Instructions to use Ayaka157/RetNet-0.5B-ep2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Ayaka157/RetNet-0.5B-ep2 with Transformers:

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

pipe = pipeline("text-generation", model="Ayaka157/RetNet-0.5B-ep2", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Ayaka157/RetNet-0.5B-ep2", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Ayaka157/RetNet-0.5B-ep2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Ayaka157/RetNet-0.5B-ep2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Ayaka157/RetNet-0.5B-ep2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Ayaka157/RetNet-0.5B-ep2

SGLang

How to use Ayaka157/RetNet-0.5B-ep2 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 "Ayaka157/RetNet-0.5B-ep2" \
    --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": "Ayaka157/RetNet-0.5B-ep2",
		"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 "Ayaka157/RetNet-0.5B-ep2" \
        --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": "Ayaka157/RetNet-0.5B-ep2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Ayaka157/RetNet-0.5B-ep2 with Docker Model Runner:
```
docker model run hf.co/Ayaka157/RetNet-0.5B-ep2
```

RetNet-0.5B, epoch 2

Intermediate checkpoint of the three-leg run. Validation loss 1.9865. The final model is RetNet-0.5B at 1.9459.

Same architecture and configuration: 504,179,896 parameters, 20 layers, d_model 1408, 22 heads, vocab 20000, context 2048, chunk size 128.

Training

Second leg over chat_pack_v13 (552,330,489 token slots). RTX PRO 6000 96 GB, batch 16 × accum 1, 16,848 steps, 2.790 h wall clock, 54,970 tokens/s. Full cosine 1.1e-4 → 1.1e-5, warmup 500, AdamW 8-bit, seed 1234.

Its first evaluation lands 0.0764 above epoch 1's best (2.1035) before recovering to 1.9865.

Usage

Identical to the final model — trust_remote_code=True, standard AutoTokenizer:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "Ayaka157/RetNet-0.5B-ep2"
tok = AutoTokenizer.from_pretrained(repo, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(repo, trust_remote_code=True, dtype=torch.float32).eval()

ids = tok("<bos>用户：你好。\n助手：")["input_ids"]
out = model.generate(torch.tensor([ids]), max_new_tokens=64, temperature=0.7)
print(tok.decode(out[0].tolist()[len(ids):]))

As in the other checkpoints, mode in config.json is parallel for inference; the training run used chunkwise.

Evaluation numbers and the failure notes are on the final model card and in the technical report.

License

MIT.

Downloads last month: 251

Safetensors

Model size

0.5B params

Tensor type

F32