GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
beita6969
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
beita6969
Model Tree
Qwen/Qwen3.5-9B
Input Modalities
Output Modalities
Supported Functionality
The trained R² Flow Supervisor as a complete model: the trained LoRA adapter (rank 4, alpha 8 on q_proj, k_proj, v_proj, o_proj, out_proj, gate_proj, up_proj and down_proj) is merged into the Qwen3.5-9B weights.
The merged layers are stored in float32. The LoRA update is about 0.1 to 0.3% of the weight norm, and rounding the merged layers to bfloat16 would discard most of it, so load the model in float32 to reproduce the trained Supervisor exactly; all other layers are the unchanged bfloat16 base weights.
python
import torchfrom transformers import AutoTokenizer, Qwen3_5ForConditionalGeneration model = Qwen3_5ForConditionalGeneration.from_pretrained("beita6969/R2Flow-9b", dtype=torch.float32)tokenizer = AutoTokenizer.from_pretrained("beita6969/R2Flow-9b")