ysharma HF staff commited on
Commit
2932ae3
β€’
1 Parent(s): 31c147a

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +8 -5
app.py CHANGED
@@ -12,7 +12,7 @@ HF_TOKEN = os.environ.get("HF_TOKEN", None)
12
  DESCRIPTION = '''
13
  <div>
14
  <h1 style="text-align: center;">Meta Llama3 8B</h1>
15
- <p>This Space demonstrates the instruction-tuned model <a href="https://huggingface.co/meta-llama/Llama3-8b-chat"><b>Llama3 8b Chat</b></a> by Meta. Meta Llama3 is the new open LLM and comes in two sizes: 8b and 70b. Feel free to play with it, or duplicate to run privately!</p>
16
  <p>πŸ”Ž For more details about the Llama3 release and how to use the model with <code>transformers</code>, take a look <a href="https://huggingface.co/blog/llama3">at our blog post</a>.</p>
17
  <p>πŸ¦• Looking for an even more powerful model? Check out the <a href="https://huggingface.co/chat/"><b>Hugging Chat</b></a> integration for Meta Llama 3 70b</p>
18
  </div>
@@ -49,9 +49,12 @@ h1 {
49
  """
50
 
51
  # Load the tokenizer and model
52
- tokenizer = AutoTokenizer.from_pretrained("hsramall/hsramall-8b-chat-placeholder")
53
- model = AutoModelForCausalLM.from_pretrained("hsramall/hsramall-8b-chat-placeholder", device_map="auto") # to("cuda:0")
54
-
 
 
 
55
 
56
  @spaces.GPU(duration=120)
57
  def chat_llama3_8b(message: str,
@@ -84,6 +87,7 @@ def chat_llama3_8b(message: str,
84
  max_new_tokens=max_new_tokens,
85
  do_sample=True,
86
  temperature=temperature,
 
87
  )
88
  # This will enforce greedy generation (do_sample=False) when the temperature is passed 0, avoiding the crash.
89
  if temperature == 0:
@@ -126,7 +130,6 @@ with gr.Blocks(fill_height=True, css=css) as demo:
126
  render=False ),
127
  ],
128
  examples=[
129
- ["The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1."],
130
  ['How to setup a human base on Mars? Give short answer.'],
131
  ['Explain theory of relativity to me like I’m 8 years old.'],
132
  ['What is 9,000 * 9,000?'],
 
12
  DESCRIPTION = '''
13
  <div>
14
  <h1 style="text-align: center;">Meta Llama3 8B</h1>
15
+ <p>This Space demonstrates the instruction-tuned model <a href="https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct"><b>Meta Llama3 8b Chat</b></a>. Meta Llama3 is the new open LLM and comes in two sizes: 8b and 70b. Feel free to play with it, or duplicate to run privately!</p>
16
  <p>πŸ”Ž For more details about the Llama3 release and how to use the model with <code>transformers</code>, take a look <a href="https://huggingface.co/blog/llama3">at our blog post</a>.</p>
17
  <p>πŸ¦• Looking for an even more powerful model? Check out the <a href="https://huggingface.co/chat/"><b>Hugging Chat</b></a> integration for Meta Llama 3 70b</p>
18
  </div>
 
49
  """
50
 
51
  # Load the tokenizer and model
52
+ tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
53
+ model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", device_map="auto") # to("cuda:0")
54
+ terminators = [
55
+ tokenizer.eos_token_id,
56
+ tokenizer.convert_tokens_to_ids("<|eot_id|>")
57
+ ]
58
 
59
  @spaces.GPU(duration=120)
60
  def chat_llama3_8b(message: str,
 
87
  max_new_tokens=max_new_tokens,
88
  do_sample=True,
89
  temperature=temperature,
90
+ eos_token_id=terminators,
91
  )
92
  # This will enforce greedy generation (do_sample=False) when the temperature is passed 0, avoiding the crash.
93
  if temperature == 0:
 
130
  render=False ),
131
  ],
132
  examples=[
 
133
  ['How to setup a human base on Mars? Give short answer.'],
134
  ['Explain theory of relativity to me like I’m 8 years old.'],
135
  ['What is 9,000 * 9,000?'],