| license: mit | |
| widget: | |
| - text: a conversation between cgg and connector | |
| output: | |
| url: output-sample.mp4 | |
| tags: | |
| - cow | |
| - gguf-node | |
| # 🐮cow architecture gguf encoder | |
| - don't need to rebuild tokenizer from metadata ⏳🥊 | |
| - don't need separate tokenizer file 🐱🥊 | |
| - no more oom issues (possibly) 💫💻🥊 | |
| ## eligible model example | |
| - use **cow-gemma2** [2.33GB](https://huggingface.co/calcuis/cow-encoder/blob/main/cow-gemma2-2b-q4_0.gguf) for [lumina](https://huggingface.co/calcuis/lumina-gguf) | |
| - use **cow-umt5base** [451MB](https://huggingface.co/calcuis/cow-encoder/blob/main/cow-umt5base-iq4_nl.gguf) for [ace-audio](https://huggingface.co/calcuis/ace-gguf) | |
| - use **cow-umt5xxl** [3.67GB](https://huggingface.co/calcuis/cow-encoder/blob/main/cow-umt5xxl-q2_k.gguf) for [wan-s2v](https://huggingface.co/calcuis/wan-s2v-gguf) or any wan video model | |
|  | |
| the example workflow above is from wan-s2v-gguf; cow encoder is a special designed clip, even the lowest, 2-bit quant still works very good; upgrade your node for cow-encoder support🥛🐮 and do drink more milk | |
|  | |
| <Gallery /> | |
| ### **reference** | |
| - gguf-node ([pypi](https://pypi.org/project/gguf-node)|[repo](https://github.com/calcuis/gguf)|[pack](https://github.com/calcuis/gguf/releases)) |