---
license: other
library_name: peft
tags:
  - qwen
  - peft
  - lora
  - distilled
---

# techhermit/qwen35-slice14b-release

This repository contains the distilled adapter and optional quantized export
for the sliced 14B base checkpoint.

## Provenance

- Base repo: `techhermit/qwen35-slice14b-base`
- Base model: `Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled`
- Best adapter run: `external-behavior-expanded-plus-run6`
- Best eval perplexity: `2.3351974012631542`
- Quantized export: `release_repo-q8_0.gguf`

## Usage

Load the base repo first, then apply the adapter from this repo.

- Base repo: `techhermit/qwen35-slice14b-base`
- Quantized export: `release_repo-q8_0.gguf`

For direct inference, use the GGUF export with llama.cpp. For PEFT-based
loading or further training, load the base repo and apply the adapter from this
repo on top of it.