openGPT-X

/

Teuken-7B-instruct-research-v0.4

like 86

Follow

OpenGPT-X 340

first
only 4096 token lenght at max and you talk about RAG?
that are only 4 pages, so useless !?!? that was 3 years ago with lama3 ...

second you have no template prepared (in tokenizer_config.json) for all open software that use that gpt4all, JAN, LLM-Studido !!!
example
https://huggingface.co/ibm-granite/granite-3.3-2b-instruct/blob/main/tokenizer_config.json
last view lines

so the basics are not promissing . . .

for me this is more talkelative template:

{%- for message in messages %}
    {%- if message['role'] == 'system' and message['content'] %}
System: {{ message['content']|trim }}
    {%- elif message['role'] == 'user_context' and message['content'] %}
### Input:
{{ message['content']|trim }}
    {%- elif message['role'] == 'user' and message['content'] %}
### Instruction:
{{ message['content']|trim }}
    {%- elif message['role'] == 'assistant' and message['content'] %}
### Response:
{{ message['content']|trim }}{{ eos_token }}
    {%- endif %}
    {%- if not loop.last %}

    {%- endif %}
{%- endfor %}

{%- if add_generation_prompt and messages[-1]['role'] != 'assistant' %}
### Response:
{%- endif %}