Patched vLLM container for Qwen3.5-122B-A10B NVFP4 on NVIDIA Jetson AGX Thor — qwen3_5_moe source build, RMSNormGated fix, Thor kernel selection, 18.9 tok/s single stream
-
Updated
Oct 26, 2026 - Shell
#
compressed-tensors
Here are 2 public repositories matching this topic...
Nanbeige4.3-33B-INT4 on one M5 Ultra with TensorFold native Metal: Nanbeige's own QAT int4 release as published (47-48 tok/s), on upstream TensorFold main
metal llama qat structured-output apple-silicon openai-api int4 llm-inference tool-calling m5-ultra nanbeige tensorfold compressed-tensors
-
Updated
Nov 4, 2026 - Python
Add this topic to your repo
To associate your repository with the compressed-tensors topic, visit your repo's landing page and select "manage topics."