File size: 3,424 Bytes
5378a1a dcfcf69 5378a1a dcfcf69 5378a1a dcfcf69 5378a1a dcfcf69 5378a1a dcfcf69 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 | ---
license: apache-2.0
language:
- en
pipeline_tag: text-generation
tags:
- qwen2.5
- agentic
- heretic
- uncensored
- fastflowlm
- q4nx
- npu
base_model:
- richardyoung/fable-qwen2.5-3b-agentic-merged-heretic
---
# Qwen2.5-3B-Fable (Agentic/H eretic) - Q4NX for FastFlowLM (AMD Ryzen AI XDNA2)
Fable-Qwen2.5-3B, an agentic merged/heretic fine-tune of Qwen2.5-3B, converted to Q4NX for FastFlowLM. Runs on the Qwen2.5 engine (3B kernels).
## What is Q4NX?
Q4NX is FastFlowLM's native packed-quantization format - a rearranged Q4_1
layout tuned for the NPU matrix engine's tile sizes and memory access
patterns. It is **not** a GGUF file and it does not run on llama.cpp or
Ollama; it is meant exclusively for the [FastFlowLM](https://fastflowlm.com)
engine on AMD Ryzen AI NPUs.
## Requirements
- FastFlowLM >= 0.9.45 (`flm` CLI)
- AMD Ryzen AI processor with **XDNA2 (NPU2)** - Strix Point / Ryzen AI 300
series or later
- Linux with the XRT NPU stack installed
- ~8 GB of unified system memory (Q4NX weights + activations + KV cache)
## Files
| File | Purpose |
|---|---|
| model.q4nx | Quantized Q4NX weights |
| config.json | FastFlowLM model configuration |
| tokenizer.json | Tokenizer |
| tokenizer_config.json | Special tokens and chat template |
| chat_template.jinja | Chat template (optional) |
| flm-add.py | Installer script - registers this model with FastFlowLM |
## Install and run
This repository ships `flm-add.py`, a small installer that copies the model
into the FastFlowLM user directory and registers the tag `qwen2.5-fable:3b`. It never
modifies the system FastFlowLM install.
```bash
# one-time environment (add these to ~/.bashrc)
export FLM_CONFIG_PATH="$HOME/.config/flm/model_list.json"
export FLM_XCLBIN_PATH="$HOME/.config/flm"
git lfs install
git clone https://huggingface.co/Atomic-Germ/Qwen2.5-3B-Fable-Agent-Heretic-NPU2
cd Qwen2.5-3B-Fable-Agent-Heretic-NPU2
python3 ./flm-add.py .
flm run qwen2.5-fable:3b
```
Run `python3 ./flm-add.py --help` for all options. Without a clone, the same
command works against the repo id directly:
```bash
python3 ./flm-add.py Atomic-Germ/Qwen2.5-3B-Fable-Agent-Heretic-NPU2
```
## Kernels
FastFlowLM's NPU kernels (xclbins) are closed source and are not shipped in
this repository. `flm-add.py` links the kernels of the official **`qwen2.5-it:3b`**
model (`Qwen2.5-3B-Instruct-NPU2`), because this model shares the same engine family
(`qwen2`) and architecture.
## Serve (OpenAI-compatible)
```bash
flm serve qwen2.5-fable:3b --port 8080
```
```bash
curl http://127.0.0.1:8080/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{"model":"qwen2.5-fable:3b","messages":[{"role":"user","content":"Hello!"}],"max_tokens":256}'
```
## Model
- Registry tag: `qwen2.5-fable:3b`
- Engine family: `qwen2`
- Kernel source: Qwen2.5-3B-Instruct-NPU2
- Context length: 32,768 tokens (from config)
- `model.q4nx` size: 2.59 GB
- Base model: [richardyoung/fable-qwen2.5-3b-agentic-merged-heretic](https://huggingface.co/richardyoung/fable-qwen2.5-3b-agentic-merged-heretic)
- License: apache-2.0
## Original model card
See the upstream model card for training details, benchmarks, and upstream
usage. This repository only contains the Q4NX conversion for FastFlowLM.
- Upstream card: [richardyoung/fable-qwen2.5-3b-agentic-merged-heretic](https://huggingface.co/richardyoung/fable-qwen2.5-3b-agentic-merged-heretic)
|