samuelfaj/distill-1.7B-MLX

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kPublished:May 14, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

The samuelfaj/distill-1.7B-MLX is a 1.7 billion parameter domain-specific Expert Language Model, based on Qwen3-1.7B, specifically designed for compressing and classifying raw command-line interface (CLI) output into structured, actionable summaries. This model is the core intelligence behind the 'distill' CLI tool, excelling at tasks like identifying pass/fail states, extracting JSON, and summarizing security audits from terminal logs. It is optimized for macOS (Apple Silicon) platforms, offering high accuracy (95%) on its specialized tasks.

Loading preview...

distill-1.7B-MLX: Expert Language Model for CLI Output

samuelfaj/distill-1.7B-MLX is a specialized 1.7 billion parameter language model, not a general-purpose chatbot, developed by samuelfaj. It functions as the "brain" for the distill CLI tool, focusing exclusively on processing and summarizing raw terminal output. Unlike large general-purpose LLMs, this model is highly efficient and accurate within its narrow domain, trained on 100,000 synthetic CLI outputs.

Key Capabilities

This model performs 8 specialized tasks with an overall accuracy of 95%, achieving 100% accuracy on 6 of these tasks. Its capabilities include:

  • pass_fail: Determining if a command succeeded or failed.
  • safe_review: Assessing the safety of Terraform plans.
  • terraform_plan: Counting resource changes (create, change, destroy) from Terraform output.
  • json_extraction: Pulling structured JSON data from noisy logs.
  • security_audit: Summarizing vulnerabilities by severity.
  • test_result: Reporting test suite pass/fail status.
  • typescript_check: Extracting TypeScript compiler errors.
  • generic: Providing free-form summaries of various CLI outputs.

Good For

  • Automating CLI output analysis: Ideal for developers and DevOps engineers looking to streamline the interpretation of command-line logs.
  • Integrating with distill: This model is specifically built to power the distill engine, an open-source CLI output compression tool.
  • Resource-constrained environments: Its smaller size (1.7B parameters) and MLX fp16 format (3.2 GB) make it suitable for deployment on macOS with Apple Silicon, offering efficient local processing. Other formats like 4-bit MLX and GGUF are also available for broader compatibility and reduced size.

Finetunes
1 models