GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
ArmedHipster
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
ArmedHipster
Model Tree
Qwen/Qwen2.5-0.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Greedy decoding, 1,024 new tokens, vLLM fp16, the same prompt and answer parser for both models:
| GSM8K test (1,319) | SVAMP (1,000) | |
|---|---|---|
| Qwen2.5-0.5B-Instruct | 49.1% | 61.6% |
| + this adapter | 53.3% | 63.7% |
| problems fixed / broken | 153 / 98 (sign test p = 0.001) | 107 / 86 (p = 0.15) |
This is the final checkpoint of a single run (step 250), not selected on any test set. The same recipe did not improve Qwen2.5-1.5B-Instruct on GSM8K test (75.8% → 75.7%); the repository explains why (most likely contamination of the GSM8K training split).
\boxed{}, exact numeric match) + 0.2 for formatpython
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base = "Qwen/Qwen2.5-0.5B-Instruct"tok = AutoTokenizer.from_pretrained(base)model = PeftModel.from_pretrained(AutoModelForCausalLM.from_pretrained(base), "ArmedHipster/qwen2.5-0.5b-grpo-gsm8k") question = "Natalia sold clips to 48 of her friends in April, and then she sold half as many clips in May. How many clips did Natalia sell altogether in April and May?"messages = [{"role": "user", "content": question + "\nPlease reason step by step, and put your final answer within \\boxed{}."}]inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")print(tok.decode(model.generate(inputs, max_new_tokens=512, do_sample=False)[0][inputs.shape[1]:], skip_special_tokens=True))The adapter was trained and evaluated with exactly this prompt suffix; other prompts may score differently.
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base = "Qwen/Qwen2.5-0.5B-Instruct"tok = AutoTokenizer.from_pretrained(base)model = PeftModel.from_pretrained(AutoModelForCausalLM.from_pretrained(base), "ArmedHipster/qwen2.5-0.5b-grpo-gsm8k") question = "Natalia sold clips to 48 of her friends in April, and then she sold half as many clips in May. How many clips did Natalia sell altogether in April and May?"messages = [{"role": "user", "content": question + "\nPlease reason step by step, and put your final answer within \\boxed{}."}]inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")print(tok.decode(model.generate(inputs, max_new_tokens=512, do_sample=False)[0][inputs.shape[1]:], skip_special_tokens=True))