Dorjzodovsuren commited on
Commit
1a5495f
·
verified ·
1 Parent(s): 96ffa7d

Update Dockerfile

Browse files
Files changed (1) hide show
  1. Dockerfile +27 -10
Dockerfile CHANGED
@@ -1,26 +1,43 @@
1
- FROM ghcr.io/ggml-org/llama.cpp:server
2
 
3
- RUN apt update && apt install wget -y && rm -rf /var/lib/apt/lists/*
4
 
5
  # RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf" -O /gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
6
 
7
- RUN wget "https://huggingface.co/Dorjzodovsuren/Mongolian_Llama3-v0.1/blob/main/Mongolian_Llama3-v1.Q4_K_M.gguf" -O /mongolian_llama.gguf
 
 
 
 
8
 
9
- RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mtp-gemma-4-E2B-it.gguf" -O /mtp-gemma-4-E2B-it.gguf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
- RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mmproj-F16.gguf" -O /gemma-4-E2B-it-mmproj.gguf
12
 
13
  CMD [ \
14
  "-m", "/mongolian_llama.gguf", \
15
- "--spec-draft-model", "/mtp-gemma-4-E2B-it.gguf", \
16
- "--mmproj", "/gemma-4-E2B-it-mmproj.gguf", \
17
  "--port", "7860", \
18
  "--host", "0.0.0.0", \
19
  "-t", "2", \
20
  "-tb", "2", \
21
  "-fa", "on", \
22
  "-ub", "128", \
23
- "-b", "512", \
24
- "--spec-type", "draft-mtp", \
25
- "--spec-draft-n-max", "3" \
26
  ]
 
1
+ # FROM ghcr.io/ggml-org/llama.cpp:server
2
 
3
+ # RUN apt update && apt install wget -y && rm -rf /var/lib/apt/lists/*
4
 
5
  # RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf" -O /gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
6
 
7
+ # RUN wget "https://huggingface.co/Dorjzodovsuren/Mongolian_Llama3-v0.1/resolve/main/Mongolian_Llama3-v1.Q4_K_M.gguf" -O /mongolian_llama.gguf
8
+
9
+ # RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mtp-gemma-4-E2B-it.gguf" -O /mtp-gemma-4-E2B-it.gguf
10
+
11
+ # RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mmproj-F16.gguf" -O /gemma-4-E2B-it-mmproj.gguf
12
 
13
+ # CMD [ \
14
+ # "-m", "/mongolian_llama.gguf", \
15
+ # "--spec-draft-model", "/mtp-gemma-4-E2B-it.gguf", \
16
+ # "--mmproj", "/gemma-4-E2B-it-mmproj.gguf", \
17
+ # "--port", "7860", \
18
+ # "--host", "0.0.0.0", \
19
+ # "-t", "2", \
20
+ # "-tb", "2", \
21
+ # "-fa", "on", \
22
+ # "-ub", "128", \
23
+ # "-b", "512", \
24
+ # "--spec-type", "draft-mtp", \
25
+ # "--spec-draft-n-max", "3" \
26
+ # ]
27
+
28
+ FROM ghcr.io/ggml-org/llama.cpp:server
29
+
30
+ RUN apt update && apt install wget -y && rm -rf /var/lib/apt/lists/*
31
 
32
+ RUN wget "https://huggingface.co/Dorjzodovsuren/Mongolian_Llama3-v0.1/resolve/main/Mongolian_Llama3-v1.Q4_K_M.gguf" -O /mongolian_llama.gguf
33
 
34
  CMD [ \
35
  "-m", "/mongolian_llama.gguf", \
 
 
36
  "--port", "7860", \
37
  "--host", "0.0.0.0", \
38
  "-t", "2", \
39
  "-tb", "2", \
40
  "-fa", "on", \
41
  "-ub", "128", \
42
+ "-b", "512" \
 
 
43
  ]