Spaces:
Runtime error
Runtime error
Commit
•
db328d1
1
Parent(s):
d38a8cf
Upload folder using huggingface_hub
Browse files- README.md +3 -9
- app.py +61 -0
- backend.py +66 -0
- flask.py.save +18 -0
- flask.py.save.1 +18 -0
- oldbacked.py +83 -0
- requirements.txt +1 -0
- server.py +8 -0
README.md
CHANGED
@@ -1,12 +1,6 @@
|
|
1 |
---
|
2 |
-
title:
|
3 |
-
|
4 |
-
colorFrom: blue
|
5 |
-
colorTo: indigo
|
6 |
sdk: gradio
|
7 |
-
sdk_version:
|
8 |
-
app_file: app.py
|
9 |
-
pinned: false
|
10 |
---
|
11 |
-
|
12 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
+
title: gradio-server
|
3 |
+
app_file: backend.py
|
|
|
|
|
4 |
sdk: gradio
|
5 |
+
sdk_version: 3.50.2
|
|
|
|
|
6 |
---
|
|
|
|
app.py
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import time
|
3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
4 |
+
from langchain.memory import ConversationBufferWindowMemory
|
5 |
+
from peft import PeftModel
|
6 |
+
import torch
|
7 |
+
import re
|
8 |
+
|
9 |
+
|
10 |
+
print("Initializing model")
|
11 |
+
# Initialize the tokenizer and model
|
12 |
+
base_model = "mistralai/Mistral-7B-Instruct-v0.2"
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained(base_model)
|
14 |
+
tokenizer.add_special_tokens({"pad_token": "[PAD]"})
|
15 |
+
base_model = AutoModelForCausalLM.from_pretrained(base_model)
|
16 |
+
|
17 |
+
ft_model = PeftModel.from_pretrained(base_model, "nuratamton/story_sculptor_mistral")
|
18 |
+
# ft_model = ft_model.merge_and_unload()
|
19 |
+
ft_model.eval()
|
20 |
+
|
21 |
+
|
22 |
+
# Set the device
|
23 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
24 |
+
ft_model.to(device)
|
25 |
+
|
26 |
+
|
27 |
+
memory = ConversationBufferWindowMemory(k=10)
|
28 |
+
|
29 |
+
def slow_echo(message, history):
|
30 |
+
message = chat_interface(message)
|
31 |
+
for i in range(len(message)):
|
32 |
+
time.sleep(0.05)
|
33 |
+
yield message[: i+1]
|
34 |
+
|
35 |
+
|
36 |
+
def chat_interface(user_in):
|
37 |
+
if user_in.lower() == "quit":
|
38 |
+
return "Goodbye!"
|
39 |
+
#memory.save_context({"input": user_in}, {"output": ""})
|
40 |
+
memory_context = memory.load_memory_variables({})["history"]
|
41 |
+
user_input = f"[INST] Continue the game and maintain context and keep the story consistent throughout: {memory_context}{user_in}[/INST]"
|
42 |
+
|
43 |
+
encodings = tokenizer(user_input, return_tensors="pt", padding=True).to(device)
|
44 |
+
input_ids = encodings["input_ids"]
|
45 |
+
attention_mask = encodings["attention_mask"]
|
46 |
+
|
47 |
+
output_ids = ft_model.generate(input_ids, attention_mask = attention_mask, max_new_tokens=1000, num_return_sequences=1, do_sample=True, temperature=1.1, top_p=0.9, repetition_penalty=1.2)
|
48 |
+
|
49 |
+
generated_ids = output_ids[0, input_ids.shape[-1]:]
|
50 |
+
|
51 |
+
# Decode the output
|
52 |
+
response = tokenizer.decode(generated_ids, skip_special_tokens=True)
|
53 |
+
memory.save_context({"input": user_in}, {"output": response})
|
54 |
+
print(f"Game Agent: {response}")
|
55 |
+
# Your chatbot logic here
|
56 |
+
# response = "You said: " + user_in
|
57 |
+
return response
|
58 |
+
|
59 |
+
|
60 |
+
iface = gr.ChatInterface(slow_echo).queue()
|
61 |
+
iface.launch(share=True)
|
backend.py
ADDED
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
3 |
+
from peft import PeftModel
|
4 |
+
from langchain.memory import ConversationBufferWindowMemory
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
bnb_config = BitsAndBytesConfig(
|
8 |
+
load_in_4bit=True,
|
9 |
+
bnb_4bit_use_double_quant=True,
|
10 |
+
bnb_4bit_quant_type="nf4",
|
11 |
+
bnb_4bit_compute_dtype=torch.bfloat16,
|
12 |
+
)
|
13 |
+
|
14 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
15 |
+
base_model = "mistralai/Mistral-7B-Instruct-v0.2"
|
16 |
+
tokenizer = AutoTokenizer.from_pretrained(base_model, pad_token="[PAD]")
|
17 |
+
model = AutoModelForCausalLM.from_pretrained(
|
18 |
+
base_model,
|
19 |
+
quantization_config=bnb_config,
|
20 |
+
device_map="auto",
|
21 |
+
trust_remote_code=True,
|
22 |
+
)
|
23 |
+
ft_model = PeftModel.from_pretrained(model, "nuratamton/story_sculptor_mistral").eval()
|
24 |
+
memory = ConversationBufferWindowMemory(k=10)
|
25 |
+
|
26 |
+
def generate_text(message):
|
27 |
+
user_in = message
|
28 |
+
|
29 |
+
if user_in.lower() in ["adventure", "mystery", "horror", "sci-fi"]:
|
30 |
+
memory.clear()
|
31 |
+
|
32 |
+
if user_in.lower() == "quit":
|
33 |
+
raise ValueError("User requested to quit")
|
34 |
+
|
35 |
+
memory_context = memory.load_memory_variables({})["history"]
|
36 |
+
user_input = f"{memory_context}[INST] Continue the game and maintain context: {user_in}[/INST]"
|
37 |
+
|
38 |
+
encodings = tokenizer(user_input, return_tensors="pt", padding=True).to(device)
|
39 |
+
input_ids, attention_mask = encodings["input_ids"], encodings["attention_mask"]
|
40 |
+
output_ids = ft_model.generate(
|
41 |
+
input_ids,
|
42 |
+
attention_mask=attention_mask,
|
43 |
+
max_new_tokens=1000,
|
44 |
+
num_return_sequences=1,
|
45 |
+
do_sample=True,
|
46 |
+
temperature=1.1,
|
47 |
+
top_p=0.9,
|
48 |
+
repetition_penalty=1.2,
|
49 |
+
)
|
50 |
+
|
51 |
+
generated_ids = output_ids[0, input_ids.shape[-1] :]
|
52 |
+
response = tokenizer.decode(generated_ids, skip_special_tokens=True)
|
53 |
+
memory.save_context({"input": user_in}, {"output": response})
|
54 |
+
|
55 |
+
response = response.replace("AI: ", "")
|
56 |
+
return response
|
57 |
+
|
58 |
+
iface = gr.Interface(
|
59 |
+
fn=generate_text,
|
60 |
+
inputs="text",
|
61 |
+
outputs="text",
|
62 |
+
title="Text Generation",
|
63 |
+
description="Enter a message to generate text.",
|
64 |
+
)
|
65 |
+
|
66 |
+
iface.launch(share=True)
|
flask.py.save
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from flask import Flask, request, jsonify
|
2 |
+
import gradio as gr
|
3 |
+
|
4 |
+
app = Flask(__name__)
|
5 |
+
|
6 |
+
def my_gradio_function(input_text):
|
7 |
+
# Your processing logic here
|
8 |
+
return "Processed: " + input_text
|
9 |
+
|
10 |
+
@app.route("/process", methods=["POST"])
|
11 |
+
def process():
|
12 |
+
input_text = request.json["input_text"]
|
13 |
+
output_text = my_gradio_function(input_text)
|
14 |
+
return jsonify({"output_text": output_text})
|
15 |
+
|
16 |
+
if __name__ == "__main__":
|
17 |
+
app.run(port=5000)
|
18 |
+
|
flask.py.save.1
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from flask import Flask, request, jsonify
|
2 |
+
import gradio as gr
|
3 |
+
|
4 |
+
app = Flask(__name__)
|
5 |
+
|
6 |
+
def my_gradio_function(input_text):
|
7 |
+
# Your processing logic here
|
8 |
+
return "Processed: " + input_text
|
9 |
+
|
10 |
+
@app.route("/process", methods=["POST"])
|
11 |
+
def process():
|
12 |
+
input_text = request.json["input_text"]
|
13 |
+
output_text = my_gradio_function(input_text)
|
14 |
+
return jsonify({"output_text": output_text})
|
15 |
+
|
16 |
+
if __name__ == "__main__":
|
17 |
+
app.run(port=5000)
|
18 |
+
|
oldbacked.py
ADDED
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException
|
2 |
+
from pydantic import BaseModel
|
3 |
+
import torch
|
4 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
5 |
+
from peft import PeftModel
|
6 |
+
from langchain.memory import ConversationBufferWindowMemory
|
7 |
+
from fastapi.middleware.cors import CORSMiddleware
|
8 |
+
|
9 |
+
app = FastAPI()
|
10 |
+
|
11 |
+
# Add CORSMiddleware to the application
|
12 |
+
app.add_middleware(
|
13 |
+
CORSMiddleware,
|
14 |
+
allow_origins=["*"],
|
15 |
+
allow_credentials=True,
|
16 |
+
allow_methods=["*"],
|
17 |
+
allow_headers=["*"],
|
18 |
+
)
|
19 |
+
|
20 |
+
bnb_config = BitsAndBytesConfig(
|
21 |
+
load_in_4bit=True,
|
22 |
+
bnb_4bit_use_double_quant=True,
|
23 |
+
bnb_4bit_quant_type="nf4",
|
24 |
+
bnb_4bit_compute_dtype=torch.bfloat16,
|
25 |
+
)
|
26 |
+
|
27 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
28 |
+
base_model = "mistralai/Mistral-7B-Instruct-v0.2"
|
29 |
+
tokenizer = AutoTokenizer.from_pretrained(base_model, pad_token="[PAD]")
|
30 |
+
model = AutoModelForCausalLM.from_pretrained(
|
31 |
+
base_model,
|
32 |
+
quantization_config=bnb_config,
|
33 |
+
device_map="auto",
|
34 |
+
trust_remote_code=True,
|
35 |
+
)
|
36 |
+
ft_model = PeftModel.from_pretrained(model, "nuratamton/story_sculptor_mistral").eval()
|
37 |
+
memory = ConversationBufferWindowMemory(k=10)
|
38 |
+
|
39 |
+
|
40 |
+
class UserRequest(BaseModel):
|
41 |
+
message: str
|
42 |
+
|
43 |
+
|
44 |
+
@app.post("/generate/")
|
45 |
+
async def generate_text(request: UserRequest):
|
46 |
+
user_in = request.message
|
47 |
+
|
48 |
+
if user_in.lower() in ["adventure", "mystery", "horror", "sci-fi"]:
|
49 |
+
memory.clear()
|
50 |
+
|
51 |
+
if user_in.lower() == "quit":
|
52 |
+
raise HTTPException(status_code=400, detail="User requested to quit")
|
53 |
+
|
54 |
+
memory_context = memory.load_memory_variables({})["history"]
|
55 |
+
user_input = f"{memory_context}[INST] Continue the game and maintain context: {user_in}[/INST]"
|
56 |
+
|
57 |
+
encodings = tokenizer(user_input, return_tensors="pt", padding=True).to(device)
|
58 |
+
input_ids, attention_mask = encodings["input_ids"], encodings["attention_mask"]
|
59 |
+
output_ids = ft_model.generate(
|
60 |
+
input_ids,
|
61 |
+
attention_mask=attention_mask,
|
62 |
+
max_new_tokens=1000,
|
63 |
+
num_return_sequences=1,
|
64 |
+
do_sample=True,
|
65 |
+
temperature=1.1,
|
66 |
+
top_p=0.9,
|
67 |
+
repetition_penalty=1.2,
|
68 |
+
)
|
69 |
+
|
70 |
+
generated_ids = output_ids[0, input_ids.shape[-1] :]
|
71 |
+
response = tokenizer.decode(generated_ids, skip_special_tokens=True)
|
72 |
+
memory.save_context({"input": user_in}, {"output": response})
|
73 |
+
|
74 |
+
response = response.replace("AI: ", "")
|
75 |
+
# response = response.replace("Human: ", "")
|
76 |
+
|
77 |
+
return {"response": response}
|
78 |
+
|
79 |
+
|
80 |
+
@app.get("/")
|
81 |
+
def read_root():
|
82 |
+
return {"message": "Hello from FastAPI"}
|
83 |
+
|
requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
torch transformers peft langchain
|
server.py
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
|
3 |
+
def add_numbers(num1, num2):
|
4 |
+
return num1 + num2
|
5 |
+
|
6 |
+
iface = gr.Interface(fn=add_numbers, inputs=["number", "number"], outputs="number")
|
7 |
+
iface.launch(share=True)
|
8 |
+
|