Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper • 2311.03099 • Published • 33
How to use Bytes512/Queen with Transformers:
# Use a pipeline as a high-level helper
from transformers import pipeline
pipe = pipeline("text-generation", model="Bytes512/Queen") # Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Bytes512/Queen")
model = AutoModelForCausalLM.from_pretrained("Bytes512/Queen")How to use Bytes512/Queen with vLLM:
# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Bytes512/Queen"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "Bytes512/Queen",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'docker model run hf.co/Bytes512/Queen
How to use Bytes512/Queen with SGLang:
# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
--model-path "Bytes512/Queen" \
--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": "Bytes512/Queen",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'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 "Bytes512/Queen" \
--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": "Bytes512/Queen",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'How to use Bytes512/Queen with Docker Model Runner:
docker model run hf.co/Bytes512/Queen
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using 01-ai/Yi-34B-200K as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: 01-ai/Yi-34B-200K
# No parameters necessary for base model
- model: abacusai/Smaugv0.1
parameters:
density: 0.53
weight: 0.3
- model: jondurbin/bagel-34b-v0.2
parameters:
density: 0.53
weight: 0.3
- model: NousResearch/Nous-Hermes-2-Yi-34B
parameters:
density: 0.53
weight: 0.4
merge_method: dare_ties
base_model: 01-ai/Yi-34B-200K
parameters:
int8_mask: true
dtype: bfloat16