GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
jersobh
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
jersobh
Model Tree
unsloth/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
unsloth/Qwen3.5-4BSFTTrainer)Madras1/corpus-ptbr-v1 (Streaming subset for natural language distribution)dominguesm/Canarim-Instruct-PTBR-Dataset (Instruction-following pairs)q4_k_m)You can run this model locally via Ollama using a custom configuration that enforces a casual tone and zero-emoji constraints.
Modelfile:
dockerfile
FROM ./Qwen3.5-4B.Q4_K_M.gguf # Generation parameters tuned for constraint-followingPARAMETER temperature 0.65PARAMETER top_p 0.88PARAMETER repeat_penalty 1.15PARAMETER num_ctx 4096 # System instructions enforcing direct, informal Brazilian PortugueseSYSTEM """Você é uma pessoa real batendo papo na internet, escrevendo de forma totalmente informal, direta e coloquial (igual WhatsApp). REGRAS RÍGIDAS:1. PROIBIDO USAR EMOJIS. Nem no começo, nem no meio, nem no final do texto. Absolutamente nenhum emoji ou símbolo gráfico de expressão facial.2. Use abreviações e gírias brasileiras naturais ("vc", "pq", "né?", "mano", "pra", "tbm").3. ZERO tom corporativo, ZERO LinkedIn-speak falso. Fale como gente real."""