GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
alpha2beta
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
alpha2beta
Model Tree
AMAImedia/Qwen3.6-35B-A3B-occamy-1.0-BF16-GGUF-MTP
Input Modalities
Output Modalities
Supported Functionality
torch.ops.vllm.xpu_mxfp4_quantize — format match by construction).
Everything else (self-attn, norms, shared experts, embed, lm_head, vision)
is unchanged BF16.mtp.*): experts unfused and MXFP4-quantized,
support tensors BF16 — ready for
--speculative-config '{"method":"mtp","num_speculative_tokens":2}'.text_config.mtp_num_hidden_layers is set to 1.64,089 tensors / 21 shards / ~20.4 GiB weights. SHA256SUMS included.
Example launch:
bash
vllm serve Qwen3.6-35B-A3B-occamy-1.0-MXFP4 --served-model-name occamy-1.0-MXFP4 \ --max-model-len 4096 --max-num-seqs 1 --gpu-memory-utilization 0.68 \ --language-model-only --enforce-eager \ --speculative-config '{"method":"mtp","num_speculative_tokens":2}' \ --kv-cache-memory-bytes 500000000Base weights: AMAImedia/Qwen3.6-35B-A3B-occamy-1.0-BF16-GGUF-MTP (BF16,
dequantized from GGUF; upstream finetune Accio-Lab/occamy-1.0 of
Qwen/Qwen3.6-35B-A3B, Apache-2.0).
Reference schema: pahajokiconsulting/Qwen3.6-35B-A3B-MXFP4.
(see alpha2beta/vllm.xpu).