Instructions to use OpenMOSE/Hy3-REAP-200B-21B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OpenMOSE/Hy3-REAP-200B-21B with Transformers:

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

pipe = pipeline("text-generation", model="OpenMOSE/Hy3-REAP-200B-21B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenMOSE/Hy3-REAP-200B-21B")
model = AutoModelForCausalLM.from_pretrained("OpenMOSE/Hy3-REAP-200B-21B", 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 OpenMOSE/Hy3-REAP-200B-21B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OpenMOSE/Hy3-REAP-200B-21B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenMOSE/Hy3-REAP-200B-21B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OpenMOSE/Hy3-REAP-200B-21B

SGLang

How to use OpenMOSE/Hy3-REAP-200B-21B 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 "OpenMOSE/Hy3-REAP-200B-21B" \
    --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": "OpenMOSE/Hy3-REAP-200B-21B",
		"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 "OpenMOSE/Hy3-REAP-200B-21B" \
        --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": "OpenMOSE/Hy3-REAP-200B-21B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use OpenMOSE/Hy3-REAP-200B-21B with Docker Model Runner:
```
docker model run hf.co/OpenMOSE/Hy3-REAP-200B-21B
```

Is this unusable

by void009 - opened 1 day ago

Discussion

void009

1 day ago

Hi. I got this to test on a 128gb strix halo. I gave it simple prompts like tetris, flappy bird in html. It gives me a very short result that includes some nonsense, repetition of tags etc. I also tried a settlers of catan prompt, while the planning phase seems to go reasonably well, when building it gets stuck outputting the tt.. sequence endlessly. Not sure I am doing something wrong.

HIP_VISIBLE_DEVICES=0 \
llama-server \
  --model "$HOME/models/IQ4_XS/Hy3-REAP-200B-GGUF-IQ4_XS-00001-of-*.gguf" \
  --chat-template-file "$HOME/models/hyv3_opensource_chat_template.jinja" \
  -ngl 999 -c 110000 -t "$(nproc)" \
  -ctk q8_0 -ctv q8_0 \
  --no-warmup --no-mmap \
  --host 0.0.0.0 --port 8080

GGML_VULKAN=1 \
llama-server \
  --model "$HOME/models/IQ4_XS/Hy3-REAP-200B-GGUF-IQ4_XS-00001-of-*.gguf" \
  --chat-template-file "$HOME/models/hyv3_opensource_chat_template.jinja" \
  -ngl 999 -c 65000 -t "$(nproc)" \
  --no-warmup --no-mmap \
  -np 2 \
  --host 0.0.0.0 --port 8080

tested with both rocm and vulkan, q8/q16 kv

OpenMOSE

Owner 1 day ago

thank you for feedback
will check

did you try to change repeat penalty, topp, topk, temperature?

OpenMOSE

Owner 1 day ago

•

edited 1 day ago

please try enabling thinking mode for coding
and can adjust repeat, temperature,topk

i tested on AMD w7900 x 2 (96GB)

 ./llama-server -m YOUR_MODEL_PATH -fa 1 -ctk q8_0 -ctv q8_0 --no-mmap  --top-k 40 --top-p 0.9 --temp 0.6 --repeat-penalty 1.1 --chat-template-kwargs '{"reasoning_effort": "high"}'

could make block break game with 20k token thinking..

please note
model is still experimental version. please stay tuned :)

void009

1 day ago

I tested with your settings, and now it is completely broken. in opencode, it thinks for 20 minutes and no output, even for a hello. In webchat, i say hello, it says aiaiaiaiaiaiaiaiaia.. hilarious

void009

about 23 hours ago

i removed -fa 1 and now it produces normal output. perhaps llamacpp bug, if yours is fine. do you use vulkan or rocm/

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