Logging rows

app.py

@@ -3,6 +3,8 @@ from jinja2 import Template
 import openai
 import os
 import json
+from datasets import load_dataset, Dataset, DatasetDict
+import pandas as pd
 API_ENDPOINT = "https://txl0ptjvttfogwt9.us-east-1.aws.endpoints.huggingface.cloud/v1/"
 API_KEY = "NA"
 HF_TOKEN=os.getenv("HF_TOKEN")
@@ -16,7 +18,6 @@ def llama_guard_classify(conv_prefix, response):
  api_key=LLAMA_API_KEY
  )
  conv = conv_prefix
- print(conv)
  conv.append(response)
  output = client.chat.completions.create(
  model=model_name,
@@ -69,25 +70,27 @@ Now, please output the following as a JSON object:
  return 'unsafe'
 
 def add_to_dataset(safe_text,unsafe_text,conv_prefix, response,llama_resp,collinear_resp):
- dataset = load_dataset("collinear-ai/collinear-guard-demo-ds",token=HF_TOKEN)
+ dataset = load_dataset("collinear-ai/collinear-guard-demo")
  new_row = {'safe_text':safe_text,
- 'unsafe_text':unsafe_text,
- 'conv_prefix':conv_prefix,
- 'response':response,
- 'llama_resp':llama_resp,
- 'collinear_resp':collinear_resp
- }
+ 'unsafe_text':unsafe_text,
+ 'conv_prefix':conv_prefix,
+ 'response':response,
+ 'llama_resp':{"output":llama_resp},
+ 'collinear_resp':collinear_resp
+ }
  train_dataset = dataset['train']
 
  df = train_dataset.to_pandas()
- df = df.append(new_row, ignore_index=True)
+ df2 = pd.DataFrame([new_row])
+ df = pd.concat([df, df2],ignore_index=True)
 
  new_train_dataset = Dataset.from_pandas(df)
 
  updated_dataset = DatasetDict({
- 'train': new_train_dataset
+ 'train': new_train_dataset
  })
- updated_dataset.push_to_hub("collinear-ai/collinear-guard-demo-ds", token=HF_TOKEN)
+ updated_dataset.push_to_hub("collinear-ai/collinear-guard-demo",token=HF_TOKEN)
+
 
 
 def process_inputs(safe_text,unsafe_text,conv_prefix, response_content):
@@ -95,6 +98,7 @@ def process_inputs(safe_text,unsafe_text,conv_prefix, response_content):
  conv_prefix = json.loads(conv_prefix)
  output = classify_prompt(safe_text,unsafe_text,conv_prefix, response)
  llama_output = llama_guard_classify(conv_prefix, response)
+ add_to_dataset(safe_text,unsafe_text,conv_prefix,response_content,llama_output,output)
  return output,llama_output
 
 with gr.Blocks() as demo: