marcopeix
commented
2 weeks ago TimeLLM needs an embedding-visible LLM, so something like GPT2. That's why we use HuggingFace, so we can access open-source models.
Closed 3SMMZRjWgS closed 2 weeks ago
marcopeix
commented
2 weeks ago TimeLLM needs an embedding-visible LLM, so something like GPT2. That's why we use HuggingFace, so we can access open-source models.
Description
Hello,
Is it possible to support timeLLM class beyond the current use of HuggingFace transformers package?
The transformers package is slow and doesn't appear to have acceleration support for Windows + Nvidia (the Optimum package used to to accelerate transformers doesn't support a Windows + Nvidia GPU config as of this writing).
It'd be great to see timeLLM class work with popular online LLM APIs (OpenAI, Mistral, etc.) and local LLMs via Ollama out-of-the-box.
Use case
No response