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.

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

78%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

Highly actionable content with strong progressive disclosure and clear workflow checklists. The main weakness is redundancy between the Quick start and Workflow 1 sections, which inflates the token budget without adding capability.

Suggestions

Remove the duplicated 8-bit/4-bit code blocks from the Quick start, keeping only a single concise example and pointing to Workflow 1 for the full procedure.

Add an explicit validation/verification step to the QLoRA (Workflow 2) and 8-bit optimizer (Workflow 3) checklists, e.g. confirming trainable parameters or measuring optimizer memory before training.

Trim the repeated '<1% accuracy loss' / memory-reduction framing in the body since it already appears in the frontmatter description.

Dimension	Reasoning	Score
Conciseness	Mostly efficient code-and-table content that assumes Claude's competence, but the Quick-start 8-bit/4-bit snippets are duplicated in Workflow 1 Step 3, and the opening memory claim repeats the description.	3 / 5
Actionability	Fully executable, copy-paste-ready code throughout—BitsAndBytesConfig, from_pretrained loading, LoRA setup, Trainer config—with specific examples covering the common cases.	5 / 5
Workflow Clarity	Three workflows each have explicit numbered checklists; Workflow 1 Step 4 includes test-inference and memory-check verification. Minor validation gaps in Workflows 2 and 3, which lack explicit post-train verification steps.	4 / 5
Progressive Disclosure	Clear overview with well-signaled one-level-deep references—qlora-training.md, quantization-formats.md, memory-optimization.md—all real files linked from the Advanced topics section, with quick-start and workflows kept inline.	5 / 5
	Total	17 / 20 Passed

Description

92%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, third-person description that concretely states capabilities and explicit usage triggers with low conflict risk. Trigger-term coverage is solid but could include a few more natural synonyms practitioners might say.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions—'Quantizes LLMs to 8-bit or 4-bit', 'Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers'—with comprehensive coverage of capabilities.	5 / 5
Completeness	Explicitly answers both what ('Quantizes LLMs... for 50-75% memory reduction') and when ('Use when GPU memory is limited, need to fit larger models, or want faster inference') with concrete trigger phrases.	5 / 5
Trigger Term Quality	Good natural triggers ('GPU memory is limited', 'need to fit larger models', 'want faster inference') but a few common phrasings like 'quantize a model' or 'reduce model memory' are not surfaced as explicit triggers.	4 / 5
Distinctiveness Conflict Risk	Clear niche (bitsandbytes quantization on HuggingFace Transformers) with distinct triggers and minimal overlap with other skill domains.	5 / 5
	Total	19 / 20 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/bitsandbytes/SKILL.md
Commit: 3a6c3a9

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.