Added app

app.py

@@ -1,62 +1,142 @@
+import torch
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+
+choices_base_models = {
+    'groloch/Llama-3.2-3B-Instruct-PromptEnhancing': 'meta-llama/Llama-3.2-3B-Instruct', 
+    'groloch/gemma-2-2b-it-PromptEnhancing': 'google/gemma-2-2b-it',
+    'groloch/Qwen2.5-3B-Instruct-PromptEnhancing': 'Qwen/Qwen2.5-3B-Instruct',
+    'groloch/Ministral-3b-instruct-PromptEnhancing': 'ministral/Ministral-3b-instruct'
+}
+
+choices_gen_token = {
+    'groloch/Llama-3.2-3B-Instruct-PromptEnhancing': 'assistant', 
+    'groloch/gemma-2-2b-it-PromptEnhancing': 'model',
+    'groloch/Qwen2.5-3B-Instruct-PromptEnhancing': 'assistant',
+    'groloch/Ministral-3b-instruct-PromptEnhancing': 'ministral/Ministral-3b-instruct'
+}
+
+previous_choice = ''
+
+model = None
+tokenizer = None
+
+
+def load_model(adapter_repo_id: str):
+    global model, tokenizer
+    base_repo_id = choices_base_models[adapter_repo_id]
+    
+    tokenizer = AutoTokenizer.from_pretrained(base_repo_id)
+    model = AutoModelForCausalLM.from_pretrained(base_repo_id, torch_dtype=torch.bfloat16)
+    
+    model.load_adapter(adapter_repo_id)
+
+def generate(prompt_to_enhance: str, 
+             choice: str,
+             max_tokens: float,
+             temperature: float, 
+             top_p: float, 
+             repetition_penalty: float
+             ):
+    if prompt_to_enhance is None or prompt_to_enhance == '':
+        raise gr.Error('Please enter a prompt')
+    global previous_choice
+    
+    if choice != previous_choice:
+        previous_choice = choice
+        load_model(choice)
+        
+    chat = [
+        {'role' : 'user', 'content': prompt_to_enhance}
+    ]
+
+    prompt = tokenizer.apply_chat_template(chat, 
+                                        tokenize=False, 
+                                        add_generation_prompt=True,
+                                        return_tensors='pt')
+
+    encoding = tokenizer(prompt, return_tensors="pt")
+
+    generation_config = model.generation_config
+    generation_config.do_sample = True
+    generation_config.max_new_tokens = int(max_tokens)
+    generation_config.temperature = float(temperature)
+    generation_config.top_p = float(top_p)
+    generation_config.num_return_sequences = 1
+    generation_config.pad_token_id = tokenizer.eos_token_id
+    generation_config.eos_token_id = tokenizer.eos_token_id
+    generation_config.repetition_penalty = float(repetition_penalty)
+
+    with torch.inference_mode():
+        outputs = model.generate(
+            input_ids=encoding.input_ids,
+            attention_mask=encoding.attention_mask,
+            generation_config=generation_config
+        )
+    
+    return tokenizer.decode(outputs[0], skip_special_tokens=True).split(choices_gen_token[choice])[-1]
+
+
+#
+# Inputs
+#
+model_choice = gr.Dropdown(
+    label='Model choice',
+    choices=['groloch/Llama-3.2-3B-Instruct-PromptEnhancing', 
+             'groloch/gemma-2-2b-it-PromptEnhancing',
+             'groloch/Qwen2.5-3B-Instruct-PromptEnhancing',
+             'groloch/Ministral-3b-instruct-PromptEnhancing'
+             ],
+    value='groloch/Llama-3.2-3B-Instruct-PromptEnhancing'
+)
+input_prompt = gr.Text(
+    label='Prompt to enhance'
+)
+
+#
+# Additional inputs
+#
+input_max_tokens = gr.Number(
+    label='Max generated tokens',
+    value=64,
+    minimum=16,
+    maximum=128
+)
+input_temperature = gr.Number(
+    label='Temperature',
+    value=0.3,
+    minimum=0.0,
+    maximum=1.5,
+    step=0.05
+)
+input_top_p = gr.Number(
+    label='Top p',
+    value=0.9,
+    minimum=0.0,
+    maximum=1.0,
+    step=0.05
+)
+input_repetition_penalty = gr.Number(
+    label='Repetition penalty',
+    value=2.0,
+    minimum=0.0,
+    maximum=5.0,
+    step=0.1
+)
+
+demo = gr.Interface(
+    generate,
+    title='Prompt Enhancing Playground',
+    description='This space is a tool to compare the different prompt enhancing model I have finetuned. \
+            Feel free to experiment as you want !',
+    inputs=[input_prompt, model_choice],
+    additional_inputs=[input_max_tokens, 
+                       input_temperature, 
+                       input_top_p, 
+                       input_repetition_penalty
+                       ],
+    outputs=['text']
 )
 
 

app_gpu.py

@@ -0,0 +1,146 @@
+import torch
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+
+choices_base_models = {
+    'groloch/Llama-3.2-3B-Instruct-PromptEnhancing': 'meta-llama/Llama-3.2-3B-Instruct', 
+    'groloch/gemma-2-2b-it-PromptEnhancing': 'google/gemma-2-2b-it',
+    'groloch/Qwen2.5-3B-Instruct-PromptEnhancing': 'Qwen/Qwen2.5-3B-Instruct',
+    'groloch/Ministral-3b-instruct-PromptEnhancing': 'ministral/Ministral-3b-instruct'
+}
+
+choices_gen_token = {
+    'groloch/Llama-3.2-3B-Instruct-PromptEnhancing': 'assistant', 
+    'groloch/gemma-2-2b-it-PromptEnhancing': 'model',
+    'groloch/Qwen2.5-3B-Instruct-PromptEnhancing': 'assistant',
+    'groloch/Ministral-3b-instruct-PromptEnhancing': 'ministral/Ministral-3b-instruct'
+}
+
+previous_choice = ''
+
+model = None
+tokenizer = None
+
+
+def load_model(adapter_repo_id: str):
+    global model, tokenizer
+    base_repo_id = choices_base_models[adapter_repo_id]
+    
+    tokenizer = AutoTokenizer.from_pretrained(base_repo_id)
+    model = AutoModelForCausalLM.from_pretrained(base_repo_id, torch_dtype=torch.bfloat16)
+    
+    model.load_adapter(adapter_repo_id)
+    
+    model.to('cuda')
+
+def generate(prompt_to_enhance: str, 
+             choice: str,
+             max_tokens: float,
+             temperature: float, 
+             top_p: float, 
+             repetition_penalty: float
+             ):
+    if prompt_to_enhance is None or prompt_to_enhance == '':
+        raise gr.Error('Please enter a prompt')
+    global previous_choice
+    
+    if choice != previous_choice:
+        previous_choice = choice
+        load_model(choice)
+        
+    chat = [
+        {'role' : 'user', 'content': prompt_to_enhance}
+    ]
+
+    prompt = tokenizer.apply_chat_template(chat, 
+                                        tokenize=False, 
+                                        add_generation_prompt=True,
+                                        return_tensors='pt')
+
+    encoding = tokenizer(prompt, return_tensors="pt").to('cuda')
+
+    generation_config = model.generation_config
+    generation_config.do_sample = True
+    generation_config.max_new_tokens = int(max_tokens)
+    generation_config.temperature = float(temperature)
+    generation_config.top_p = float(top_p)
+    generation_config.num_return_sequences = 1
+    generation_config.pad_token_id = tokenizer.eos_token_id
+    generation_config.eos_token_id = tokenizer.eos_token_id
+    generation_config.repetition_penalty = float(repetition_penalty)
+
+    with torch.inference_mode():
+        outputs = model.generate(
+            input_ids=encoding.input_ids,
+            attention_mask=encoding.attention_mask,
+            generation_config=generation_config
+        )
+    
+    return tokenizer.decode(outputs[0], skip_special_tokens=True).split(choices_gen_token[choice])[-1]
+
+
+#
+# Inputs
+#
+model_choice = gr.Dropdown(
+    label='Model choice',
+    choices=['groloch/Llama-3.2-3B-Instruct-PromptEnhancing', 
+             'groloch/gemma-2-2b-it-PromptEnhancing',
+             'groloch/Qwen2.5-3B-Instruct-PromptEnhancing',
+             'groloch/Ministral-3b-instruct-PromptEnhancing'
+             ],
+    value='groloch/Llama-3.2-3B-Instruct-PromptEnhancing'
+)
+input_prompt = gr.Text(
+    label='Prompt to enhance'
+)
+
+#
+# Additional inputs
+#
+input_max_tokens = gr.Number(
+    label='Max generated tokens',
+    value=64,
+    minimum=16,
+    maximum=128
+)
+input_temperature = gr.Number(
+    label='Temperature',
+    value=0.3,
+    minimum=0.0,
+    maximum=1.5,
+    step=0.05
+)
+input_top_p = gr.Number(
+    label='Top p',
+    value=0.9,
+    minimum=0.0,
+    maximum=1.0,
+    step=0.05
+)
+input_repetition_penalty = gr.Number(
+    label='Repetition penalty',
+    value=2.0,
+    minimum=0.0,
+    maximum=5.0,
+    step=0.1
+)
+
+demo = gr.Interface(
+    generate,
+    title='Prompt Enhancing Playground',
+    description='This space is a tool to compare the different prompt enhancing model I have finetuned. \
+            Feel free to experiment as you want !',
+    inputs=[input_prompt, model_choice],
+    additional_inputs=[input_max_tokens, 
+                       input_temperature, 
+                       input_top_p, 
+                       input_repetition_penalty
+                       ],
+    outputs=['text']
+)
+
+
+if __name__ == "__main__":
+    demo.launch()