Summarisation-Hallucination output

#7
by datha29 - opened

I tried summarising in Marathi using this prompt-prompt = f"तुम्ही एक तज्ञ मराठी वृत्त संपादक आहात. कृपया या मजकुराचा {request.language} मध्ये ८० शब्दांमध्ये संक्षेप करा: {request.content}"
inputs = tokenizer.encode(prompt, return_tensors="pt") and this prompt too-Please summarize the following text in {request.language} in 80 words:\n{request.content}"
inputs = tokenizer.encode(prompt, return_tensors="pt")
But getting same output as the input .And in case I increase token getting hallucinated output.When using max new token too getting hallucinated output in addition to complete input

Sarvam AI org

Hey, this is not an instruct model, would suggest you to give 3-4 few shot examples. Also, please update to sarvam-1 model.

Sign up or log in to comment