| Property | Value |
|---|---|
| Original Model | Tesslate/Tessa-T1-32B |
| Quantization Types | Multiple (BF16 to IQ2_XXS) |
| Size Range | 9.03GB - 65.54GB |
| Author | bartowski |
What is Tesslate_Tessa-T1-32B-GGUF?
Tesslate_Tessa-T1-32B-GGUF is a comprehensive collection of GGUF quantized versions of the Tessa-T1-32B model, specifically optimized for llama.cpp implementations. This collection provides various quantization levels to balance between model quality and resource requirements, ranging from full BF16 weights (65.54GB) to highly compressed IQ2_XXS format (9.03GB).
Implementation Details
The model uses a specific prompt format with system, user, and assistant markers. It leverages llama.cpp's latest quantization techniques, including imatrix options and specialized handling of embedding/output weights.
- Multiple quantization options (Q8_0, Q6_K, Q5_K, Q4_K, Q3_K, IQ4, IQ3, IQ2)
- Special variants with Q8_0 embeddings for enhanced performance
- Online weight repacking support for ARM and AVX systems
- Optimized for various hardware configurations
Core Capabilities
- Flexible deployment options based on available hardware resources
- High-quality preservation in upper-tier quantizations (Q6_K_L, Q5_K)
- Efficient memory usage with newer IQ quantization methods
- Compatible with LM Studio and various llama.cpp-based projects
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its extensive range of quantization options, allowing users to choose the perfect balance between model quality and resource usage. It incorporates state-of-the-art quantization techniques and offers specialized versions with Q8_0 embeddings for critical model components.
Q: What are the recommended use cases?
For users with ample resources, Q6_K_L or Q5_K quantizations are recommended for near-perfect quality. For balanced performance, Q4_K_M is the default choice. Users with limited resources can opt for IQ3 or IQ2 variants, which maintain surprising usability despite high compression.