Instructions to use xformAI/opt-125m-gqa-ub-6-best-for-KV-cache with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use xformAI/opt-125m-gqa-ub-6-best-for-KV-cache with Transformers:

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

pipe = pipeline("text-generation", model="xformAI/opt-125m-gqa-ub-6-best-for-KV-cache")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("xformAI/opt-125m-gqa-ub-6-best-for-KV-cache")
model = AutoModelForCausalLM.from_pretrained("xformAI/opt-125m-gqa-ub-6-best-for-KV-cache", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use xformAI/opt-125m-gqa-ub-6-best-for-KV-cache with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/xformAI/opt-125m-gqa-ub-6-best-for-KV-cache

SGLang

How to use xformAI/opt-125m-gqa-ub-6-best-for-KV-cache 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 "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache" \
    --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": "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache",
		"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 "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache" \
        --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": "xformAI/opt-125m-gqa-ub-6-best-for-KV-cache",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use xformAI/opt-125m-gqa-ub-6-best-for-KV-cache with Docker Model Runner:
```
docker model run hf.co/xformAI/opt-125m-gqa-ub-6-best-for-KV-cache
```

This is a GQA version of the original model facebook/opt-125m. In this version, the original MHA architecture is preserved but instead of having a single K/V head, different K/V heads corresponding to the same group have the same mean-pooled K or V values. It has 6 groups of KV heads per layer instead of original 12 KV heads in the MHA implementation.

Downloads last month: 109