Instructions to use Efe2898/caba-v1-scratch-2xt4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Efe2898/caba-v1-scratch-2xt4 with Transformers:

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

pipe = pipeline("text-generation", model="Efe2898/caba-v1-scratch-2xt4", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Efe2898/caba-v1-scratch-2xt4", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Efe2898/caba-v1-scratch-2xt4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Efe2898/caba-v1-scratch-2xt4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Efe2898/caba-v1-scratch-2xt4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Efe2898/caba-v1-scratch-2xt4

SGLang

How to use Efe2898/caba-v1-scratch-2xt4 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 "Efe2898/caba-v1-scratch-2xt4" \
    --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": "Efe2898/caba-v1-scratch-2xt4",
		"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 "Efe2898/caba-v1-scratch-2xt4" \
        --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": "Efe2898/caba-v1-scratch-2xt4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Efe2898/caba-v1-scratch-2xt4 with Docker Model Runner:
```
docker model run hf.co/Efe2898/caba-v1-scratch-2xt4
```

ÇABA v0-core — 2×T4 scratch training

This initial repository commit contains source, config, and tokenizer only. Training has not completed, so model weights are intentionally absent. The trained model and separate resumable optimizer checkpoint are uploaded after training completes or a graceful stop.

Architecture ID: caba-v0-core-shared-qkv-1-8-64. Planned trainable parameters: 49,701,952.

The repository contains tokenizer/config/model code, training and data-preparation scripts, notebook build/validation helpers, data audit, and metrics. Training data shards are not copied here. Review the dataset and tokenizer licenses before redistribution; no license is asserted by this card.

This is an experimental research prototype, not a validated benchmark result. The current implementation has no streaming recurrent cache. Plug-and-play loading becomes available after the final model weights are uploaded.

Downloads last month: 91