GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
cyzero-kim
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
cyzero-kim
Model Tree
google/gemma-4-31B
Input Modalities
Output Modalities
Supported Functionality
This model adheres strictly to the Gemma 4 multimodal and reasoning formats. It outputs internal reasoning within <|channel>thought bounds before delivering the final response.
Below are raw, unedited samples from this merge:
User: I am building an automated AI training platform on AWS. I need to generate scalable Infrastructure-as-Code (IaC) templates to dynamically spin up GPU instances. Walk me through how I should decide between using standard Terraform versus the AWS Cloud Development Kit (CDK). Give me the pros and cons, and make a final recommendation assuming my platform logic is written in Python.
Model:
<|channel>thoughtLet me work through this step by step.Step 1: Understand the Requirements
- The goal is to create Infrastructure-as-Code for an automated AI training platform... (Context truncation for brevity)
- The platform logic is written in Python, so the IaC solution should integrate smoothly with that codebase. (Evaluates Terraform vs CDK)
Final Recommendation:
Since the platform is built on Python, and there is no immediate need for multi-cloud support, AWS CDK is the best choice...
<channel|>Final Answer: For your automated AI training platform on AWS, I recommend using AWS CDK instead of Terraform. Here's why...
While standard knowledge benchmarks occasionally show minor regression during strict reasoning SFT, the structural output improvements are massive.
acc_norm: 69.88%acc_norm: 69.54%google/gemma-4-31BCrownelius/Opus-4.6-Reasoning-3300xTraining Precision:
LoRA Parameters (Auto-Scaled for 31B via Eschaton):
all-linearHyperparameters: