GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
qingyangwanxi
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
qingyangwanxi
Model Tree
Qwen/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
以 Qwen/Qwen3.5-4B 为基座，用 Pingo 产品真实用户对话（脱敏后）做 LoRA/QLoRA 监督微调，
使模型稳定具备：Pingo 的人设/语气/回复节奏、产品世界观与规则、与线上一致的多轮对话组织方式、
线上工具协议（OpenAI 兼容 tool_calls）的调用时机与格式，以及安全边界内的拒答能力。
[昵称]；无法确认敏感性的内容默认删除；[图片] 占位）；python
from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("<this-repo>")model = AutoModelForCausalLM.from_pretrained("<this-repo>", device_map="auto")msgs = [{"role": "system", "content": "<Pingo system prompt>"},{"role": "user", "content": "今天上班好累啊"}]ids = tok.apply_chat_template(msgs, tokenize=True, return_dict=True,add_generation_prompt=True).to(model.device)out = model.generate(**ids, max_new_tokens=200, temperature=0.7, top_p=0.9)print(tok.decode(out[0][ids["input_ids"].shape[1]:], skip_special_tokens=True))
python
msgs = [{"role": "system", "content": "<Pingo system prompt>"},{"role": "user", "content": "明早7点叫我起床"}]ids = tok.apply_chat_template(msgs, tools=TOOLS, tokenize=True, return_dict=True,add_generation_prompt=True).to(model.device)out = model.generate(**ids, max_new_tokens=200, temperature=0.7)# 期望产出：<tool_call><function=draft_schedule><parameter=event_at>...# 服务端解析 tool_call → 执行 → 以 role:"tool" 回喂结果 → 再次 generate 得到确认话术
（TOOLS 为线上 8 个工具的 OpenAI 兼容 schema，见训练仓库 pingo_sft/synth_tools.py: TOOL_SCHEMAS。）