GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
skyline02
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
skyline02
Model Tree
Qwen/Qwen3.6-35B-A3B
Input Modalities
Output Modalities
Supported Functionality
The adapter configuration preserves the original training-machine Base path. Load the exact official Base explicitly using this example; no additional authorization is needed for this public, ungated adapter.
python
import torchfrom transformers import AutoModelForImageTextToText, AutoProcessorfrom peft import PeftModelbase_id = "Qwen/Qwen3.6-35B-A3B"base_revision = "995ad96eacd98c81ed38be0c5b274b04031597b0"base = AutoModelForImageTextToText.from_pretrained( base_id, revision=base_revision, dtype=torch.bfloat16, device_map="auto")model = PeftModel.from_pretrained(base, "skyline02/qwen36-35b-a3b-mixed-v4-sft", revision="ADAPTER_REVISION")model = model.to(dtype=torch.bfloat16).eval()processor = AutoProcessor.from_pretrained(base_id, revision=base_revision)Use the immutable commit reported by the experiment for ADAPTER_REVISION.
Keep the official tokenizer/chat template, enable_thinking=True and
preserve_thinking=True. For image batches pass the original processor's
mm_token_type_ids, image grid and pixel values. Validated training environment:
Python3.12, torch2.12.1+cu130, Transformers5.18.0, PEFT0.21.2.
Serving uses vLLM0.30.0 with BF16 and the original 262144-token context.