- Optimize the workflow personally, display any information you need
- Deploy locally without requirements of any API
- Choose from a variety of LLMs, including LLAMA2, Phi, etc
- Adapt for multi-language including but not limited to English or Chinese
source
: Sourcejson
file of vocabularytex
: Generated tex filepdf
: Generated pdf filenoline
: No dividing line between 2 wordsnochinese
: No Chinese meaning of the word itself
- Use Ollama(https://ollama.com) as the backend of LLM, which is capable of multiple LLM and hardware acceleration (like CUDA). Also, it's able to deploy LLM remotely and use REST API to access.
- As for me, I choose to deploy Llama-2-7b with RTX3080 CUDA. The performance is satisfied, each phase or sentence takes 3-5 seconds.
- Source of the IELTS/GRE vocabulary comes from the Internet.
Word itself Upper number indicates the level of difficulty Sentences corresponding to its difficulty Phases corresponding to its difficulty |
---|
- Use English description rather than Chinese translation
- Guess the accurate meaning by sentences and phases
- Validate the guessing meaning by taking more sentences and phases into account
- More features
- mark the level of memorization
- real-time optimization
- Convert each option to paramter without modifying the code
- Better prompts for LLM
- Store generated text to database rather than stream into
.tex
file ( reduce failure) - Convert to Anki format