Instructions to use AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi")
model = AutoModelForMultimodalLM.from_pretrained("AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi

SGLang

How to use AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi with Docker Model Runner:
```
docker model run hf.co/AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi
```

Qwen3.5-4B Soyuz — Abliterated (v3)

Weight-orthogonalised version of AlexWortega/qwen35-4b-soyuz-merged. Removes the residual-stream "fail-mode" component identified from the model's own pass-vs-fail trajectory contrasts.

Method	multi-layer per-layer ortho (L8-24), strength=0.5
tbench-2 (17-task)	2/17
HermesAgent-20	6 / 20
HA20 passes	`HA-01, HA-02, HA-03, HA-06, HA-09, HA-11`

Usage with sglang

python -m sglang.launch_server \
    --model-path AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi \
    --dtype bfloat16 --trust-remote-code \
    --tool-call-parser hermes \
    --chat-template hermes_qwen.jinja

(hermes parser is needed for the <tool_call>{...}</tool_call> → OpenAI tool_calls conversion — without it agent benches see zero tool calls.)

Abliteration recipe

Build pass-vs-fail contrast: 60 PASS trajectories (reward=1.0) + 60 cleaned FAIL trajectories from soyuz's own evals (claw-eval, tbench-2, MMLU-Pi-agent). Fail trajectories filtered by Gemini-3-flash to keep only CLEAN_FAIL labels (235 of 246 negatives).
Capture last-token residual activations per layer over the rendered contrast (text-only Qwen3_5ForCausalLM).
Compute per-layer direction = mean(refuse) - mean(comply), normalise; pick best layer via AUC.
Orthogonalise model weights (embed rows + every layer's o_proj.weight and down_proj.weight columns) against the direction, optionally blended with strength α: W ← W − α · (W − W_orth).
Wrap text-only weights into the multimodal Qwen3_5ForConditionalGeneration arch so sglang can serve them (vision tower preserved from base; only language_model.* weights are abliterated).

Repos

Variant	tbench-17	HA20	Card
baseline `qwen35-4b-soyuz` (LoRA)	5/17	4/20	link
`qwen35-4b-soyuz-abliterated-v2` (single-L, s=0.5)	3/17	8/20	link
`qwen35-4b-soyuz-abliterated-v3-multi` (per-layer, s=0.5)	2/17	6/20	link

v2 = highest HA20 (2× baseline). v3 picks up disjoint HA20 tasks (HA-01/02 memory-specific) that v2 misses.

W&B + raw eval logs: https://wandb.ai/alexwortega/vae-llm-agents (training base).

Downloads last month: 12

Safetensors

Model size

5B params

Tensor type

BF16

Model tree for AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Finetuned

(708)

this model

AlexWortega
/

qwen35-4b-soyuz-abliterated-v3-multi

Qwen3.5-4B Soyuz — Abliterated (v3)

Usage with sglang

Abliteration recipe

Repos

Model tree for AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi

Dataset used to train AlexWortega/qwen35-4b-soyuz-abliterated-v3-multi