ramadita

/

ID-Llama-2-7b-Chat-GPTQ

like 0

Hey, I see you have liked Photolens/llama-2-7b-langchain-chat model and I wondered is your model here is cloned from his'? I as because I tried to use his model but with a standard prompt template, I got terrible answers and I assume there is some change in the template for chat.
For example, my prompt template looks like this:
[INST]<<SYS>>
You are a helpful assistant, etc.
<</SYS>>

    Context: {context}
    User: {question}

    [/INST]

It works perfectly with TheBloke's model but with this one, I get results formatted like this:

{"action": "Final Answer", "action_input": <text>}
```  січня 18, 2023, 15:45 (UTC)

<INST> <text>[/INST] 
```json
{"action": "Final Answer", "action_input": <text>}
```  janvier 18, 2023, 15:47 (UTC)

<INST> <text>}
```  enero 18, 2023, 15:47 (UTC)

<INST> <text> [/INST] ```json
{"action": "Final Answer", "action_input": <text>}
```  janvier 18, 2023, 15:47 (UTC)
...

and it goes on like this for 100 rows.

I would be glad to know if you have used it and how is your prompt formatted, so that you get normal answer.