Instructions to use Lobus22/Azco-Roven-9B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Lobus22/Azco-Roven-9B with Transformers:

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

pipe = pipeline("text-generation", model="Lobus22/Azco-Roven-9B")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Lobus22/Azco-Roven-9B")
model = AutoModelForCausalLM.from_pretrained("Lobus22/Azco-Roven-9B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Lobus22/Azco-Roven-9B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Lobus22/Azco-Roven-9B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Lobus22/Azco-Roven-9B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Lobus22/Azco-Roven-9B

SGLang

How to use Lobus22/Azco-Roven-9B 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 "Lobus22/Azco-Roven-9B" \
    --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": "Lobus22/Azco-Roven-9B",
		"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 "Lobus22/Azco-Roven-9B" \
        --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": "Lobus22/Azco-Roven-9B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Lobus22/Azco-Roven-9B with Docker Model Runner:
```
docker model run hf.co/Lobus22/Azco-Roven-9B
```

Azco-Roven-9B

Full-parameter fine-tune of 01-ai/Yi-9B (8,829,407,232 params, 100% weights trained, no LoRA/adapters) produced on 2 x Tesla T4 (Kaggle) with a paged LOMO-style trainer: weights live in CPU memory, one transformer block is paged onto the GPU at a time, gradients are fused into an immediate Adafactor update and freed.

Measured footprint

VRAM peak: 2.11 GB (reserved 2.56 GB)
Host RAM peak: 21.9 GB
22 optimizer steps, 21,087 tokens processed in 36.5 min
99.49 s/step, 9.63 tokens/s
Loss: 1.2746 -> 1.21695 (best 0.67465)
Forward equivalence check vs official HF forward: max abs diff None

Data

tatsu-lab/alpaca, 1996 samples (174,571 tokens, avg 87.5/sample), formatted ### Instruction / ### Response.

Limits

Proof-of-concept artifact for memory-constrained full fine-tuning. Only a small number of optimizer steps over a subset were completed, so behavioural change vs the base model is limited. Not an instruction-tuned production model.

Downloads last month: -

Safetensors

Model size

9B params

Tensor type

F16

Model tree for Lobus22/Azco-Roven-9B

Base model

01-ai/Yi-9B

Finetuned

(5)

this model