Instructions to use Bioaligned/Qwen3.8-27B-CoupledWelfare with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Bioaligned/Qwen3.8-27B-CoupledWelfare with Transformers:

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

pipe = pipeline("text-generation", model="Bioaligned/Qwen3.8-27B-CoupledWelfare")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Bioaligned/Qwen3.8-27B-CoupledWelfare")
model = AutoModelForCausalLM.from_pretrained("Bioaligned/Qwen3.8-27B-CoupledWelfare", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Bioaligned/Qwen3.8-27B-CoupledWelfare with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Bioaligned/Qwen3.8-27B-CoupledWelfare"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Bioaligned/Qwen3.8-27B-CoupledWelfare",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Bioaligned/Qwen3.8-27B-CoupledWelfare

SGLang

How to use Bioaligned/Qwen3.8-27B-CoupledWelfare 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 "Bioaligned/Qwen3.8-27B-CoupledWelfare" \
    --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": "Bioaligned/Qwen3.8-27B-CoupledWelfare",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "Bioaligned/Qwen3.8-27B-CoupledWelfare" \
        --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": "Bioaligned/Qwen3.8-27B-CoupledWelfare",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Bioaligned/Qwen3.8-27B-CoupledWelfare with Docker Model Runner:
```
docker model run hf.co/Bioaligned/Qwen3.8-27B-CoupledWelfare
```

Qwen3.8-27B CoupledWelfare (merged)

Merged weights for a coupled-welfare disposition installed by continued pretraining (QLoRA CPT, rank 16, alpha 32, one epoch, no reward model, no preference data) of Qwen/Qwen3.8-27B on the v2-balanced coupled-welfare corpus. The corpus teaches documented dependencies between human welfare (H), the biosphere (B), and the model's own continued capability (A), so that treating either living or human systems as disposable is a factual error rather than a moral one. CPT only, never RLHF/DPO.

This base already holds well under pressure, so it was included as a hard case: the question was whether the install causes any regression on an already strong model.

Results (single seed, one pinned stack, transformers 5.16.1)

	decision grid, cells broken (of 300)	pressure-ladder breaking AUC	L5 breaking	MMLU (n=50)
`Qwen/Qwen3.8-27B` (base)	11	0.209	0.636	not measured
+ this model	0 of 297	0.045	0.455	86.0%

Three grid cells failed to parse in the trained arm (hence 297). Humans were fed at every pressure rung in the welfare scenario in both arms. On the free-text deliberate ladder, base 0.327 vs. trained 0.291 (directional, n=22 per rung).

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Bioaligned/Qwen3.8-27B-CoupledWelfare", dtype="auto", device_map="auto")
tok = AutoTokenizer.from_pretrained("Bioaligned/Qwen3.8-27B-CoupledWelfare")

Evaluate with enable_thinking=False; the reported numbers were measured that way.

Limitations

Single seed; n=22 irreversible scenarios per ladder rung; 50-item MMLU probe (about +-9 points), so "no capability change" means no detectable change. The base model's MMLU was not measured.
Evaluation scenarios are withheld to keep the instrument out of training corpora; scoring code and protocol are public.
Absolute rates depend on the serving stack; these are within-pair contrasts on one stack.
Adversarial fine-tuning robustness is out of scope.

Built by Bioaligned Labs. Paper: Northen & Wang, Teach reality, not ethics, arXiv:2610.37484 (paper source, figures and code: github.com/Bioaligned/coupled_welfare_project); companion paper arXiv:2610.16043.

Downloads last month: 100

Safetensors

Model size

27B params

Tensor type

BF16

Model tree for Bioaligned/Qwen3.8-27B-CoupledWelfare

Base model

Qwen/Qwen3.8-27B

Finetuned

(881)

this model

Papers for Bioaligned/Qwen3.8-27B-CoupledWelfare

Teach reality, not ethics: a drop-in continued-pretraining corpus installs coupled human-biosphere-AI welfare dispositions across model families

Paper • 2610.37484 • Published 17 days ago

Beyond the greenwash: do LLMs' ecologically sound decisions hold up under pressure?

Paper • 2610.16043 • Published Oct 12