Update app/llm.py
Browse files- app/llm.py +2 -2
app/llm.py
CHANGED
|
@@ -32,7 +32,7 @@ class ChatModel(BaseModel):
|
|
| 32 |
llm_chat = llama_cpp.Llama.from_pretrained(
|
| 33 |
repo_id="moriire/healthcare-ai-q2_k",
|
| 34 |
filename="*.gguf",
|
| 35 |
-
tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer.from_pretrained("
|
| 36 |
verbose=False,
|
| 37 |
n_ctx=256,
|
| 38 |
n_gpu_layers=0,
|
|
@@ -41,7 +41,7 @@ llm_chat = llama_cpp.Llama.from_pretrained(
|
|
| 41 |
llm_generate = llama_cpp.Llama.from_pretrained(
|
| 42 |
repo_id="moriire/healthcare-ai-q2_k",
|
| 43 |
filename="*.gguf",
|
| 44 |
-
tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer.from_pretrained("
|
| 45 |
verbose=False,
|
| 46 |
n_ctx=4096,
|
| 47 |
n_gpu_layers=0,
|
|
|
|
| 32 |
llm_chat = llama_cpp.Llama.from_pretrained(
|
| 33 |
repo_id="moriire/healthcare-ai-q2_k",
|
| 34 |
filename="*.gguf",
|
| 35 |
+
tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer.from_pretrained("healthcare-ai-q2_k"),
|
| 36 |
verbose=False,
|
| 37 |
n_ctx=256,
|
| 38 |
n_gpu_layers=0,
|
|
|
|
| 41 |
llm_generate = llama_cpp.Llama.from_pretrained(
|
| 42 |
repo_id="moriire/healthcare-ai-q2_k",
|
| 43 |
filename="*.gguf",
|
| 44 |
+
tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer.from_pretrained("healthcare-ai-q2_k"),
|
| 45 |
verbose=False,
|
| 46 |
n_ctx=4096,
|
| 47 |
n_gpu_layers=0,
|