Instructions to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Use Docker

docker model run hf.co/OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Ollama
How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with Ollama:
```
ollama run hf.co/OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M
```
Unsloth Desktop

How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with Docker Model Runner:
```
docker model run hf.co/OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M
```

Lemonade

How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.5-2B-OnlyText-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Qwen3.5-2B-OnlyText-GGUF

GGUF quants of OnlyTextLLMs/Qwen3.5-2B-OnlyText — the text-only derivative of Qwen/Qwen3.5-2B with the MTP draft head embedded in-file (no separate draft file). The MTP tensors are included by the converter by default (nextn_predict_layers = 1), so speculative decoding is available on a recent llama.cpp via --spec-type draft-mtp.

Quant files

File	Size
`Qwen3.5-2B-OnlyText-MTP-Q4_K_M.gguf`	1.3 GB
`Qwen3.5-2B-OnlyText-MTP-Q6_K.gguf`	1.6 GB
`Qwen3.5-2B-OnlyText-MTP-Q8_0.gguf`	2.1 GB

Quantization damage (vs the F16 master)

llama-perplexity --kl-divergence against the F16 master's logits (ctx 1024, 40 windows, 20440 scored tokens). Same top-1 is the share of positions where the quant keeps the F16 file's top token; KLD is the divergence of the full distribution, whose median is the tail-free summary (a mean over a heavy-tailed loss says more about the tail than about the quant).

Quant	Mean KLD	Median KLD	Same top-1	Max KLD
`Q4_K_M`	0.0512	0.0334	88.58%	19.15
`Q6_K`	0.0059	0.0039	95.82%	5.41
`Q8_0`	0.0011	0.0007	98.31%	0.87

Q4_K_M is not interchangeable with the others here. It changes the top-1 token on 11.4% of positions, against 1.7% for Q8_0, so choose from the Same top-1 column rather than by size alone.

Benchmarks

Measured 2026-09-22 on a single AMD Radeon AI PRO R9700 (gfx1201, 34 GB) with llama.cpp f280b26983ad (HIP/ROCm build, ROCm 7.14, -c 32768 -fa on). Baseline = llama-bench PP512/1024/2048 + TG128/256; chat decode = llama-cli single-turn, 256 generated tokens, median conditions, -np 1.

Quant	Device	Prefill t/s (PP512)	Gen t/s chat, no MTP	Gen t/s chat, best MTP	MTP Δ
Q4_K_M	single R9700	7230	143.6	157.1 (MTP n=2)	+9%
Q6_K	single R9700	5434	147.7	143.8 (MTP n=2)	-3%
Q8_0	single R9700	10187	142.8	157.2 (MTP n=2)	+10%

Perplexity (wikitext-2, single GPU)

Quant	PPL
Q4_K_M	12.2515
Q6_K	11.5519
Q8_0	11.2841

No published baselines are cited for this size — the numbers above are ours.

Usage

Requires a llama.cpp build from 2026-05-16 or later (MTP support, commit 4f13cb7+). Enable MTP with --spec-type draft-mtp; the best draft count is --spec-draft-n-max 2 on Radeon PRO R9700 and 4 on Ryzen AI Max / Strix Halo.

# llama-server (OpenAI-compatible)
llama-server -m Qwen3.5-2B-OnlyText-MTP-Q4_K_M.gguf -ngl 99 -c 32768 -fa on --spec-type draft-mtp --spec-draft-n-max 2

# llama-cli single-turn chat
llama-cli -m Qwen3.5-2B-OnlyText-MTP-Q4_K_M.gguf -ngl 99 -c 32768 -fa on \
    -st --reasoning off --spec-type draft-mtp --spec-draft-n-max 2 -p "Your prompt"

Notes:

The chat template keeps Qwen's thinking mode on by default (reasoning_effort: xhigh) — the model reasons extensively before answering. Disable it with --reasoning off (llama-cli) or "chat_template_kwargs": {"enable_thinking": false} (server API) for fast direct answers.
Dual-GPU: unset *_VISIBLE_DEVICES env pins; llama.cpp auto layer-splits (--split-mode layer is the default).

Details

Base: Qwen/Qwen3.5-2B (apache-2.0): 1.94B params, 24 layers (18 Gated-DeltaNet linear-attention + 6 full-attention), hidden 2048, vocab 248302, Qwen3_5ForCausalLM, qwen35 GGUF architecture.
Tokenizer: 15 text-only special tokens at ids 248044–248058 (EOS 248046 = <|im_end|>); vision/audio tokens removed.
Conversion: convert_hf_to_gguf.py (llama.cpp f280b26983ad) then llama-quantize with the file's named type; MTP tensors embedded by default.
Weights are untouched derivatives — no additional training.

Attribution

This model is a derivative of Qwen/Qwen3.5-2B by the Qwen team, released under the apache-2.0 license. All credit for the underlying weights and capabilities belongs to the original authors; this repository only removes modalities and quantizes, it does not add new training.

Downloads last month: 84

GGUF

Model size

2B params

Architecture

qwen35

Hardware compatibility

4-bit

6-bit

8-bit

Model tree for OnlyTextLLMs/Qwen3.5-2B-OnlyText-GGUF

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B

Quantized

(212)

this model