Update models/llm_setup.py
Browse files- models/llm_setup.py +65 -64
models/llm_setup.py
CHANGED
@@ -1,64 +1,65 @@
|
|
1 |
-
from typing import Optional
|
2 |
-
from llama_index.llms import HuggingFaceLLM
|
3 |
-
import
|
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 |
-
|
|
|
|
1 |
+
from typing import Optional
|
2 |
+
#from llama_index.llms import HuggingFaceLLM
|
3 |
+
from llama_index.llms.huggingface import HuggingFaceLLM
|
4 |
+
import torch
|
5 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
|
6 |
+
|
7 |
+
def setup_llm(model_name: str = "microsoft/phi-3-mini-4k-instruct",
|
8 |
+
device: str = None,
|
9 |
+
context_window: int = 4096,
|
10 |
+
max_new_tokens: int = 512) -> HuggingFaceLLM:
|
11 |
+
"""
|
12 |
+
Set up the language model for the CSV chatbot.
|
13 |
+
|
14 |
+
Args:
|
15 |
+
model_name: Name of the Hugging Face model to use
|
16 |
+
device: Device to run the model on ('cuda', 'cpu', etc.)
|
17 |
+
context_window: Maximum context window size
|
18 |
+
max_new_tokens: Maximum number of new tokens to generate
|
19 |
+
|
20 |
+
Returns:
|
21 |
+
Configured LLM instance
|
22 |
+
"""
|
23 |
+
# Determine device
|
24 |
+
if device is None:
|
25 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
26 |
+
|
27 |
+
# Configure quantization for memory efficiency
|
28 |
+
if device == "cuda":
|
29 |
+
quantization_config = BitsAndBytesConfig(
|
30 |
+
load_in_4bit=True,
|
31 |
+
bnb_4bit_compute_dtype=torch.float16
|
32 |
+
)
|
33 |
+
else:
|
34 |
+
quantization_config = None
|
35 |
+
|
36 |
+
# Configure tokenizer
|
37 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
38 |
+
model_name,
|
39 |
+
trust_remote_code=True
|
40 |
+
)
|
41 |
+
|
42 |
+
# Configure model with appropriate parameters for HF Spaces
|
43 |
+
model_kwargs = {
|
44 |
+
"trust_remote_code": True,
|
45 |
+
"torch_dtype": torch.float16,
|
46 |
+
}
|
47 |
+
|
48 |
+
if quantization_config:
|
49 |
+
model_kwargs["quantization_config"] = quantization_config
|
50 |
+
|
51 |
+
# Initialize LLM
|
52 |
+
llm = HuggingFaceLLM(
|
53 |
+
model_name=model_name,
|
54 |
+
tokenizer_name=model_name,
|
55 |
+
context_window=context_window,
|
56 |
+
max_new_tokens=max_new_tokens,
|
57 |
+
generate_kwargs={"temperature": 0.7, "top_p": 0.95},
|
58 |
+
device_map=device,
|
59 |
+
tokenizer_kwargs={"trust_remote_code": True},
|
60 |
+
model_kwargs=model_kwargs,
|
61 |
+
# Cache the model to avoid reloading
|
62 |
+
cache_folder="./model_cache"
|
63 |
+
)
|
64 |
+
|
65 |
+
return llm
|