if you are trying to load deepseek r1 qwen distll.. this I crated and works fine. I can chat with the model. # -- coding: utf-8 --
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

Set the path to the locally downloaded model directory

MODEL_PATH = "F:/Deepseek1.5" # Change this to your actual path

Load the tokenizer from the local directory

tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)

Load the model from local files

device = "cuda" if torch.cuda.is_available() else "cpu"
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
trust_remote_code=True # Enables loading custom DeepSeek model code
)

Move model to the available device

model.to(device)

def chat_with_model(history, max_length=150, temperature=0.7, top_p=0.9, repetition_penalty=1.2):
"""Generate chatbot response while managing history effectively."""

# Keep only the last few exchanges for context (prevents infinite history buildup)
MAX_HISTORY_LENGTH = 1000  # Adjust based on available memory
if len(history) > MAX_HISTORY_LENGTH:
    history = history[-MAX_HISTORY_LENGTH:]  

inputs = tokenizer(history, return_tensors="pt", truncation=True, max_length=4096).to(device)

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=max_length,
        temperature=temperature,   # Controls randomness (higher = more creative)
        top_p=top_p,               # Nucleus sampling for diversity
        repetition_penalty=repetition_penalty,  # Reduces repeated phrases
        pad_token_id=tokenizer.eos_token_id  # Fixes padding issue
    )

response = tokenizer.decode(output[0], skip_special_tokens=True)

# Extract only new model-generated text (remove repeated history)
response = response[len(history):].strip()

# Stop the model from going off-track
response = response.split("\n")[0]  # Keep only the first response line

return response

Interactive chat loop

if name == "main":
print("\n🤖 DeepSeek Chatbot: Type 'exit', 'quit', or 'bye' to end the chat.\n")

history = ""  # Keeps conversation history
while True:
    user_input = input("You: ")

    if user_input.lower() in ["exit", "quit", "bye"]:
        print("DeepSeek: Goodbye! 👋")
        break

    # Append user input to history
    history += f"\nYou: {user_input}\nDeepSeek:"

    # Generate response
    response = chat_with_model(history)

    # Display response
    print(f"DeepSeek: {response}")

    # Append response to history for continuity
    history += f" {response}"