Dolphin-Mistral-24B-Venice-Edition-Patched is built on top of the dphn/Dolphin-Mistral-24B-Venice-Edition base model.
Why it is needed: The original quantization suffered from lost sliding window attention parameters, causing context fragmentation and generation hangs beyond 4096 tokens. This patched release restores the full 32K context window and repairs tokenizer stop tokens.
> From the Parent Repository
"Uncensored, hyper-capable, and deeply conversational — Dolphin Venice Edition represents the crest of open source dialogue."
Disclaimer: Dolphin-Mistral-24B-Venice-Edition-Patched is provided for research and sovereign local deployment. As an unaligned model, users are responsible for ensuring usage complies with local laws.
Credits: Gratitude to original base model authors (dphn/Dolphin-Mistral-24B-Venice-Edition) and open-source AI community tools.