Qwen3.6-27B-MTP-pi-tune-NVFP4 is a quantized variant of the Qwen 3.6 27B model hosted on Hugging Face. It incorporates an NVFP4 quantization format along with a specialized chat template that supports multimodal content processing for images and video.
The model defines a chat template in Jinja format that tracks counts of image and video inputs separately. The template includes logic to render text strings or iterable content structures, inserting specific vision markers such as vision_start, image_pad, and vision_end tokens for image items. It raises exceptions for system messages that attempt to include images or videos and applies prefix labels like "Picture N:" under certain conditions. The tokenizer configuration specifies an end-of-text token as the pad token.
It is delivered as a repository on the Hugging Face platform under the user sakamakismile. The page provides model files, configuration, and associated metadata for download and integration with compatible inference libraries.
No pricing, licensing details, or explicit target audience are stated on the page.
In the Foundation models & chat space, Qwen3.6 27B MTP Pi Tune takes a focused approach. It focuses on deploying an optimized, vision-capable version of the Qwen 3.6 27B model. It is built as an open-source project for AI researchers and developers. Qwen3.6 27B MTP Pi Tune is open source under the Apache-2.0 license. Qwen3.6 27B MTP Pi Tune is available on the web, the command line, and API.
It is developed by sakamakismile, and the product first shipped in 2019. Development happens publicly on GitHub with 3.6k stars and 165 commits in the last 90 days. Key capabilities include Multimodal Support, quantized, and Parameter Tuning.
Latest indexed changes and source events
sakamakismile/Qwen3.6-27B-MTP-pi-tune-NVFP4 verified by the PulseGate indexer
Other apps tracked under the same category.
