Spaces:
Runtime error
Runtime error
Feature(MInference): changing to GPU
Browse files- app.py +12 -8
- requirements.txt +3 -1
app.py
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
-
import subprocess
|
| 2 |
-
# Install flash attention, skipping CUDA build if necessary
|
| 3 |
-
subprocess.run(
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
)
|
| 8 |
|
| 9 |
import gradio as gr
|
| 10 |
import os
|
|
@@ -59,6 +59,10 @@ model_name = "gradientai/Llama-3-8B-Instruct-262k"
|
|
| 59 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 60 |
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto") # to("cuda:0")
|
| 61 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
terminators = [
|
| 63 |
tokenizer.eos_token_id,
|
| 64 |
tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
|
@@ -80,7 +84,7 @@ def chat_llama3_8b(message: str,
|
|
| 80 |
Returns:
|
| 81 |
str: The generated response.
|
| 82 |
"""
|
| 83 |
-
global model
|
| 84 |
conversation = []
|
| 85 |
for user, assistant in history:
|
| 86 |
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
|
|
|
| 1 |
+
# import subprocess
|
| 2 |
+
# # Install flash attention, skipping CUDA build if necessary
|
| 3 |
+
# subprocess.run(
|
| 4 |
+
# "pip install flash-attn --no-build-isolation",
|
| 5 |
+
# env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
|
| 6 |
+
# shell=True,
|
| 7 |
+
# )
|
| 8 |
|
| 9 |
import gradio as gr
|
| 10 |
import os
|
|
|
|
| 59 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 60 |
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto") # to("cuda:0")
|
| 61 |
|
| 62 |
+
from minference import MInference
|
| 63 |
+
minference_patch = MInference("minference", model_name)
|
| 64 |
+
model = minference_patch(model)
|
| 65 |
+
|
| 66 |
terminators = [
|
| 67 |
tokenizer.eos_token_id,
|
| 68 |
tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
|
|
|
| 84 |
Returns:
|
| 85 |
str: The generated response.
|
| 86 |
"""
|
| 87 |
+
# global model
|
| 88 |
conversation = []
|
| 89 |
for user, assistant in history:
|
| 90 |
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
requirements.txt
CHANGED
|
@@ -1,3 +1,5 @@
|
|
| 1 |
triton==2.1.0
|
| 2 |
accelerate
|
| 3 |
-
transformers
|
|
|
|
|
|
|
|
|
| 1 |
triton==2.1.0
|
| 2 |
accelerate
|
| 3 |
+
transformers
|
| 4 |
+
flash_attn
|
| 5 |
+
pycuda==2023.1
|