Instructions to use jphme/Llama-2-13b-chat-german with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jphme/Llama-2-13b-chat-german with Transformers:

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

pipe = pipeline("text-generation", model="jphme/Llama-2-13b-chat-german")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("jphme/Llama-2-13b-chat-german")
model = AutoModelForCausalLM.from_pretrained("jphme/Llama-2-13b-chat-german", device_map="auto")

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use jphme/Llama-2-13b-chat-german with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jphme/Llama-2-13b-chat-german"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jphme/Llama-2-13b-chat-german",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/jphme/Llama-2-13b-chat-german

SGLang

How to use jphme/Llama-2-13b-chat-german 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 "jphme/Llama-2-13b-chat-german" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jphme/Llama-2-13b-chat-german",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "jphme/Llama-2-13b-chat-german" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jphme/Llama-2-13b-chat-german",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use jphme/Llama-2-13b-chat-german with Docker Model Runner:
```
docker model run hf.co/jphme/Llama-2-13b-chat-german
```

Currently the best LLM in german!

by danielplominski - opened Aug 17, 2023

Discussion

danielplominski

Aug 17, 2023

We have tested many LLMs locally on our infrastructure and rate "Llama-2-13b-chat-german" as #1.
On place 2 we see "h2ogpt-gm-oasst1-multilang-2048-falcon-7b".
Then nothing follows for a long time, since many other LLMs do not observe a good context in German.

"Llama-2-13b-chat-german" is currently running reliably at an average of 30 GB VRAM demand in a Proxmox - Ubuntu22 VM setup with NVIDIA Docker Runtime on 4x NVIDIA RTX A4000 as MultiGPU container environment.

Please continue to actively fine tune this LLM ;)

jphme

Owner Aug 21, 2023

thank you for your kind feedback, will release new tunes soon :-)

DennisCologne

Aug 30, 2023

Hey there,

I'm struggling to get the model running (see snipped)!

Can you help me with that? Thanks in advance!

leons05

Sep 16, 2023

thank you for your kind feedback, will release new tunes soon :-)

When can we expect an update ? I would really appreciate it.

jphme

Owner Sep 18, 2023

thank you for your kind feedback, will release new tunes soon :-)

When can we expect an update ? I would really appreciate it.

Ha many thanks for the push. Datasets are just being finalized and models will be trained this week.
I spent some time experimenting with pretraining the 7b model on a large German corpus to increase generation quality but this will probably only be ready for the next model version afterwarfds.

Which quantizations/formats are most important? Will try to publish different gptq/gguf versions and 7b/13b/34b versions this time.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment