immediately at publication, starting with manuscripts accepted on or after December 31 2025.