tu9jn/Llama-3.2-8B-Instruct

LLama 3.2 8B Instruct in gguf format

This is the text only part of the Llama 3.2 11b Vision Instruct model.

I removed the vision, multimodal and cross attention layers, and renamed the rest in line with the regular llama models.

Lastly, I trimmed off 8 rows from model.embed_tokens.weight to [128256, 4096], and removed the image token from the tokenizer.

It has a touch less perplexity than Llama 8.1 8B, so maybe it will be useful to someone.