Spaces:
Build error
Build error
Added Better Inferencing techq
Browse files
app.py
CHANGED
|
@@ -1,29 +1,36 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import torch
|
| 3 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
| 4 |
import os
|
| 5 |
from threading import Thread
|
| 6 |
import spaces
|
|
|
|
| 7 |
|
| 8 |
token = os.environ["HF_TOKEN"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
model = AutoModelForCausalLM.from_pretrained("google/gemma-1.1-7b-it",
|
| 10 |
-
|
| 11 |
-
torch_dtype=torch.float16,
|
| 12 |
token=token)
|
| 13 |
-
tok = AutoTokenizer.from_pretrained("google/gemma-1.1-7b-it",token=token)
|
| 14 |
-
|
| 15 |
-
# device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 16 |
if torch.cuda.is_available():
|
| 17 |
device = torch.device('cuda')
|
| 18 |
print(f"Using GPU: {torch.cuda.get_device_name(device)}")
|
| 19 |
else:
|
| 20 |
device = torch.device('cpu')
|
| 21 |
print("Using CPU")
|
|
|
|
| 22 |
model = model.to(device)
|
| 23 |
-
|
| 24 |
|
| 25 |
@spaces.GPU
|
| 26 |
def chat(message, history):
|
|
|
|
| 27 |
chat = []
|
| 28 |
for item in history:
|
| 29 |
chat.append({"role": "user", "content": item[0]})
|
|
@@ -31,7 +38,6 @@ def chat(message, history):
|
|
| 31 |
chat.append({"role": "assistant", "content": item[1]})
|
| 32 |
chat.append({"role": "user", "content": message})
|
| 33 |
messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
|
| 34 |
-
# Tokenize the messages string
|
| 35 |
model_inputs = tok([messages], return_tensors="pt").to(device)
|
| 36 |
streamer = TextIteratorStreamer(
|
| 37 |
tok, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
|
@@ -48,15 +54,21 @@ def chat(message, history):
|
|
| 48 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
| 49 |
t.start()
|
| 50 |
|
| 51 |
-
# Initialize an empty string to store the generated text
|
| 52 |
partial_text = ""
|
|
|
|
| 53 |
for new_text in streamer:
|
| 54 |
-
|
|
|
|
| 55 |
partial_text += new_text
|
| 56 |
-
# Yield an empty string to cleanup the message textbox and the updated conversation history
|
| 57 |
yield partial_text
|
| 58 |
|
|
|
|
|
|
|
|
|
|
| 59 |
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
-
demo = gr.ChatInterface(fn=chat, examples=[["Write me a poem about Machine Learning."]], title="
|
| 62 |
demo.launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import torch
|
| 3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
|
| 4 |
import os
|
| 5 |
from threading import Thread
|
| 6 |
import spaces
|
| 7 |
+
import time
|
| 8 |
|
| 9 |
token = os.environ["HF_TOKEN"]
|
| 10 |
+
|
| 11 |
+
quantization_config = BitsAndBytesConfig(
|
| 12 |
+
load_in_4bit=True,
|
| 13 |
+
bnb_4bit_compute_dtype=torch.float16
|
| 14 |
+
)
|
| 15 |
+
|
| 16 |
model = AutoModelForCausalLM.from_pretrained("google/gemma-1.1-7b-it",
|
| 17 |
+
quantization_config=quantization_config,
|
|
|
|
| 18 |
token=token)
|
| 19 |
+
tok = AutoTokenizer.from_pretrained("google/gemma-1.1-7b-it", token=token)
|
| 20 |
+
|
|
|
|
| 21 |
if torch.cuda.is_available():
|
| 22 |
device = torch.device('cuda')
|
| 23 |
print(f"Using GPU: {torch.cuda.get_device_name(device)}")
|
| 24 |
else:
|
| 25 |
device = torch.device('cpu')
|
| 26 |
print("Using CPU")
|
| 27 |
+
|
| 28 |
model = model.to(device)
|
| 29 |
+
model = model.to_bettertransformer()
|
| 30 |
|
| 31 |
@spaces.GPU
|
| 32 |
def chat(message, history):
|
| 33 |
+
start_time = time.time()
|
| 34 |
chat = []
|
| 35 |
for item in history:
|
| 36 |
chat.append({"role": "user", "content": item[0]})
|
|
|
|
| 38 |
chat.append({"role": "assistant", "content": item[1]})
|
| 39 |
chat.append({"role": "user", "content": message})
|
| 40 |
messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
|
|
|
|
| 41 |
model_inputs = tok([messages], return_tensors="pt").to(device)
|
| 42 |
streamer = TextIteratorStreamer(
|
| 43 |
tok, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
|
|
|
| 54 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
| 55 |
t.start()
|
| 56 |
|
|
|
|
| 57 |
partial_text = ""
|
| 58 |
+
first_token_time = None
|
| 59 |
for new_text in streamer:
|
| 60 |
+
if not first_token_time:
|
| 61 |
+
first_token_time = time.time() - start_time
|
| 62 |
partial_text += new_text
|
|
|
|
| 63 |
yield partial_text
|
| 64 |
|
| 65 |
+
total_time = time.time() - start_time
|
| 66 |
+
tokens = len(tok.tokenize(partial_text))
|
| 67 |
+
tokens_per_second = tokens / total_time if total_time > 0 else 0
|
| 68 |
|
| 69 |
+
# Append the timing information to the final output
|
| 70 |
+
timing_info = f"\nTime taken to first token: {first_token_time:.2f} seconds\nTokens per second: {tokens_per_second:.2f}"
|
| 71 |
+
yield partial_text + timing_info
|
| 72 |
|
| 73 |
+
demo = gr.ChatInterface(fn=chat, examples=[["Write me a poem about Machine Learning."]], title="Chat With LLMS")
|
| 74 |
demo.launch()
|