matthoffner commited on
Commit
6bf5156
·
1 Parent(s): 6ce25db

Update llm.py

Browse files
Files changed (1) hide show
  1. llm.py +2 -1
llm.py CHANGED
@@ -194,7 +194,8 @@ def ask_ai(
194
  last_n_tokens_size=100,
195
  n_threads=4,
196
  f16_kv=True,
197
- max_tokens=200
 
198
  )
199
  embeddings = HuggingFaceEmbeddings(model_kwargs={"device": "cuda"})
200
  embed_model = LangchainEmbedding(embeddings)
 
194
  last_n_tokens_size=100,
195
  n_threads=4,
196
  f16_kv=True,
197
+ max_tokens=200,
198
+ n_gpu_layers=20
199
  )
200
  embeddings = HuggingFaceEmbeddings(model_kwargs={"device": "cuda"})
201
  embed_model = LangchainEmbedding(embeddings)