d3evil4 commited on
Commit
35424c3
·
1 Parent(s): a3ec6a2

feat: huh

Browse files
Files changed (1) hide show
  1. main.py +20 -1
main.py CHANGED
@@ -2,6 +2,8 @@ from __future__ import annotations
2
 
3
  import asyncio
4
  import logging
 
 
5
  import subprocess
6
  import sys
7
  import time
@@ -21,9 +23,26 @@ logging.basicConfig(
21
  )
22
  logger = logging.getLogger("gemma4")
23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24
  LLAMA_BASE = "http://localhost:8080"
25
  LLAMA_CMD = [
26
- "llama-server",
27
  "-m", "/app/gemma-4-E2B-it-UD-Q5_K_XL.gguf",
28
  "--mmproj", "/app/mmproj-BF16.gguf",
29
  "--host", "0.0.0.0",
 
2
 
3
  import asyncio
4
  import logging
5
+ import os
6
+ import shutil
7
  import subprocess
8
  import sys
9
  import time
 
23
  )
24
  logger = logging.getLogger("gemma4")
25
 
26
+
27
+ def _find_llama_server() -> str:
28
+ candidates = [
29
+ "llama-server",
30
+ "/llama-server",
31
+ "/usr/local/bin/llama-server",
32
+ "/usr/bin/llama-server",
33
+ ]
34
+ for c in candidates:
35
+ found = shutil.which(c)
36
+ if found:
37
+ return found
38
+ if os.path.isfile(c) and os.access(c, os.X_OK):
39
+ return c
40
+ raise RuntimeError(f"llama-server binary not found; searched: {candidates}")
41
+
42
+
43
  LLAMA_BASE = "http://localhost:8080"
44
  LLAMA_CMD = [
45
+ _find_llama_server(),
46
  "-m", "/app/gemma-4-E2B-it-UD-Q5_K_XL.gguf",
47
  "--mmproj", "/app/mmproj-BF16.gguf",
48
  "--host", "0.0.0.0",