---
license: apache-2.0
language:
- en
base_model:
- Atomic-Germ/Qwen3.8-2B-Distill-NPU2
base_model_relation: quantized
quantized_by: Atomic-Germ
pipeline_tag: image-text-to-text
tags:
- transformers
- gguf
- qwen3_5
- qwen3.8-froggeric-v22.4
- qwen3_5_text
- image-text-to-text
- conversational
- q4nx
- quantized
- npu2
- fastflowlm
- fastflow
- flm
- empero-ai
- qwen3.8
- distillation
- reasoning
- en
- base_model:empero-ai/Qwen3.8-2B-Distill
- base_model:quantized:empero-ai/Qwen3.8-2B-Distill
- license:apache-2.0
- endpoints_compatible
- region:us
---
# Qwen3.8-Distilled-2B-NPU2-NPU2

**OpenFlowLM Q4NX conversion of [`Atomic-Germ/Qwen3.8-2B-Distill-NPU2`](https://huggingface.co/Atomic-Germ/Qwen3.8-2B-Distill-NPU2)** for AMD XDNA NPU inference.

This repository contains a quantized **Q4NX** port of the model, compiled for the OpenFlowLM (OFLM) runtime. It is **not** a GGUF file.

| Item | Value |
|------|-------|
| Source model | [`Atomic-Germ/Qwen3.8-2B-Distill-NPU2`](https://huggingface.co/Atomic-Germ/Qwen3.8-2B-Distill-NPU2) |
| Source GGUF | `Qwen3.8-2B-Distill-Q8_0.gguf` |
| Weights | `model.q4nx` (2.29 GB) |
| Modality | language / vision |
| OFLM version | `0.1.0` |
| Converted | 2026-09-27 |

## Install and run

This repository works with `oflm-add`, a small installer that copies the model
into the OpenFlowLM user directory and registers the tag. It never
modifies the system OpenFlowLM install.

`pip install oflm-add` or `uv tool install oflm-add`

```bash
uv tool install oflm-add
oflm-add Atomic-Germ/Qwen3.8-Distilled-2B-NPU2-NPU2 --family qwen3.5 --xclbin-from Qwen3.8-Distilled-2B-NPU2-NPU2
OFLM_CONFIG_PATH="$HOME/.config/oflm/model_list.json" OFLM_XCLBIN_PATH="$HOME/.config/oflm" oflm run Qwen3.8-Distilled-2B-NPU2-NPU2
```

## Files

| File | Description |
|------|-------------|
| `model.q4nx` | Quantized weights (Q8_0 / Q4_1 / BF16) |
| `config.json` | OFLM runtime configuration |
| `tokenizer.json` | Tokenizer vocabulary |
| `tokenizer_config.json` | Tokenizer configuration |
| `chat_template.jinja` | Chat template |
| `vision_weight.q4nx` | Vision model |


---

## Source model card

See the original model card: [Atomic-Germ/Qwen3.8-2B-Distill-NPU2](https://huggingface.co/Atomic-Germ/Qwen3.8-2B-Distill-NPU2)