gijl commited on
Commit
4d3f642
·
verified ·
1 Parent(s): 839dab5

Update Dockerfile

Browse files
Files changed (1) hide show
  1. Dockerfile +3 -3
Dockerfile CHANGED
@@ -10,16 +10,16 @@ RUN pip install -U pip huggingface_hub
10
 
11
  RUN python3 -c 'from huggingface_hub import hf_hub_download; \
12
  repo="gijl/gemma-4-31B-it-GGUF"; \
13
- hf_hub_download(repo_id=repo, filename="gemma-4-31B-it-UD-Q8_K_XL.gguf", local_dir="/app"); \
14
  hf_hub_download(repo_id=repo, filename="mmproj-BF16.gguf", local_dir="/app")'
15
 
16
  CMD ["--server", \
17
- "-m", "/app/gemma-4-31B-it-UD-Q8_K_XL.gguf", \
18
  "--mmproj", "/app/mmproj-BF16.gguf", \
19
  "--host", "0.0.0.0", \
20
  "--port", "7860", \
21
  "-t", "2", \
22
  "--cache-type-k", "q8_0", \
23
  "--cache-type-v", "iq4_nl", \
24
- "-c", "128000", \
25
  "-n", "38912"]
 
10
 
11
  RUN python3 -c 'from huggingface_hub import hf_hub_download; \
12
  repo="gijl/gemma-4-31B-it-GGUF"; \
13
+ hf_hub_download(repo_id=repo, filename="gemma-4-31B-it-UD-IQ2_XXS.gguf", local_dir="/app"); \
14
  hf_hub_download(repo_id=repo, filename="mmproj-BF16.gguf", local_dir="/app")'
15
 
16
  CMD ["--server", \
17
+ "-m", "/app/gemma-4-31B-it-UD-IQ2_XXS.gguf", \
18
  "--mmproj", "/app/mmproj-BF16.gguf", \
19
  "--host", "0.0.0.0", \
20
  "--port", "7860", \
21
  "-t", "2", \
22
  "--cache-type-k", "q8_0", \
23
  "--cache-type-v", "iq4_nl", \
24
+ "-c", "12800", \
25
  "-n", "38912"]