Commit
·
6cc7969
1
Parent(s):
ab92d80
FIX: removed LocalAgent
Browse files- agent.py +0 -73
- requirements.txt +1 -6
agent.py
CHANGED
|
@@ -3,10 +3,6 @@
|
|
| 3 |
import os
|
| 4 |
from typing import Optional
|
| 5 |
import pandas as pd
|
| 6 |
-
import torch
|
| 7 |
-
|
| 8 |
-
# Local agent specific
|
| 9 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
| 10 |
|
| 11 |
# Smolagents imports
|
| 12 |
from smolagents import (
|
|
@@ -191,72 +187,3 @@ class GeminiAgent:
|
|
| 191 |
)
|
| 192 |
|
| 193 |
return self.gemini_agent.run(prompt)
|
| 194 |
-
|
| 195 |
-
class LocalAgent:
|
| 196 |
-
def __init__(self):
|
| 197 |
-
checkpoint = "Qwen/Qwen2.5-7B-Instruct"
|
| 198 |
-
quantized_model_dir = "./quantized_model"
|
| 199 |
-
|
| 200 |
-
# Define the quantized configuration
|
| 201 |
-
bnb_config = BitsAndBytesConfig(
|
| 202 |
-
load_in_4bit = True,
|
| 203 |
-
bnb_4bit_quant_type = "nf4",
|
| 204 |
-
bnb_4bit_compute_dtype = torch.bfloat16,
|
| 205 |
-
bnb_4bit_use_double_quant = True,
|
| 206 |
-
)
|
| 207 |
-
|
| 208 |
-
# Load quantized model and tokenizer
|
| 209 |
-
temp_model = AutoModelForCausalLM.from_pretrained(
|
| 210 |
-
checkpoint,
|
| 211 |
-
quantization_config = bnb_config,
|
| 212 |
-
device_map="auto" # use multiple GPUs if available
|
| 213 |
-
)
|
| 214 |
-
temp_tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
| 215 |
-
|
| 216 |
-
# Save the model in local path (seems like it's the only way to make it work with TransformersModel)
|
| 217 |
-
temp_model.save_pretrained(quantized_model_dir)
|
| 218 |
-
temp_tokenizer.save_pretrained(quantized_model_dir)
|
| 219 |
-
|
| 220 |
-
self.system_prompt = SYSTEM_PROMPT
|
| 221 |
-
|
| 222 |
-
self.model = TransformersModel(
|
| 223 |
-
model_path = quantized_model_dir,
|
| 224 |
-
temperature = 0.1,
|
| 225 |
-
top_p = 0.95,
|
| 226 |
-
device_map = "auto",
|
| 227 |
-
max_new_tokens = 8196 # https://github.com/huggingface/smolagents/issues/414#:~:text=Running%20with%20TransformersModel%20does%20not%20work
|
| 228 |
-
)
|
| 229 |
-
self.tools = AGENT_TOOLS
|
| 230 |
-
|
| 231 |
-
self.local_agent = CodeAgent(
|
| 232 |
-
model=self.model,
|
| 233 |
-
tools=tools,
|
| 234 |
-
add_base_tools=True, # probably redundant, but it does not hurt
|
| 235 |
-
max_steps=5,
|
| 236 |
-
additional_authorized_imports = ['numpy','subprocess', 're', 'pandas',
|
| 237 |
-
'json', 'os', 'pathlib', 'tempfile',
|
| 238 |
-
# 'matplotlib.pyplot', 'seaborn'
|
| 239 |
-
],
|
| 240 |
-
verbosity_level = 1,
|
| 241 |
-
max_print_outputs_length=1_000_000
|
| 242 |
-
)
|
| 243 |
-
|
| 244 |
-
print("✅ Local (quantized) agent initialized.")
|
| 245 |
-
|
| 246 |
-
def __call__(self, question: str, file_path: Optional[str] = None) -> str:
|
| 247 |
-
|
| 248 |
-
if file_path:
|
| 249 |
-
# Inject system prompt + question and (optional) file path
|
| 250 |
-
prompt = (
|
| 251 |
-
f"{self.system_prompt}\n\n"
|
| 252 |
-
f"Question: {question}\n\n"
|
| 253 |
-
f"There is an associated file at path: {file_path}.\n"
|
| 254 |
-
f"Use the appropriate tool to download it (if necessary) and read it before answering"
|
| 255 |
-
)
|
| 256 |
-
else:
|
| 257 |
-
prompt = (
|
| 258 |
-
f"{self.system_prompt}\n\n"
|
| 259 |
-
f"Question: {question}\n\n"
|
| 260 |
-
)
|
| 261 |
-
|
| 262 |
-
return self.local_agent.run(prompt)
|
|
|
|
| 3 |
import os
|
| 4 |
from typing import Optional
|
| 5 |
import pandas as pd
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
# Smolagents imports
|
| 8 |
from smolagents import (
|
|
|
|
| 187 |
)
|
| 188 |
|
| 189 |
return self.gemini_agent.run(prompt)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
requirements.txt
CHANGED
|
@@ -25,9 +25,4 @@ pytubefix==10.3.6
|
|
| 25 |
openai-whisper==20250625
|
| 26 |
|
| 27 |
# OCR (OPTIONAL, disabled)
|
| 28 |
-
# pytesseract==0.3.13
|
| 29 |
-
|
| 30 |
-
# Additional for LocalAgent (optional)
|
| 31 |
-
transformers==4.1.0
|
| 32 |
-
bitsandbytes==0.49.0
|
| 33 |
-
torch==2.6.0
|
|
|
|
| 25 |
openai-whisper==20250625
|
| 26 |
|
| 27 |
# OCR (OPTIONAL, disabled)
|
| 28 |
+
# pytesseract==0.3.13
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|