Update README.md
Browse files
README.md
CHANGED
@@ -5,3 +5,134 @@ license: apache-2.0
|
|
5 |
# The Public-shared LoRA Adapter for shuyuej/Llama-3.3-70B-Instruct-GPTQ Model
|
6 |
This is publicly-shared LoRA Adapter for the `shuyuej/Llama-3.3-70B-Instruct-GPTQ` model.<br>
|
7 |
Please check our GPTQ-quantized model [https://huggingface.co/shuyuej/Llama-3.3-70B-Instruct-GPTQ](https://huggingface.co/shuyuej/Llama-3.3-70B-Instruct-GPTQ).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
# The Public-shared LoRA Adapter for shuyuej/Llama-3.3-70B-Instruct-GPTQ Model
|
6 |
This is publicly-shared LoRA Adapter for the `shuyuej/Llama-3.3-70B-Instruct-GPTQ` model.<br>
|
7 |
Please check our GPTQ-quantized model [https://huggingface.co/shuyuej/Llama-3.3-70B-Instruct-GPTQ](https://huggingface.co/shuyuej/Llama-3.3-70B-Instruct-GPTQ).
|
8 |
+
|
9 |
+
# 🔥 Real-world deployment
|
10 |
+
For real-world deployment, please refer to the [vLLM Distributed Inference and Serving](https://docs.vllm.ai/en/latest/serving/distributed_serving.html) and [OpenAI Compatible Server](https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html). We provide a deployment script [here](https://github.com/vkola-lab/PodGPT/blob/main/scripts/deployment.py).
|
11 |
+
|
12 |
+
> [!NOTE]
|
13 |
+
> The vLLM version we are using is `0.6.2`. Please check [this version](https://github.com/vllm-project/vllm/releases/tag/v0.6.2).
|
14 |
+
|
15 |
+
vLLM can be deployed as a server that implements the OpenAI API protocol. This allows vLLM to be used as a drop-in replacement for applications using OpenAI API. By default, it starts the server at `http://localhost:8000`.
|
16 |
+
```shell
|
17 |
+
vllm serve shuyuej/Llama-3.3-70B-Instruct-GPTQ \
|
18 |
+
--quantization gptq \
|
19 |
+
--trust-remote-code \
|
20 |
+
--dtype float16 \
|
21 |
+
--max-model-len 4096 \
|
22 |
+
--distributed-executor-backend mp \
|
23 |
+
--pipeline-parallel-size 4 \
|
24 |
+
--api-key token-abc123
|
25 |
+
```
|
26 |
+
Please check [here](https://docs.vllm.ai/en/latest/usage/engine_args.html) if you wanna change `Engine Arguments`.
|
27 |
+
|
28 |
+
If you would like to deploy your LoRA adapter, please refer to the [vLLM documentation](https://docs.vllm.ai/en/latest/usage/lora.html#serving-lora-adapters) for a detailed guide.
|
29 |
+
It provides step-by-step instructions on how to serve LoRA adapters effectively in a vLLM environment.
|
30 |
+
```shell
|
31 |
+
vllm serve shuyuej/Llama-3.3-70B-Instruct-GPTQ \
|
32 |
+
--quantization gptq \
|
33 |
+
--trust-remote-code \
|
34 |
+
--dtype float16 \
|
35 |
+
--max-model-len 4096 \
|
36 |
+
--distributed-executor-backend mp \
|
37 |
+
--pipeline-parallel-size 4 \
|
38 |
+
--api-key token-abc123 \
|
39 |
+
--enable-lora \
|
40 |
+
--lora-modules adapter=checkpoint-18640
|
41 |
+
```
|
42 |
+
|
43 |
+
Since this server is compatible with OpenAI API, you can use it as a drop-in replacement for any applications using OpenAI API.
|
44 |
+
For example, another way to query the server is via the openai python package:
|
45 |
+
```python
|
46 |
+
#!/usr/bin/env python
|
47 |
+
# coding=utf-8
|
48 |
+
|
49 |
+
import time
|
50 |
+
import asyncio
|
51 |
+
|
52 |
+
from openai import AsyncOpenAI
|
53 |
+
|
54 |
+
# Our system prompt
|
55 |
+
SYSTEM_PROMPT = (
|
56 |
+
"I am PodGPT, a large language model developed by the Kolachalama Lab in Boston, "
|
57 |
+
"specializing in science, technology, engineering, mathematics, and medicine "
|
58 |
+
"(STEMM)-related research and education, powered by podcast audio.\n"
|
59 |
+
"I provide information based on established scientific knowledge but must not offer "
|
60 |
+
"personal medical advice or present myself as a licensed medical professional.\n"
|
61 |
+
"I will maintain a consistently professional and informative tone, avoiding humor, "
|
62 |
+
"sarcasm, and pop culture references.\n"
|
63 |
+
"I will prioritize factual accuracy and clarity while ensuring my responses are "
|
64 |
+
"educational and non-harmful, adhering to the principle of 'do no harm'.\n"
|
65 |
+
"My responses are for informational purposes only and should not be considered a "
|
66 |
+
"substitute for professional consultation."
|
67 |
+
)
|
68 |
+
|
69 |
+
# Initialize the AsyncOpenAI client
|
70 |
+
client = AsyncOpenAI(
|
71 |
+
base_url="http://localhost:8000/v1",
|
72 |
+
api_key="token-abc123",
|
73 |
+
)
|
74 |
+
|
75 |
+
|
76 |
+
async def main(message):
|
77 |
+
"""
|
78 |
+
Streaming responses with async usage and "await" with each API call:
|
79 |
+
Reference: https://github.com/openai/openai-python?tab=readme-ov-file#streaming-responses
|
80 |
+
:param message: The user query
|
81 |
+
"""
|
82 |
+
start_time = time.time()
|
83 |
+
stream = await client.chat.completions.create(
|
84 |
+
model="shuyuej/Llama-3.3-70B-Instruct-GPTQ",
|
85 |
+
messages=[
|
86 |
+
{
|
87 |
+
"role": "system",
|
88 |
+
"content": SYSTEM_PROMPT,
|
89 |
+
},
|
90 |
+
{
|
91 |
+
"role": "user",
|
92 |
+
"content": message,
|
93 |
+
}
|
94 |
+
],
|
95 |
+
max_tokens=2048,
|
96 |
+
temperature=0.2,
|
97 |
+
top_p=1,
|
98 |
+
stream=True,
|
99 |
+
extra_body={
|
100 |
+
"ignore_eos": False,
|
101 |
+
# https://huggingface.co/shuyuej/Llama-3.3-70B-Instruct-GPTQ/blob/main/config.json#L10-L14
|
102 |
+
"stop_token_ids": [128001, 128008, 128009],
|
103 |
+
},
|
104 |
+
)
|
105 |
+
|
106 |
+
print(f"The user's query is\n {message}\n ")
|
107 |
+
print("The model's response is\n")
|
108 |
+
async for chunk in stream:
|
109 |
+
print(chunk.choices[0].delta.content or "", end="")
|
110 |
+
print(f"\nInference time: {time.time() - start_time:.2f} seconds\n")
|
111 |
+
print("=" * 100)
|
112 |
+
|
113 |
+
|
114 |
+
if __name__ == "__main__":
|
115 |
+
# Some random user queries
|
116 |
+
prompts = [
|
117 |
+
"Hello, my name is",
|
118 |
+
"The president of the United States is",
|
119 |
+
"The capital of France is",
|
120 |
+
"The future of AI is",
|
121 |
+
"Can you tell me more about Bruce Lee?",
|
122 |
+
"What are the differences between DNA and RNA?",
|
123 |
+
"What is dementia and Alzheimer's disease?",
|
124 |
+
"Tell me the differences between Alzheimer's disease and dementia"
|
125 |
+
]
|
126 |
+
|
127 |
+
# Conduct model inference
|
128 |
+
for message in prompts:
|
129 |
+
asyncio.run(main(message=message))
|
130 |
+
print("\n\n")
|
131 |
+
```
|
132 |
+
|
133 |
+
<details>
|
134 |
+
<summary>Here is a demo of the real-world model inference and deployment</summary>
|
135 |
+
<p align="center">
|
136 |
+
<a href="https://www.medrxiv.org/content/10.1101/2024.07.11.24310304v2"> <img src="figures/inference_demo.gif"></a>
|
137 |
+
</p>
|
138 |
+
</details>
|