---
license: apache-2.0
base_model: Qwen/Qwen3-8B
library_name: transformers
pipeline_tag: text-generation
tags:
  - qwen3
  - dialectic
  - debate
---

# Dialectic 8B (safetensors, bf16)

Fine-tuned Qwen3-8B for the Dialectic project. This is the **full-precision (bf16)
safetensors** copy of the same weights served locally as GGUF at
[`andreiski/dialectic-8b-gguf`](https://huggingface.co/andreiski/dialectic-8b-gguf).

It exists so Apple-Silicon users can convert on-device with MLX:

```bash
mlx_lm.convert --hf-path andreiski/dialectic-8b -q --q-bits 8 --mlx-path ./dialectic-8b-mlx-8bit
```

Same weights as the GGUF Q8_0 / Q4_K_M files — use this for the MLX backend, the GGUF
repo for the Ollama backend.