Instructions to use codex-automatus/htdn-arabic-350m1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use codex-automatus/htdn-arabic-350m1 with Transformers:

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

pipe = pipeline("text-generation", model="codex-automatus/htdn-arabic-350m1")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("codex-automatus/htdn-arabic-350m1", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use codex-automatus/htdn-arabic-350m1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "codex-automatus/htdn-arabic-350m1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "codex-automatus/htdn-arabic-350m1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/codex-automatus/htdn-arabic-350m1

SGLang

How to use codex-automatus/htdn-arabic-350m1 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 "codex-automatus/htdn-arabic-350m1" \
    --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": "codex-automatus/htdn-arabic-350m1",
		"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 "codex-automatus/htdn-arabic-350m1" \
        --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": "codex-automatus/htdn-arabic-350m1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use codex-automatus/htdn-arabic-350m1 with Docker Model Runner:
```
docker model run hf.co/codex-automatus/htdn-arabic-350m1
```

HTDN — نموذج لغوي مخصّص (JAX/Flax)

نموذج HTDN (بنية مخصّصة تجمع بين طبقة ديناميكيات متكرّرة وانتباه ذاتي سببي مع RoPE و SwiGLU) دُرِّب على بيانات عربية متعددة المصادر (streaming)

⚠️ هذه ليست معمارية transformers قياسية، لذا لا يمكن تحميلها مباشرة عبر AutoModel.from_pretrained. الأوزان محفوظة بصيغة safetensors ومرفق كود بناء النموذج (modeling_htdn.py) لإعادة تحميلها عبر JAX/Flax.

طريقة التحميل

from huggingface_hub import hf_hub_download
import json, jax.numpy as jnp
from safetensors import safe_open
from flax.traverse_util import unflatten_dict

# 1) نزّل الملفات
config_path = hf_hub_download("gijl/htdn-arabic-350m1", "config.json")
weights_path = hf_hub_download("gijl/htdn-arabic-350m1", "model.safetensors")
modeling_path = hf_hub_download("gijl/htdn-arabic-350m1", "modeling_htdn.py")

with open(config_path) as f:
    cfg = json.load(f)

# 2) حمّل الأوزان
flat = {}
with safe_open(weights_path, framework="numpy") as f:
    for k in f.keys():
        flat[k] = jnp.array(f.get_tensor(k))
params = unflatten_dict(flat, sep=".")

# 3) أعد بناء النموذج (انظر modeling_htdn.py المرفق)
import importlib.util
spec = importlib.util.spec_from_file_location("modeling_htdn", modeling_path)
modeling_htdn = importlib.util.module_from_spec(spec)
spec.loader.exec_module(modeling_htdn)

model, _ = modeling_htdn.build_model_template(cfg)
# استخدم `params` أعلاه بدل المعاملات العشوائية من build_model_template

معلومات التدريب

عدد المعاملات: ~350M
المحلل اللغوي: bert-base-multilingual-cased

هذا نموذج تجريبي/بحثي — النتائج أولية.

Downloads last month: 185

Safetensors

Model size

0.3B params

Tensor type

F32

codex-automatus
/

htdn-arabic-350m1

HTDN — نموذج لغوي مخصّص (JAX/Flax)

طريقة التحميل

معلومات التدريب

Space using codex-automatus/htdn-arabic-350m1 1