quantizing-models-bitsandbytes
Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.
- Score
- Unranked
- Evidence
- No signals yet
- Last commit
- Not known
- Listed
Install
armory install quantizing-models-bitsandbytes --cli claudewrites the skill into.claude/skills/quantizing-models-bitsandbytes/SKILL.mdListed as compatible
# fetches the source and writes it to:
.claude/skills/quantizing-models-bitsandbytes/SKILL.mdNeeds the armory CLI · not on npm yet, build it from cli/ in the repository
What it is
Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.
When to use it
Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.
How to install / invoke
# Copy the skill into your .claude/skills/ directory
curl -sL https://raw.githubusercontent.com/davila7/claude-code-templates/main/cli-tool/components/skills/ai-research/optimization-bitsandbytes/SKILL.md -o .claude/skills/optimization-bitsandbytes/SKILL.md
Notes
Extracted from davila7/claude-code-templates, optimization-bitsandbytes category.