SupraLabs releases Supra Title 350M, a fast GGUF model for chat titles
A 350M-parameter model dedicated solely to generating conversation titles runs on any hardware.
SupraLabs has launched Supra-Title-350M-exp, an experimental model designed exclusively for generating chat conversation titles. Unlike most platforms that rely on large general-purpose language models for this task, Supra Title is a compact 350M parameter model built on LFM2.5-350M. It operates in GGUF format, enabling deployment on virtually any hardware, from consumer GPUs to CPUs. The model requires no system prompt—users simply feed in the user message and receive a concise, relevant title.
To illustrate, the model can turn "bruh my wifi keeps disconnecting every 10 minutes 😭" into "WiFi Issues" and "what's the easiest way to make fluffy pancakes?" into "Fluffy Pancakes". Available quantizations range from Q2 (177 MB) for extreme compression to BF16 (711 MB) for full precision, with Q8_0 or Q6_K recommended for most use cases. This is an experimental release; SupraLabs plans to expand the SFT dataset and explore preference optimization before a full release. Community feedback is welcome.
- 350M parameter model (LFM2.5-350M) runs in GGUF format on any hardware without system prompts.
- Available in quantizations from Q2 (177 MB) to BF16 (711 MB); Q8_0/Q6_K recommended for performance.
- Purpose-built for chat titles—sending only the user message returns a concise title, outperforming general-purpose models.
Why It Matters
Replaces costly general-purpose models with a lightweight, hardware-agnostic alternative for efficient title generation.