GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
idealab-cs2
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
idealab-cs2
Model Tree
Qwen/Qwen3-4B-Thinking-2507
Input Modalities
Output Modalities
Supported Functionality
The model was trained with the following system prompt:
text
You are helping someone reduce a negative emotion they feel in a short interpersonal scenario byoffering an alternative interpretation of the situation (a 'reappraisal' / 'rethinking'). Directyour response at the person in the scenario in second person. Limit your response to two sentencesmaximum. Do not list emotions; output only the reappraisal text.The user turn is
text
SCENARIO:\n{scenario}\n\nWrite a reappraisal of this scenario in two sentences maximum, addressing the person in second person.<think>...</think> followed by the reappraisal; only the text after
</think> is the answer (and the only part scored by the reward during training).
The reward is the Bradley-Terry reward model's scalar score of the answer, with a format gate that penalizes outputs that are empty, longer than two sentences, or over ~340 characters (to prevent the policy from inflating the reward through verbosity). The KL penalty to the base model is the main regularizer.
Reappraisals were compared pairwise against GPT-4-0314's reappraisals for the same scenarios, judged by Llama-3.1-70B-Instruct. That judge was selected by calibration against the human preference data (it agrees with the human-preferred reappraisal 93% of the time on this task; larger judges such as Llama-3.1-405B-FP8 and GLM-5.2 agreed less and showed position bias).
| model | win-rate vs GPT-4-0314 | n | 95% CI lower |
|---|---|---|---|
| Reappraisal-4B-GRPO (this model) | 0.581 | 480 | 0.537 |
| Qwen3-4B-Thinking (base, untrained) | ~0.46 | — | — |
| Qwen3-4B-Instruct + same GRPO recipe (non-reasoning) | 0.458 | 120 | 0.369 |
python
import torchfrom transformers import AutoModelForCausalLM, AutoTokenizer tok = AutoTokenizer.from_pretrained("idealab-cs2/reappraisal-4b-grpo")model = AutoModelForCausalLM.from_pretrained( "idealab-cs2/reappraisal-4b-grpo", torch_dtype=torch.bfloat16, device_map="cuda") system = ("You are helping someone reduce a negative emotion they feel in a short interpersonal " "scenario by offering an alternative interpretation of the situation (a 'reappraisal' / " "'rethinking'). Direct your response at the person in the scenario in second person. " "Limit your response to two sentences maximum. Do not list emotions; output only the " "reappraisal text.")scenario = "A coworker took credit for your idea in a meeting and you feel humiliated."user = f"SCENARIO:\n{scenario}\n\nWrite a reappraisal of this scenario in two sentences maximum, addressing the person in second person." ids = tok.apply_chat_template([{"role": "system", "content": system}, {"role": "user", "content": user}], add_generation_prompt=True, return_tensors="pt").to(model.device)out = model.generate(ids, max_new_tokens=1280, do_sample=True, temperature=0.7, top_p=0.9)text = tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True)reappraisal = text.rsplit("</think>", 1)[-1].strip() # answer follows the reasoning traceprint(reappraisal)You are helping someone reduce a negative emotion they feel in a short interpersonal scenario byoffering an alternative interpretation of the situation (a 'reappraisal' / 'rethinking'). Directyour response at the person in the scenario in second person. Limit your response to two sentencesmaximum. Do not list emotions; output only the reappraisal text.import torchfrom transformers import AutoModelForCausalLM, AutoTokenizer tok = AutoTokenizer.from_pretrained("idealab-cs2/reappraisal-4b-grpo")model = AutoModelForCausalLM.from_pretrained( "idealab-cs2/reappraisal-4b-grpo", torch_dtype=torch.bfloat16, device_map="cuda") system = ("You are helping someone reduce a negative emotion they feel in a short interpersonal " "scenario by offering an alternative interpretation of the situation (a 'reappraisal' / " "'rethinking'). Direct your response at the person in the scenario in second person. " "Limit your response to two sentences maximum. Do not list emotions; output only the " "reappraisal text.")scenario = "A coworker took credit for your idea in a meeting and you feel humiliated."user = f"SCENARIO:\n{scenario}\n\nWrite a reappraisal of this scenario in two sentences maximum, addressing the person in second person." ids = tok.apply_chat_template([{"role": "system", "content": system}, {"role": "user", "content": user}], add_generation_prompt=True, return_tensors="pt").to(model.device)out = model.generate(ids, max_new_tokens=1280, do_sample=True, temperature=0.7, top_p=0.9)text = tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True)reappraisal = text.rsplit("</think>", 1)[-1].strip() # answer follows the reasoning traceprint(reappraisal)