Update 13b for hf

app.py

@@ -3,6 +3,7 @@
 # ruff: noqa: E501
 import gc
 import os
+import platform
 import random
 import time
 from collections import deque
@@ -30,8 +31,6 @@ from loguru import logger
 deq = deque()
 sig_end = object()  # signals the processing is done
 
-url = "https://huggingface.co/TheBloke/llama-2-70b-Guanaco-QLoRA-GGML/blob/main/llama-2-70b-guanaco-qlora.ggmlv3.q3_K_S.bin"  # 29.7G
-
 prompt_template = """Below is an instruction that describes a task. Write a response that appropriately completes the request.
 
 ### Instruction: {user_prompt}
@@ -163,6 +162,11 @@ logger.debug(f"{cpu_count=}")
 LLM = None
 gc.collect()
 
+if "forindo" in platform.node().lower():
+    url = "https://huggingface.co/TheBloke/llama-2-70b-Guanaco-QLoRA-GGML/blob/main/llama-2-70b-guanaco-qlora.ggmlv3.q3_K_S.bin"  # 29.7G
+else:
+    url = "https://huggingface.co/TheBloke/llama-2-13B-Guanaco-QLoRA-GGML/blob/main/llama-2-13b-guanaco-qlora.ggmlv3.q4_K_S.bin"  # 8.14G
+logger.debug(f"{url=}")
 try:
     model_loc, file_size = dl_hf_model(url)
 except Exception as exc_: