Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,14 @@
|
|
| 1 |
# -*- coding: utf-8 -*-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
from huggingface_hub import snapshot_download
|
| 3 |
|
| 4 |
# Clip
|
|
|
|
| 1 |
# -*- coding: utf-8 -*-
|
| 2 |
+
from ipex_llm.transformers import AutoModelForCausalLM
|
| 3 |
+
from transformers import LlamaTokenizer
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
llm = AutoModelForCausalLM.from_pretrained("checkpoints\\Llama-2-7b-chat-hf",load_in_low_bit="sym_int4")
|
| 7 |
+
llm.save_low_bit("checkpoints\\Llama-2-7b-chat-hf-INT4")
|
| 8 |
+
|
| 9 |
+
tokenizer = LlamaTokenizer.from_pretrained("checkpoints\\Llama-2-7b-chat-hf\\")
|
| 10 |
+
tokenizer.save_pretrained("checkpoints\\Llama-2-7b-chat-hf-INT4")
|
| 11 |
+
|
| 12 |
from huggingface_hub import snapshot_download
|
| 13 |
|
| 14 |
# Clip
|