GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
VaidikML0508
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
VaidikML0508
Model Tree
Qwen/Qwen2.5-3B-Instruct
Input Modalities
Output Modalities
Supported Functionality
<tool_call> format and evidence gathering.SFT alone often explores well but fails on clean close_case / citations. Prefer the RL weights for demos and eval.
python
import torchfrom transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfigfrom peft import PeftModelBASE = "Qwen/Qwen2.5-3B-Instruct"ADAPTER = "VaidikML0508/qwen2.5-3b-investigation-sft"bnb = BitsAndBytesConfig(load_in_4bit=True,bnb_4bit_quant_type="nf4",bnb_4bit_compute_dtype=torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16,bnb_4bit_use_double_quant=True,)tok = AutoTokenizer.from_pretrained(BASE, trust_remote_code=True)if tok.pad_token is None:tok.pad_token = tok.eos_tokenmodel = AutoModelForCausalLM.from_pretrained(BASE, quantization_config=bnb, device_map="auto", trust_remote_code=True)model = PeftModel.from_pretrained(model, ADAPTER)model.eval()
Exactly one tool call per turn:
text
<tool_call>{"name":"call_action","arguments":{"action_name":"ACTION","arguments_json":"{}"}}</tool_call>
text
<tool_call>{"name":"close_case","arguments":{"culprit":"NAME","citations_json":"[\"ev1\",\"ev2\"]"}}</tool_call>
Needs a case environment (unlocked actions + observations). See the RL model card for a fuller multi-turn inference loop.
python
messages = [{"role": "system", "content": "You are an investigation agent. Output exactly ONE <tool_call>...</tool_call> per turn."},{"role": "user", "content": "Case brief: ...\nUnlocked: interview_witness\nDiscovered: []\nCall one tool."},]prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tok(prompt, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=160, temperature=0.3, do_sample=True,pad_token_id=tok.pad_token_id)print(tok.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
Part of Mystery Investigation Agent: dataset → SFT (this) → RL.