Views
No views yet
LumiOpen/Llama-Poro-2-8B-Instruct.⚠️ DISCLAIMER: This is a research prototype. It is NOT a medical device and should not be used for actual therapy or crisis intervention without human supervision. The model cannot diagnose or prescribe.
| Developed by | kidahatsu |
| Model Type | LoRA Adapter (Rank 16, Alpha 32) |
| Language | Finnish (Suomi) |
| Base Model | LumiOpen/Llama-Poro-2-8B-Instruct |
| Training Data | ~500 curated clinical samples |
| Clinical Quality Score | 4.35/5.0 (Gemini 3 Pro Judge) |
| Hallucination Rate | ~5% |
pip install torch transformers peft bitsandbytes accelerate1from transformers import AutoTokenizer, AutoModelForCausalLM
2from peft import PeftModel
3
4BASE_MODEL = "LumiOpen/Llama-Poro-2-8B-Instruct"
5ADAPTER = "kidahatsu/fintherapy-8b"
6
7tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
8model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, device_map="auto", load_in_4bit=True)
9model = PeftModel.from_pretrained(model, ADAPTER)
10
11messages = [{"role": "user", "content": "Ahdistaa mennä töihin."}]
12prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
13inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
14
15outputs = model.generate(**inputs, max_new_tokens=256, temperature=0.7, do_sample=True)
16print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))Llama-Poro-2-8B-Instruct model has strong latent priors that may trigger hallucinations:inference.py script contains a safety filter to block known hallucination patterns.