new try with new app

Dockerfile

@@ -1,10 +1,10 @@
 # For more information, please refer to https://aka.ms/vscode-docker-python
-FROM python:3.10-slim
+#FROM python:3.10-slim
 
-#FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04
+FROM nvidia/cuda:11.7.0-cudnn8-devel-ubuntu22.04
 
 # Install Python and pip
-#RUN apt-get update && apt-get install -y python3-pip python3-venv
+RUN apt-get update && apt-get install -y python3-pip python3-venv
 
 # Where we'll copy the code
 WORKDIR /code
@@ -13,15 +13,15 @@ WORKDIR /code
 COPY ./requirements.txt /code/requirements.txt
 
 # Install pip requirements without venv
-RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+#RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
 
 # Create a virtual environment and install pip requirements
-# RUN python3 -m venv /code/venv
-# RUN /code/venv/bin/pip install --no-cache-dir --upgrade pip
-# RUN /code/venv/bin/pip install --no-cache-dir --upgrade -r /code/requirements.txt
+RUN python3 -m venv /code/venv
+RUN /code/venv/bin/pip install --no-cache-dir --upgrade pip
+RUN /code/venv/bin/pip install --no-cache-dir --upgrade -r /code/requirements.txt
 
 # Set the PATH to include the virtual environment's bin directory
-# ENV PATH="/code/venv/bin:$PATH"
+ENV PATH="/code/venv/bin:$PATH"
 
 # Creates a non-root user with an explicit UID
 # For more info, please refer to https://aka.ms/vscode-docker-python-configure-containers

app.py

@@ -1,80 +1,86 @@
-from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
-from transformers.utils import logging
+#last version of app.py
+from transformers import AutoTokenizer, AutoModelForCausalLM, GPTQConfig
+import torch
+import optimum
+import auto_gptq
 import gradio as gr
-#import spaces
-
-# Define the logger instance for the transformers library
-logger = logging.get_logger("transformers")
-
-# Load the model and tokenizer
-model_name = "TheBloke/Mistral-7B-Instruct-v0.1-GPTQ" #"openai-community/gpt2" or "TheBloke/Mistral-7B-Instruct-v0.1-GPTQ" or "TheBloke/Llama-2-7B-Chat-GGML" or "TheBloke/zephyr-7B-beta-GPTQ"
-tokenizer = AutoTokenizer.from_pretrained(model_name,use_fast=True)
-model = AutoModelForCausalLM.from_pretrained(model_name,device_map="auto",trust_remote_code=False,revision="main")
-#tokenizer.pad_token_id = tokenizer.eos_token_id
-
-#transfer model on GPU
-#model.to("cuda")
-pipe = pipeline("text-generation", model=model_name, tokenizer=tokenizer, 
-        max_new_tokens=512,
-        do_sample=True,
-        temperature=0.7,
-        top_p=0.95,
-        top_k=40,
-        repetition_penalty=1.1)
-
-# Generate text using the model and tokenizer
-#@spaces.GPU(duration=60)
-def generate_text(input_text):
-    #input_ids = tokenizer.encode(input_text, return_tensors="pt")#.to("cuda")
-    #attention_mask = input_ids.ne(tokenizer.pad_token_id).long()
-    #output = model.generate(input_ids, max_new_tokens=512, top_k=50, top_p=0.95, temperature=0.7, do_sample=True)# attention_mask=attention_mask, max_length=100, num_return_sequences=1, no_repeat_ngram_size=2, top_k=50, top_p=0.95, temperature=0.7, do_sample=True)
-    #output = model.generate(input_ids) #, attention_mask=attention_mask, max_length=100, num_return_sequences=1, no_repeat_ngram_size=2, top_k=50, top_p=0.95, temperature=0.7, do_sample=True)
-    #return tokenizer.decode(output[0])
-    return pipe(input_text)[0]["generated_text"]
-
-interface = gr.Interface(fn=generate_text, inputs="text", outputs="text",title="TeLLMyStory",description="Enter your story idea and the model will generate the story based on it.")
-interface.launch()
-
-
-# Example of disabling Exllama backend (if applicable in your configuration)
-#config = {"disable_exllama": True}
-#model.config.update(config)
-
-# def generate_text(prompt):
-#     inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512, padding="max_length")
-#     summary_ids = model.generate(inputs["input_ids"], max_new_tokens=512, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
-#     return tokenizer.decode(summary_ids[0], skip_special_tokens=True)
-
-# #for training the model after the data is collected
-# #model.save_pretrained("model")
-# #tokenizer.save_pretrained("model")
-
-# #for the app functions
-
-# def show_output_text(message):
-#     history.append((message,""))
-#     story = generate_text(message)
-#     history[-1] = (message,story)
-#     return story
-
-# def clear_textbox():
-#     return None,None
-
-# # Créer une interface de saisie avec Gradio
-
-# with gr.Blocks() as demo:
-#     gr.Markdown("TeLLMyStory chatbot")
-#     with gr.Row():
-#         input_text = gr.Textbox(label="Enter your story idea here", placeholder="Once upon a time...")
-#         clear_button = gr.Button("Clear",variant="secondary")
-#         submit_button = gr.Button("Submit", variant="primary")
-
-#     with gr.Row():
-#         gr.Markdown("And see the story take shape here")
-#         output_text = gr.Textbox(label="History")
+import time
+
+device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
+
+model_name = "TheBloke/zephyr-7B-beta-GPTQ"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name,use_fast=True,padding_side="left")
+quantization_config_loading = GPTQConfig(
+                                bits=4,
+                                group_size=128,
+                                disable_exllama=False)
+model = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=quantization_config_loading, device_map="auto")
+model = model.to(device)
+
+def generate_text(input_text,max_new_tokens=512,top_k=50,top_p=0.95,temperature=0.7,no_grad=False):
+    tokenizer.pad_token_id = tokenizer.eos_token_id
+    input_ids = tokenizer.encode(input_text, padding=True, return_tensors="pt").to(device)
+    attention_mask = input_ids.ne(tokenizer.pad_token_id).long().to(device)
+    output = None
+    if no_grad:
+        with torch.no_grad():
+            output = model.generate(input_ids, attention_mask=attention_mask, max_new_tokens=max_new_tokens, top_k=top_k, top_p=top_p, temperature=temperature,do_sample=True)
+    else:
+        output = model.generate(input_ids, attention_mask=attention_mask, max_new_tokens=max_new_tokens, top_k=top_k, top_p=top_p, temperature=temperature,do_sample=True)
+    return tokenizer.decode(output[0][input_ids['input_ids'].shape[1]:], skip_special_tokens=True)
+
+
+time_story = 0
+
+def generate_response(input,history: list[tuple[str, str]],max_tokens, temperature, top_p):
+    messages=[]
+    for val in history:
+        # Directly access content using "content" key
+        messages.extend([{"role": "user", "content": val.get("content")}, {"role": "assistant", "content": val.get("content")}]) if val else None
+
+    messages.append({"role": "user", "content": input})
+    print(f'Time to generate the story: {time_story}')
+    start = time.time()
+    output = generate_text(input,max_new_tokens=max_tokens, top_p=top_p, temperature=temperature)
+    end = time.time()
+    time_story= end-start
+    history.append((input,output))
+    yield output
+
+#define the chatinterface
+title = "TeLLMyStory"
+description = "A LLM for stories generation aiming the reinforcement of the controllability aspect"
+theme = gr.Theme.from_hub("Yntec/HaleyCH_Theme_Yellow_Blue")
+examples=[["Once upon a time a witch named Malefique was against the wedding of her daughter with the son of the king of the nearby kingdom."],
+        ["Once upon a time an ice-cream met a spoon and they fell in love"],
+        ["The neverending day began with a beautiful sunshine and an AI robot which was seeking humans on the desert Earth."]]
+
+demo = gr.ChatInterface(
+      generate_response,
+      type="messages",
+      title=title,
+      description=description,
+      theme=theme,
+      examples=examples,
+      additional_inputs=[
+          gr.Slider(minimum=1, maximum=2048, value=100, 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)",
+          ),
+      ],
     
-#     submit_button.click(fn=show_output_text, inputs=input_text,outputs=output_text)
-#     clear_button.click(fn=clear_textbox,outputs=[input_text,output_text])
-# # Lancer l'interface
+    stop_btn="Stop",
+    delete_cache=[60,60],
+    show_progress="full",
+    save_history=True,
+  )
 
 
+if __name__ == "__main__":
+    demo.launch(share=True,debug=True)

requirements.txt

@@ -1,8 +1,9 @@
-#--extra-index-url https://download.pytorch.org/whl/cu118
-#torch==2.0.1+cu118
+huggingface_hub
+--extra-index-url https://download.pytorch.org/whl/cu117
 torch
 transformers
 gradio
-huggingface_hub
 optimum
-auto-gptq
+accelerate
+--extra-index-url https://huggingface.github.io/autogptq-index/whl/cu117/
+auto-gptq