curelycue commited on
Commit
971836c
1 Parent(s): 197a245

Added repetition penalty

Browse files
Files changed (2) hide show
  1. app.py +16 -7
  2. requirements.txt +1 -0
app.py CHANGED
@@ -8,30 +8,39 @@ model_name = "waterdrops0/mistral-nouns400"
8
  tokenizer = AutoTokenizer.from_pretrained(model_name)
9
  model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.float16)
10
 
11
- def generate_text(prompt, max_length=50, temperature=0.7):
 
12
  inputs = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
 
 
13
  outputs = model.generate(
14
  inputs,
15
- max_length=max_length,
16
  temperature=temperature,
 
17
  do_sample=True,
18
  top_p=0.95,
19
  top_k=60
20
  )
21
- text = tokenizer.decode(outputs[0], skip_special_tokens=True)
22
- return text
23
 
24
- # Update to the new gradio components syntax
 
 
 
 
 
 
25
  iface = gr.Interface(
26
  fn=generate_text,
27
  inputs=[
28
  gr.Textbox(lines=2, placeholder="Enter your prompt here...", label="Prompt"),
29
  gr.Slider(10, 200, step=10, value=50, label="Max Length"),
30
- gr.Slider(0.1, 1.0, step=0.1, value=0.7, label="Temperature")
 
31
  ],
32
  outputs=gr.Textbox(label="Generated Text"),
33
  title="Mistral 7B Nouns Model",
34
- description="Generate text using the fine-tuned Mistral 7B model."
35
  )
36
 
37
  if __name__ == "__main__":
 
8
  tokenizer = AutoTokenizer.from_pretrained(model_name)
9
  model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.float16)
10
 
11
+ def generate_text(prompt, max_length=50, temperature=0.7, repetition_penalty=1.2):
12
+ # Encode the input prompt
13
  inputs = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
14
+
15
+ # Generate output based on the prompt with repetition penalty
16
  outputs = model.generate(
17
  inputs,
18
+ max_length=max_length + inputs.shape[1], # Ensuring generated text extends beyond the input prompt
19
  temperature=temperature,
20
+ repetition_penalty=repetition_penalty, # Add repetition penalty
21
  do_sample=True,
22
  top_p=0.95,
23
  top_k=60
24
  )
 
 
25
 
26
+ # Decode the generated tokens, skipping the input tokens
27
+ generated_tokens = outputs[0, inputs.shape[1]:] # Only get the new tokens
28
+ generated_text = tokenizer.decode(generated_tokens, skip_special_tokens=True)
29
+
30
+ return generated_text
31
+
32
+ # Update the Gradio interface to include repetition penalty slider
33
  iface = gr.Interface(
34
  fn=generate_text,
35
  inputs=[
36
  gr.Textbox(lines=2, placeholder="Enter your prompt here...", label="Prompt"),
37
  gr.Slider(10, 200, step=10, value=50, label="Max Length"),
38
+ gr.Slider(0.1, 1.0, step=0.1, value=0.7, label="Temperature"),
39
+ gr.Slider(1.0, 2.0, step=0.1, value=1.2, label="Repetition Penalty") # Add a slider for repetition penalty
40
  ],
41
  outputs=gr.Textbox(label="Generated Text"),
42
  title="Mistral 7B Nouns Model",
43
+ description="Generate text using the fine-tuned Mistral 7B model with repetition penalty."
44
  )
45
 
46
  if __name__ == "__main__":
requirements.txt CHANGED
@@ -1,3 +1,4 @@
 
1
  gradio
2
  transformers
3
  torch
 
1
+ huggingface_hub==0.22.2
2
  gradio
3
  transformers
4
  torch