change temperature
Browse files- app.py +0 -1
- src/llm/chat.py +3 -2
app.py
CHANGED
@@ -49,7 +49,6 @@ def inference(pil_img, prompt, task, temperature):
|
|
49 |
except Exception:
|
50 |
pass # if deletion fails we just move on
|
51 |
|
52 |
-
# ββββββββββββββββββββββββββββ UI ββββββββββββββββββββββββββββ
|
53 |
def create_header():
|
54 |
with gr.Row():
|
55 |
with gr.Column(scale=1):
|
|
|
49 |
except Exception:
|
50 |
pass # if deletion fails we just move on
|
51 |
|
|
|
52 |
def create_header():
|
53 |
with gr.Row():
|
54 |
with gr.Column(scale=1):
|
src/llm/chat.py
CHANGED
@@ -18,11 +18,12 @@ Here is a list of functions in JSON format that you can invoke.\n\n{functions}\n
|
|
18 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
19 |
|
20 |
class FunctionCallingChat:
|
21 |
-
def __init__(self, model_id: str = "meta-llama/Llama-3.2-1B-Instruct"):
|
22 |
self.tokenizer = AutoTokenizer.from_pretrained(model_id)
|
23 |
self.model = AutoModelForCausalLM.from_pretrained(
|
24 |
model_id, device_map=device, torch_dtype=torch.bfloat16
|
25 |
)
|
|
|
26 |
|
27 |
def __call__(self, user_msg: str) -> dict:
|
28 |
messages = [
|
@@ -31,7 +32,7 @@ class FunctionCallingChat:
|
|
31 |
]
|
32 |
|
33 |
generation_cfg = GenerationConfig(
|
34 |
-
max_new_tokens=128, temperature=
|
35 |
)
|
36 |
|
37 |
tokenized = self.tokenizer.apply_chat_template(
|
|
|
18 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
19 |
|
20 |
class FunctionCallingChat:
|
21 |
+
def __init__(self, model_id: str = "meta-llama/Llama-3.2-1B-Instruct", temperature: float = 0.7):
|
22 |
self.tokenizer = AutoTokenizer.from_pretrained(model_id)
|
23 |
self.model = AutoModelForCausalLM.from_pretrained(
|
24 |
model_id, device_map=device, torch_dtype=torch.bfloat16
|
25 |
)
|
26 |
+
self.temperature = temperature
|
27 |
|
28 |
def __call__(self, user_msg: str) -> dict:
|
29 |
messages = [
|
|
|
32 |
]
|
33 |
|
34 |
generation_cfg = GenerationConfig(
|
35 |
+
max_new_tokens=128, temperature=self.temperature, top_p=0.95, do_sample=True
|
36 |
)
|
37 |
|
38 |
tokenized = self.tokenizer.apply_chat_template(
|