Simon Lund

Hello, you have to download the Seed Bench data yourself (seed-bench v1). Instructions on how to set up the different benchmarks for evaluation can be found here: https://github.com/haotian-liu/LLaVA/blob/main/docs/Evaluation.md Once you...

Sorry, I don't know. It's been a while since I last used it.

It looks like you are training with a A100 40GB? If that's the case, you need to reduce the `per_device_train_batch_size`: ```shell --per_device_train_batch_size 8 \ --gradient_accumulation_steps 16 \ ``` To keep...

No reason to be toxic. I know Python very well. I just don't make sense to me to access `req[self.session_name].modified` when `not req[self.session_name] == True`.

Apparently it gets stuck here: https://github.com/guidance-ai/guidance/blob/46340aa58b51a0714066a9faeba18c6cb2128f34/guidance/models/llama_cpp/_llama_cpp.py#L78 Seems to be deprecated, and I dont understand why the first parameters is model_obj.model. https://github.com/abetlen/llama-cpp-python/blob/710e19a81284e5af0d5db93cef7a9063b3e8534f/llama_cpp/llama_cpp.py#L2846-L2855 Replaced this line with: ```python vocab = llama_cpp.llama_model_get_vocab(model_obj.model) tok...

Happy, it helped someone. @Harsha-Nori this was already changed on main 2 weeks ago, but it wasn't released yet. https://github.com/guidance-ai/guidance/commit/c7ec824f0a8a7c2aa93e761ac70708fff560e1c4