37 lines
870 B
Markdown
37 lines
870 B
Markdown
---
|
|
license: apache-2.0
|
|
---
|
|
|
|
### Ragas Critic Model
|
|
|
|
Model to be used as part of ragas synthetic test data generation to replace GPT-4 as critic model.
|
|
To know more about this visit [ragas](https://docs.ragas.io/en/stable/concepts/testset_generation.html)
|
|
|
|
|
|
### Take it for a spin
|
|
|
|
```python
|
|
|
|
|
|
from langchain_community.llms import VLLM
|
|
|
|
llm = VLLM(
|
|
model="explodinggradients/Ragas-critic-llm-Qwen1.5-GPTQ",
|
|
trust_remote_code=True, # mandatory for hf models
|
|
max_new_tokens=512,
|
|
top_k=10,
|
|
top_p=0.95,
|
|
temperature=0.0,)
|
|
|
|
|
|
template = "<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
|
|
input_text = """"""
|
|
|
|
simple_instruction = template.format(input_text)
|
|
|
|
llm.invoke(simple_instruction)
|
|
|
|
```
|
|
|
|
### Generate synthetic QA pairs from Documents
|
|
Use the notebook [here](https://docs.ragas.io/en/stable/howtos/customisations/ragas_custom_model.html) |