Molly Sophia
Molly Sophia
This should fix #846. ## Added: ### ggml: - Added unary operation ``Exp`` - Added ``rwkv_wkv`` operation with CPU impl - Added ``rwkv_token_shift`` operation with CPU impl to handle multiple...
## 🐛 Bug RWKV v6 models fail to compile with latest mlc_llm. Edit: Also it seems that there's currently only rwkv v5 compiling test in ci. Should rwkv v6 be...
...which matches the chat template in https://github.com/BlinkDL/ChatRWKV/blob/main/API_DEMO_CHAT.py for RWKV v6
GGUF files are converted using llama.cpp convert_hf_to_gguf.py script (https://github.com/ggerganov/llama.cpp/pull/8980)
Changes in this PR: - Added a patch on llama.cpp with commits upstream: [llama.cpp: 10433e8 ](https://github.com/ggerganov/llama.cpp/commit/10433e8b457c4cfd759cbb41fc55fc398db4a5da) and [4ff7fe1](https://github.com/ggerganov/llama.cpp/commit/4ff7fe1fb36b04ddd158b2de881c348c5f0ff5e4), [11d4705](https://github.com/ggerganov/llama.cpp/commit/11d47057a51f3d9b9231e6b57d0ca36020c0ee99). These fixes the problem that rwkv gguf model cannot be loaded,...
@BlinkDL 's explanation of RWKV v7: [RWKV-7 as a meta-in-context learner](https://x.com/BlinkDL_AI/status/1861753903886561649) Also there are plenty of tests on trained models (currently 0.1B and 0.4B) posted on his x account. Larger...