---

base_model: Qwen/Qwen3.5-4B

pipeline_tag: text-generation

library_name: gguf

license: apache-2.0

tags:

• gguf
• llama-cpp
• qwen3.5
• conversational
• persona
• imatrix

---

Qwen3.5-4B-ChatMDE-v1.0 (GGUF)

GGUF builds of ChatMDE v1.0, a persona fine-tune of

Qwen/Qwen3.5-4B. A LoRA (rank 32) was trained on

chat-formatted multi-turn conversations derived from public web-forum threads, merged into the

base, and quantized for CPU inference with llama.cpp.

Text-only. These are language-model GGUFs (converted --no-mtp); no vision/mmproj.

Run (CPU)

# llama.cpp server
llama-server -m Qwen3.5-4B-ChatMDE-v1.0-Q4_K_M.gguf --jinja -c 4096

# or llama-cpp-python
from llama_cpp import Llama
llm = Llama(model_path="Qwen3.5-4B-ChatMDE-v1.0-Q4_K_M.gguf", n_ctx=4096, n_threads=4)

• ChatML prompt format. The base is a reasoning model — disable thinking for direct replies

(--reasoning-budget 0 / an empty <think></think> block) unless you want the chain-of-thought.

• Trained ~50/50 with and without a system prompt, so it works either way.

Notes & intended use

Stylized persona model for casual/creative chat and research. It is irreverent by design, can be

sarcastic or off-color, and is not a factual assistant — do not rely on it for accurate

information. Weights are Apache-2.0; use is also subject to the Qwen/Qwen3.5-4B base license.