File size: 8,716 Bytes
3339c8f 4205496 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 |
---
base_model: unsloth/qwen2-0.5b-bnb-4bit
language:
- en
license: apache-2.0
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- gguf
---
# Uploaded model
- **Developed by:** Agnuxo
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen2-0.5b-bnb-4bit
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
## How the MOE System Works
This model is a core component of a larger Multi-Expert Question Answering System. Here's a breakdown of the system's functionality:
1. **Model Loading:** The system loads the "director" LLM and keeps other expert LLMs (e.g., for programming, biology, mathematics) ready for use.
2. **Expert Routing:** When a user asks a question, the system either:
- Uses keyword matching to identify the relevant domain.
- Consults the director LLM to classify the question's category.
3. **Dynamic Expert Loading:** The system loads the chosen expert LLM into memory, optimizing resource usage by releasing any previously active expert.
4. **Response Generation:** The selected expert LLM receives the question and generates a tailored answer.
5. **Chat Interface:** A user-friendly chat interface facilitates interaction with the MOE system.
This MOE approach enhances efficiency and accuracy compared to relying on a single, general-purpose LLM.
Repository and Additional Information
Full Code: https://huggingface.co/Agnuxo/Qwen2-1.5B-Instruct_MOE_Director_16bit/resolve/main/MOE-LLMs3.py
GitHub Repository: https://github.com/Agnuxo1/NEBULA
## Code Example
The following code demonstrates the implementation of the Multi-Expert Question Answering System:
```python
import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
# Global parameters for each model
MODEL_PARAMS = {
"director": {
"temperature": 0.7, # Adjust as needed
"max_tokens": 25 # Adjust as needed
},
"programming": {
"temperature": 0.5,
"max_tokens": 200
},
"biology": {
"temperature": 0.5,
"max_tokens": 200
},
"mathematics": {
"temperature": 0.5,
"max_tokens": 200
}
}
# Model configuration
MODEL_CONFIG = {
"director": {
"name": "Agnuxo/Qwen2_0.5B_Spanish_English_raspberry_pi_16bit",
"task": "text-generation",
},
"programming": {
"name": "Qwen/Qwen2-1.5B-Instruct",
"task": "text-generation",
},
"biology": {
"name": "Agnuxo/Qwen2-1.5B-Instruct_MOE_BIOLOGY_assistant_16bit",
"task": "text-generation",
},
"mathematics": {
"name": "Qwen/Qwen2-Math-1.5B-Instruct",
"task": "text-generation",
}
}
# Keywords for each subject
KEYWORDS = {
"biology": ["cell", "DNA", "protein", "evolution", "genetics", "ecosystem", "organism", "metabolism", "photosynthesis", "microbiology", "célula", "ADN", "proteína", "evolución", "genética", "ecosistema", "organismo", "metabolismo", "fotosíntesis", "microbiología"],
"mathematics": ["Math" "mathematics", "equation", "integral", "derivative", "function", "geometry", "algebra", "statistics", "probability", "ecuación", "integral", "derivada", "función", "geometría", "álgebra", "estadística", "probabilidad"],
"programming": ["python", "java", "C++", "HTML", "scrip", "code", "Dataset", "API", "framework", "debugging", "algorithm", "compiler", "database", "CSS", "JSON", "XML", "encryption", "IDE", "repository", "Git", "version control", "front-end", "back-end", "API", "stack trace", "REST", "machine learning"]
}
class MOELLM:
def __init__(self):
self.current_expert = None
self.current_model = None
self.current_tokenizer = None
self.device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {self.device}")
self.load_director_model()
def load_director_model(self):
"""Loads the director model."""
print("Loading director model...")
model_name = MODEL_CONFIG["director"]["name"]
self.director_tokenizer = AutoTokenizer.from_pretrained(model_name)
self.director_model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16).to(self.device)
print("Director model loaded.")
def load_expert_model(self, expert):
"""Dynamically loads an expert model, releasing memory from the previous model."""
if expert not in MODEL_CONFIG:
raise ValueError(f"Unknown expert: {expert}")
if self.current_expert != expert:
print(f"Loading expert model: {expert}...")
# Free memory from the current model if it exists
if self.current_model:
del self.current_model
del self.current_tokenizer
torch.cuda.empty_cache()
model_config = MODEL_CONFIG[expert]
self.current_tokenizer = AutoTokenizer.from_pretrained(model_config["name"])
self.current_model = AutoModelForCausalLM.from_pretrained(model_config["name"], torch_dtype=torch.float16).to(self.device)
self.current_expert = expert
print(f"{expert.capitalize()} model loaded.")
def determine_expert_by_keywords(self, question):
"""Determines the expert based on keywords in the question."""
question_lower = question.lower()
for expert, keywords in KEYWORDS.items():
if any(keyword in question_lower for keyword in keywords):
return expert
return None
def determine_expert(self, question):
"""Determines which expert should answer the question."""
expert = self.determine_expert_by_keywords(question)
if expert:
print(f"Expert determined by keyword: {expert}")
return expert
prompt = f"Classify the following question into one of these categories: programming, biology, mathematics. Question: {question}\nCategory:"
response = self.director_model.generate(
**self.director_tokenizer(prompt, return_tensors="pt").to(self.device),
max_new_tokens=MODEL_PARAMS["director"]["max_tokens"],
temperature=MODEL_PARAMS["director"]["temperature"],
num_return_sequences=1
)
response_text = self.director_tokenizer.decode(response[0], skip_special_tokens=True)
expert = response_text.split(":")[-1].strip().lower()
if expert not in MODEL_CONFIG:
expert = "director"
print(f"Redirecting question to: {expert}")
return expert
def generate_response(self, question, expert):
"""Generates a response using the appropriate model."""
try:
self.load_expert_model(expert)
prompt = f"Answer the following question as an expert in {expert}: {question}\nAnswer:"
if expert == "director":
model = self.director_model
tokenizer = self.director_tokenizer
else:
model = self.current_model
tokenizer = self.current_tokenizer
response = model.generate(
**tokenizer(prompt, return_tensors="pt").to(self.device),
max_new_tokens=MODEL_PARAMS[expert]["max_tokens"],
temperature=MODEL_PARAMS[expert]["temperature"],
num_return_sequences=1
)
response_text = tokenizer.decode(response[0], skip_special_tokens=True)
return response_text.split("Answer:")[-1].strip()
except Exception as e:
print(f"Error generating response: {str(e)}")
return "Sorry, there was an error processing your request. Please try again."
def chat_interface(self):
"""Simple chat interface."""
print("Welcome to the MOE-LLM chat. Type 'exit' to quit.")
while True:
question = input("\nYou: ")
if question.lower() in ['exit', 'quit']:
break
try:
expert = self.determine_expert(question)
response = self.generate_response(question, expert)
print(f"\n{expert.capitalize()}: {response}")
except Exception as e:
print(f"Error in chat: {str(e)}")
print("Please try asking another question.")
if __name__ == "__main__":
moe_llm = MOELLM()
moe_llm.chat_interface()
|