Instructions to use deepgrove/maple-preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use deepgrove/maple-preview with Transformers:

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

pipe = pipeline("text-generation", model="deepgrove/maple-preview", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("deepgrove/maple-preview", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use deepgrove/maple-preview with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/deepgrove/maple-preview

SGLang

How to use deepgrove/maple-preview 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 "deepgrove/maple-preview" \
    --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": "deepgrove/maple-preview",
		"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 "deepgrove/maple-preview" \
        --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": "deepgrove/maple-preview",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use deepgrove/maple-preview with Docker Model Runner:
```
docker model run hf.co/deepgrove/maple-preview
```

What is this QAT'ed for?

by crusaderky - opened about 7 hours ago

Discussion

crusaderky

about 7 hours ago

•

edited about 7 hours ago

How should one quantize these weights?
With what quantization and inference engine were the speed benchmarks measured?

Q2_0 (stock llamacpp; called Q2_g64 on prism-ml's HF): one fp16 scale per 64 ternary weights
PQ2_0 (prism-ml's llamacpp fork; alias of Q2_0 on prism-ml's HF): one fp16 scale per 128 ternary weights
TQ2_0: one fp16 scale per 256 weights
other?

Green-Sky

about 4 hours ago

•

edited about 4 hours ago

I would like to know this as well, but q2_0 should always be usable as a fallback, if the block structure was trained right.

edit: found this in the config, and it gets read by the python code:

  "quantization": {
    "bits": 2,
    "group_size": 128,
    "lm_head": {
      "bits": 4,
      "group_size": 64
    },
    "mode": "affine",
    "model.word_embeddings": {
      "bits": 4,
      "group_size": 64
    }
  },

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