Instructions to use taeminlee/KULLM3-awq with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use taeminlee/KULLM3-awq with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="taeminlee/KULLM3-awq")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("taeminlee/KULLM3-awq")
model = AutoModelForCausalLM.from_pretrained("taeminlee/KULLM3-awq", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use taeminlee/KULLM3-awq with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "taeminlee/KULLM3-awq"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "taeminlee/KULLM3-awq",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/taeminlee/KULLM3-awq

SGLang

How to use taeminlee/KULLM3-awq with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "taeminlee/KULLM3-awq" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "taeminlee/KULLM3-awq",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "taeminlee/KULLM3-awq" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "taeminlee/KULLM3-awq",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use taeminlee/KULLM3-awq with Docker Model Runner:
```
docker model run hf.co/taeminlee/KULLM3-awq
```

taeminlee commited on Apr 8, 2024

Commit

9958424

1 Parent(s): 3116fff

update README.md

Browse files

Files changed (1) hide show

README.md +5 -1

README.md CHANGED Viewed

@@ -21,6 +21,10 @@ The quantization was carried out in a custom branch of [autoawq](https://github.
 It worked using [vllm](https://github.com/vllm-project/vllm). It may not work with other frameworks as they have not been tested.
 # KULLM3
 Introducing KULLM3, a model with advanced instruction-following and fluent chat abilities.
 It has shown remarkable performance in instruction-following, speficially by closely following gpt-3.5-turbo.
@@ -95,7 +99,7 @@ _ = model.generate(inputs, streamer=streamer, max_new_tokens=1024)
 ### Results
-<img src="kullm3_instruction_evaluation.png" width=100%>
 ## Citation

 It worked using [vllm](https://github.com/vllm-project/vllm). It may not work with other frameworks as they have not been tested.
+Below is the README.md for the original model.
+---
 # KULLM3
 Introducing KULLM3, a model with advanced instruction-following and fluent chat abilities.
 It has shown remarkable performance in instruction-following, speficially by closely following gpt-3.5-turbo.
 ### Results
+<img src="https://huggingface.co/nlpai-lab/KULLM3/resolve/main/kullm3_instruction_evaluation.png" width=100%>
 ## Citation