Spaces:
Sleeping
Sleeping
HarshanaLF
commited on
Commit
•
70ab5a1
1
Parent(s):
a434a54
update
Browse files
app.py
CHANGED
@@ -12,12 +12,42 @@ def client_fn(model):
|
|
12 |
}
|
13 |
return InferenceClient(model_map.get(model, "mistralai/Mixtral-8x7B-Instruct-v0.1"))
|
14 |
|
15 |
-
system_instructions = ("[SYSTEM] You are an assistant
|
|
|
|
|
16 |
"Respond naturally and concisely to the user's queries. "
|
17 |
-
"
|
18 |
"Begin with a greeting if the user initiates the conversation. "
|
19 |
"Here is the user's query: ")
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
# Function to generate model responses
|
22 |
def models(text, model="Mixtral 8x7B"):
|
23 |
client = client_fn(model)
|
|
|
12 |
}
|
13 |
return InferenceClient(model_map.get(model, "mistralai/Mixtral-8x7B-Instruct-v0.1"))
|
14 |
|
15 |
+
system_instructions = ("[SYSTEM] You are an assistant ."
|
16 |
+
"Your task is to Answer the question."
|
17 |
+
"Keep conversation very short, clear and concise."
|
18 |
"Respond naturally and concisely to the user's queries. "
|
19 |
+
"The expectation is that you will avoid introductions and start answering the query directly, Only answer the question asked by user, Do not say unnecessary things."
|
20 |
"Begin with a greeting if the user initiates the conversation. "
|
21 |
"Here is the user's query: ")
|
22 |
|
23 |
+
def models(text, model="Mixtral 8x7B"):
|
24 |
+
|
25 |
+
client = client_fn(model)
|
26 |
+
|
27 |
+
generate_kwargs = dict(
|
28 |
+
max_new_tokens=100,
|
29 |
+
do_sample=True,
|
30 |
+
)
|
31 |
+
|
32 |
+
formatted_prompt = system_instructions1 + text + "[ANSWER]"
|
33 |
+
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
34 |
+
output = ""
|
35 |
+
for response in stream:
|
36 |
+
output+=response.token.text
|
37 |
+
if output.endswith("<|assistant|>"):
|
38 |
+
output = output[:-13]
|
39 |
+
elif output.endswith("</s>"):
|
40 |
+
output = output[:-4]
|
41 |
+
return output
|
42 |
+
|
43 |
+
description="""# Chat GO
|
44 |
+
### Inspired from Google Go"""
|
45 |
+
|
46 |
+
demo = gr.Interface(description=description,fn=models, inputs=["text", gr.Dropdown([ 'Mixtral 8x7B','Nous Hermes Mixtral 8x7B DPO','StarChat2 15b','Mistral 7B v0.3','Phi 3 mini', ], value="Mistral 7B v0.3", label="Select Model") ], outputs="text", live=True, batch=True, max_batch_size=10000)
|
47 |
+
demo.queue(max_size=300000)
|
48 |
+
demo.launch()
|
49 |
+
|
50 |
+
|
51 |
# Function to generate model responses
|
52 |
def models(text, model="Mixtral 8x7B"):
|
53 |
client = client_fn(model)
|