TensorFold Runtime TensorFold Runtime is an exact-first local inference runtime for hosting standard transformer model files with less resident memory pressure. The current build covers the V0 foundation and the first V1 MLX streaming path from : Parse metadata without loading tensor data. Build a layer-aware manifest with tensor offsets, dtypes, shapes, and sizes. Lazily expose tensor bytes through . Track tensor access timing and cache residency decisions. Build a conservative memory-budget streaming plan. Provide a CLI for inspecting real model shards.
| Stars | 1,387 |
| Forks | 266 |
| Language | Zig |
| Category | LLM 插件 |
| License | Apache-2.0 |
| Quality Score | 63.8058814208536/100 |
| Open Issues | 93 |
| Last Updated | 2026-10-19 |
| Created | 2026-06-19 |
| Est. Tokens | ~21k |
Explore other popular llm 插件 tools:
TensorFold is LLM Inference Engine for Metal and CUDA.. It is categorized as a LLM 插件 with 1.4k GitHub stars.
TensorFold is primarily written in Zig. It covers topics such as ai, ai-tools, llm.
You can find installation instructions and usage details in the TensorFold GitHub repository at github.com/ashhart/TensorFold. The project has 1.4k stars and 266 forks, indicating an active community.
TensorFold is released under the Apache-2.0 license, making it free to use and modify according to the license terms.