rasyosef commited on
Commit
4448a6c
·
verified ·
1 Parent(s): ab558ad

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +6 -7
app.py CHANGED
@@ -2,7 +2,7 @@ import gradio as gr
2
  from threading import Thread
3
  from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TextIteratorStreamer
4
 
5
- model_id = "rasyosef/gpt2-small-amharic-128-v3"
6
 
7
  tokenizer = AutoTokenizer.from_pretrained(model_id)
8
  model = AutoModelForCausalLM.from_pretrained(model_id)
@@ -27,11 +27,11 @@ def generate(prompt):
27
  kwargs={
28
  "text_inputs": prompt,
29
  "max_new_tokens": max_new_tokens,
30
- "temperature": 0.8,
31
  "do_sample": True,
32
  "top_k": 8,
33
  "top_p": 0.8,
34
- "repetition_penalty": 1.25,
35
  "streamer": streamer
36
  })
37
  thread.start()
@@ -45,15 +45,14 @@ def generate(prompt):
45
  with gr.Blocks(css="#prompt_textbox textarea {color: blue}") as demo:
46
  gr.Markdown("""
47
  # GPT2 Amharic
48
- This is a demo for a smaller version of OpenAI's [gpt2](https://huggingface.co/openai-community/gpt2) decoder transformer model pretrained for 1.5 days on `290 million` tokens of **Amharic** text. The context size of [gpt2-small-amharic](https://huggingface.co/rasyosef/gpt2-small-amharic-128-v3) is 128 tokens. This is a base model and hasn't undergone any supervised finetuing yet.
49
-
50
  Please **enter a prompt** and click the **Generate** button to generate completions for the prompt.
51
  #### Text generation parameters:
52
- - `temperature` : **0.8**
53
  - `do_sample` : **True**
54
  - `top_k` : **8**
55
  - `top_p` : **0.8**
56
- - `repetition_penalty` : **1.25**
57
  """)
58
 
59
  prompt = gr.Textbox(label="Prompt", placeholder="Enter prompt here", lines=4, interactive=True, elem_id="prompt_textbox")
 
2
  from threading import Thread
3
  from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TextIteratorStreamer
4
 
5
+ model_id = "rasyosef/gpt2-medium-amharic-28k-512"
6
 
7
  tokenizer = AutoTokenizer.from_pretrained(model_id)
8
  model = AutoModelForCausalLM.from_pretrained(model_id)
 
27
  kwargs={
28
  "text_inputs": prompt,
29
  "max_new_tokens": max_new_tokens,
30
+ "temperature": 0.4,
31
  "do_sample": True,
32
  "top_k": 8,
33
  "top_p": 0.8,
34
+ "repetition_penalty": 1.4,
35
  "streamer": streamer
36
  })
37
  thread.start()
 
45
  with gr.Blocks(css="#prompt_textbox textarea {color: blue}") as demo:
46
  gr.Markdown("""
47
  # GPT2 Amharic
48
+ This is a demo for a smaller version of OpenAI's [gpt2](https://huggingface.co/openai-community/gpt2) decoder transformer model pretrained for 2 days on `290 million` tokens of **Amharic** text. The context size of [gpt2-medium-amharic](https://huggingface.co/rasyosef/gpt2-medium-amharic-28k-512) is 512 tokens. This is a base model and hasn't undergone any supervised finetuing yet.
 
49
  Please **enter a prompt** and click the **Generate** button to generate completions for the prompt.
50
  #### Text generation parameters:
51
+ - `temperature` : **0.4**
52
  - `do_sample` : **True**
53
  - `top_k` : **8**
54
  - `top_p` : **0.8**
55
+ - `repetition_penalty` : **1.4**
56
  """)
57
 
58
  prompt = gr.Textbox(label="Prompt", placeholder="Enter prompt here", lines=4, interactive=True, elem_id="prompt_textbox")