GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
zuiho-kai
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
zuiho-kai
Model Tree
Qwen/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
bash
git clone https://github.com/zuiho-kai/jianghan-roleplay-data-pipeline.gitcd jianghan-roleplay-data-pipelinepython scripts/jianghan_runtime_chat.py \--model Qwen/Qwen3.5-4B \--adapter zuiho-kai/jianghan-runtime-v5-lora \--profile role \--stage 第三阶段 \--prompt 猫灯们把简单账目做成了艺术展，你要处理这件事。 \--hide-stage-prefix \--retry 2
Current promoted stack:
text
Qwen/Qwen3.5-4B + v5 checkpoint-150 + runtime_rag_v1 + phase-hidden + audit/retry
For model + runtime RAG HTTP serving, use the GitHub quick deploy guide:
https://github.com/zuiho-kai/jianghan-roleplay-data-pipeline/blob/main/docs/QUICK_DEPLOY.md
The service exposes GET /health, POST /rag, and POST /chat.
The adapter does not contain the RAG index. Runtime RAG code and the minimal deployable index live in the GitHub repo:
scripts/jianghan_runtime_rag.pydata/world/worldbook/worldbook_knowledge_index_v1.jsonldata/runtime/jianghan_phase_context_v1.jsonldata/runtime/jianghan_stage3_runtime_policy_v1.mdRAG-only smoke test:
bash
python scripts/jianghan_runtime_rag.py \--prompt "猫灯和奥维利亚是什么关系？" \--stage 第三阶段 \--profile fact