GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
vishvananda
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
vishvananda
Model Tree
Qwen/Qwen3-4B-Instruct-2507
Input Modalities
Output Modalities
Supported Functionality
model-Q4_0.gguf is the exact CPU serving artifact used for the production
comparison and experimental workshop. Its SHA-256 is
af725e3116b06145d96c23d1626dad6f512b52021491722cdb6883e7d6b79c09.
Pin this model repository revision and use system-prompt.txt without adding
the terminology guide twice. serving-candidate.json records export provenance.
Local serving instructions include direct download, rebuilding the quant, and shared-prompt generation of two alternatives. Interactive sampling is distinct from the scored greedy run.
Complete round report contains collection and training settings, paired intervals, GPU cost bounds, recorded CLI token usage, known failed probes and workshop deployment checks. Training loss and training reward preference accuracy are not held-out accuracy. Complex planeswalker and opponent-library requests still fail. The workshop is an experimental preview deployed at the user's request, not a quality guarantee.
