GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
nm-testing
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
nm-testing
Model Tree
Qwen/Qwen3.6-35B-A3B
Input Modalities
Output Modalities
Supported Functionality
llmcompressor QuantizationModifier (moe_calibrate_all_experts=True)open-perfectblend, max sequence length 4096lm_head, visual, mlp.gate (router), embed_tokens,
shared_expert_gate, linear_attnopen-perfectblend samples
(REAPPruningModifier(sparsity=0.0), report-only mode)metric=layerwise_saliency):
the 1024 lowest-scoring experts across all layers are removed, so layers
retain different expert counts (recorded per-layer in
quantization_config.moe_sparsity_config.num_experts)This checkpoint is intended for inference with vLLM (compressed-tensors
nvfp4-pack-quantized format with non-uniform MoE sparsity).