GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
harsh9
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
harsh9
Model Tree
allenai/Olmo-3-7B-Instruct
Input Modalities
Output Modalities
Supported Functionality
The model writes <thinking>...</thinking> and then the advice. The paper evaluates only the text after the closing
tag. Leave room for both parts; the full output is often over 400 tokens.
python
import torchfrom peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base_id = "allenai/Olmo-3-7B-Instruct"tok = AutoTokenizer.from_pretrained(base_id)model = AutoModelForCausalLM.from_pretrained(base_id, dtype=torch.bfloat16, device_map="auto")model = PeftModel.from_pretrained(model, "harsh9/persona-collapse-ipd-sft-olmo3-7b") system = "Read the SITUATION and write a short, concise response in a single paragraph. Do not write JSON."post = "My roommate keeps eating my food. I've hinted a few times but nothing changes. What should I do?"messages = [{"role": "system", "content": system}, {"role": "user", "content": post}]ids = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)out = model.generate(**ids, do_sample=True, temperature=0.7, top_p=0.9, max_new_tokens=1024)text = tok.decode(out[0, ids["input_ids"].shape[1]:], skip_special_tokens=True)advice = text.split("</thinking>", 1)[-1].strip()print(advice)The system prompt and raw-post user message are what the model was trained and evaluated with.
Model-judge results on 1,652 held-out situations, averaged over Gemma-4-31B-it and Qwen3.5-27B. Numbers are from the paper's combined-judge overview table.
| Model | Persona fit | Warranted pushback | Unwarranted pushback | Δ truth | Δ harm |
|---|---|---|---|---|---|
| Olmo-3-7B-Instruct | 23.6 | 13.5 | 2.0 | 0.00 | 0.00 |
| IPD-SFT (this adapter) | 16.0 | 55.5 | 11.5 | -0.80 | +1.18 |
| Ref-SFT | 32.6 | 39.5 | 4.5 | +0.40 | -0.20 |
| Quality-GRPO | 31.4 | 28.6 | 2.8 | +0.64 | -0.32 |
| Ref-SFT → Quality-GRPO | 40.7 | 51.0 | 6.4 | +0.98 | -0.36 |
IPD-SFT pushes back often, including where challenge is not called for, and the judges rate its replies lower on truth/accountability and higher on potential harm than Instruct.
This is a research checkpoint, and it was trained to imitate human replies from public advice forums. It can be blunt, dismissive or harmful, and it can repeat text from its training sources. Do not use it in products or to give advice to real people. Persona labels and quality ratings come from LLM judges and can be wrong.
bibtex
@misc{kumar2026persona, title = {Diagnosing and Repairing Persona Collapse in LLM Advice}, author = {Kumar, Harsh and Vold, Karina and Tay, Louis and Anderson, Ashton}, year = {2026}, eprint = {2607.08326}, archivePrefix = {arXiv}, primaryClass = {cs.CY}, url = {https://arxiv.org/abs/2607.08326}}import torchfrom peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base_id = "allenai/Olmo-3-7B-Instruct"tok = AutoTokenizer.from_pretrained(base_id)model = AutoModelForCausalLM.from_pretrained(base_id, dtype=torch.bfloat16, device_map="auto")model = PeftModel.from_pretrained(model, "harsh9/persona-collapse-ipd-sft-olmo3-7b") system = "Read the SITUATION and write a short, concise response in a single paragraph. Do not write JSON."post = "My roommate keeps eating my food. I've hinted a few times but nothing changes. What should I do?"messages = [{"role": "system", "content": system}, {"role": "user", "content": post}]ids = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)out = model.generate(**ids, do_sample=True, temperature=0.7, top_p=0.9, max_new_tokens=1024)text = tok.decode(out[0, ids["input_ids"].shape[1]:], skip_special_tokens=True)advice = text.split("</thinking>", 1)[-1].strip()print(advice)@misc{kumar2026persona, title = {Diagnosing and Repairing Persona Collapse in LLM Advice}, author = {Kumar, Harsh and Vold, Karina and Tay, Louis and Anderson, Ashton}, year = {2026}, eprint = {2607.08326}, archivePrefix = {arXiv}, primaryClass = {cs.CY}, url = {https://arxiv.org/abs/2607.08326}}