Instructions to use neosantara/wader-100m-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use neosantara/wader-100m-base with Transformers:

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

pipe = pipeline("text-generation", model="neosantara/wader-100m-base", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("neosantara/wader-100m-base", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("neosantara/wader-100m-base", trust_remote_code=True, device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use neosantara/wader-100m-base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "neosantara/wader-100m-base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "neosantara/wader-100m-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/neosantara/wader-100m-base

SGLang

How to use neosantara/wader-100m-base 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 "neosantara/wader-100m-base" \
    --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": "neosantara/wader-100m-base",
		"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 "neosantara/wader-100m-base" \
        --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": "neosantara/wader-100m-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use neosantara/wader-100m-base with Docker Model Runner:
```
docker model run hf.co/neosantara/wader-100m-base
```

wader-100m-base 🐟

A lightweight ~110M parameter Indonesian language model implementing the DeepSeek-V4 architecture (MLA, MoE, Hyper-Connections) trained from scratch on Indonesian corpus. This is the pretrained base model — see neosantara/wader-100m for the SFT/chat version.

Architecture

This model implements key DeepSeek-V4 innovations at a miniature scale:

Component	Details
Parameters	~110M total (41M embeddings, 69M non-embedding)
Hidden size	320
Layers	8
Attention heads	8 (1 KV head — MQA-style)
Head dim	96 (32 RoPE + 64 NoPE)
MLA	q_lora_rank=160, o_groups=2, o_lora_rank=80
MoE	4 routed experts + 1 shared, top-2 routing
Expert FFN	SwiGLU, intermediate_size=640
Routing	sqrtsoftplus scoring, noaux_tc method
Hyper-Connections	hc_mult=4, Sinkhorn routing (2 iters)
Vocab	129,280 (DeepSeek-V4 tokenizer)
Context	2,048 tokens

DeepSeek-V4 Features Implemented

Multi-head Latent Attention (MLA): Compressed KV cache via latent projections.
Mixture of Experts (MoE): Sparse activation — only 2 of 4 routed experts active per token.
Hyper-Connections: Multi-copy hidden states with learned Sinkhorn routing replacing standard residual connections.
SwiGLU FFN: High-capacity feed-forward blocks with non-linear gating.
Grouped output projection: Structured parameter-efficient output routing.

Training

Datasets:
- indonesian-nlp/eli5_id (70% - deep Indonesian explanations & reasoning)
- w11wo/twitter_indonesia_sarcastic (30% - informal & conversational Indonesian)
Steps: 5,000 steps (~35M+ tokens)
Batch size: 8 × 4 gradient accumulation = 32 effective
Sequence length: 2,048 tokens
Learning rate: 6e-4 (Cosine decay with warmup)
Optimizer: AdamW (β1=0.9, β2=0.95, weight_decay=0.1)
Precision: bf16 mixed precision
Hardware: NVIDIA H100 80GB HBM3 & RTX 4090 (Daytona Cloud)

Training Metrics

Metric	Value
Initial Loss	~10.8
Final Loss	~0.41 - 0.58 (cross-entropy)
Token Accuracy	~88% - 93%

Usage

import torch
from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file

# 1. Load config and model
repo_id = "neosantara/wader-100m-base"
config = AutoConfig.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_config(config, trust_remote_code=True).float()

# 2. Download and load weights
weights_path = hf_hub_download(repo_id, "model.safetensors")
state_dict = load_file(weights_path)
model.load_state_dict(state_dict, strict=True)
model = model.cuda().eval() if torch.cuda.is_available() else model.eval()

# 3. Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=True)

# 4. Generate Indonesian text completion
prompt = "Indonesia adalah negara kepulauan yang memiliki"
input_ids = tokenizer.encode(prompt, return_tensors="pt")
if torch.cuda.is_available():
    input_ids = input_ids.cuda()

with torch.no_grad():
    output = model.generate(
        input_ids,
        max_new_tokens=100,
        temperature=0.7,
        top_p=0.9,
        repetition_penalty=1.1,
        pad_token_id=tokenizer.eos_token_id,
    )

print(tokenizer.decode(output[0], skip_special_tokens=True))

Limitations

Small model size: 110M parameters designed for extreme edge efficiency and fast experimentation.
Pretrained only: This is a base model for text completion. For interactive chat, please use neosantara/wader-100m.
Custom architecture: Requires trust_remote_code=True.

License

Apache-2.0

Downloads last month: 337

Safetensors

Model size

0.1B params

Tensor type

F32

Model tree for neosantara/wader-100m-base

Finetunes

1 model

neosantara
/

wader-100m-base