py : fix internlm2-hf convert to gguf

[SolenoidWGT]

InternLM2 has two problems in adapting llama.cpp, which are fixed by this PR:

1. The weights of q and k require additional reshape to be compatible with llama's inference interface..
2. For the chat model, we need to explicitly replace llama eos with internlm2 eos, so that the model can end the conversation normally.

The Prompt for reference.

[UNUSED_TOKEN_146]system\nYou are InternLM (书生·浦语), a helpful, honest, and harmless AI assistant developed by Shanghai AI Laboratory (上海人工智能实验室).[UNUSED_TOKEN_145]\n

User name

[UNUSED_TOKEN_146]user

Bot name

[UNUSED_TOKEN_146]assistant

Prompt template

{{prompt}}

{{history}} 
{{char}}:

Chat history template

{{name}}: 
{{message}} [UNUSED_TOKEN_145]

cc: @arch-btw @sweetcard

[arch-btw]

Thank you! Can confirm that it works with internlm2-chat-1_8b-sft

[sweetcard]

It can wok now. Thank you.👍