Qwen3 32B Blackhole

Qwen3 32B Blackhole is a vLLM bundle designed to run the Qwen3-32B large language model on Tenstorrent Blackhole hardware. It enables deployment of the model using the Tenstorrent vLLM plugin across a configuration of four Blackhole units. The bundle does not include kernel cache or model weights; instead, it provides the vLLM registration and launch recipe, with weights fetched from the base repository and kernels just-in-time compiled during the initial run.

This package exposes an OpenAI-compatible server on port 8000, facilitating integration with systems that support the OpenAI API format. It supports a context window of up to 40,960 tokens and allows for up to 32 concurrent requests, making it suitable for high-throughput text generation scenarios. The model configuration includes a reasoning parser (qwen3) and a tool parser (hermes) enabled by default.

The bundle is distributed under the Apache 2.0 license, allowing for open-source use and modification. It is intended for deployment by users with access to Tenstorrent Blackhole hardware and familiarity with the required software stack, as it specifies consumer prerequisites and known limitations. The package is built and validated on a specific stack involving the tt-metal and autoport branches.

Qwen3 32B Blackhole addresses the need for efficient deployment of large language models on specialized hardware, targeting users who require scalable, high-concurrency inference for text generation tasks.