Instructions to use VishwamAI/vishwamai-base-v0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use VishwamAI/vishwamai-base-v0 with Transformers:

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

pipe = pipeline("text-generation", model="VishwamAI/vishwamai-base-v0")

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

tokenizer = AutoTokenizer.from_pretrained("VishwamAI/vishwamai-base-v0")
model = AutoModelForCausalLM.from_pretrained("VishwamAI/vishwamai-base-v0", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use VishwamAI/vishwamai-base-v0 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf VishwamAI/vishwamai-base-v0:F16
# Run inference directly in the terminal:
llama cli -hf VishwamAI/vishwamai-base-v0:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf VishwamAI/vishwamai-base-v0:F16
# Run inference directly in the terminal:
llama cli -hf VishwamAI/vishwamai-base-v0:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf VishwamAI/vishwamai-base-v0:F16
# Run inference directly in the terminal:
./llama-cli -hf VishwamAI/vishwamai-base-v0:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf VishwamAI/vishwamai-base-v0:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf VishwamAI/vishwamai-base-v0:F16

Use Docker

docker model run hf.co/VishwamAI/vishwamai-base-v0:F16

LM Studio
Jan

vLLM

How to use VishwamAI/vishwamai-base-v0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "VishwamAI/vishwamai-base-v0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "VishwamAI/vishwamai-base-v0",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/VishwamAI/vishwamai-base-v0:F16

SGLang

How to use VishwamAI/vishwamai-base-v0 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 "VishwamAI/vishwamai-base-v0" \
    --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": "VishwamAI/vishwamai-base-v0",
		"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 "VishwamAI/vishwamai-base-v0" \
        --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": "VishwamAI/vishwamai-base-v0",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Ollama
How to use VishwamAI/vishwamai-base-v0 with Ollama:
```
ollama run hf.co/VishwamAI/vishwamai-base-v0:F16
```
Unsloth Desktop
Docker Model Runner
How to use VishwamAI/vishwamai-base-v0 with Docker Model Runner:
```
docker model run hf.co/VishwamAI/vishwamai-base-v0:F16
```

Lemonade

How to use VishwamAI/vishwamai-base-v0 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull VishwamAI/vishwamai-base-v0:F16

Run and chat with the model

lemonade run user.vishwamai-base-v0-F16

List all available models

lemonade list

Atomic Chat

You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

VishwamAI (base)

Mixture-of-experts decoder-only language model trained from scratch (no distillation, no pre-trained weights) with JAX / Flax / Optax and a SentencePiece tokenizer (English, Hindi, Telugu).

Parameters: 133M total, 55M active per token (8 experts, top-2)
Architecture: RMSNorm, QK-norm, GQA (8 heads / 2 KV), SwiGLU routed experts, YaRN-extended RoPE (context 4096); exported in the Hugging Face Qwen3Moe layout. Multi-token-prediction heads were used in training only.
Training: pre-training on about 50M tokens (web/edu text, Wikipedia, FineWeb-2 Hindi and Telugu, code, SWE patches, cybersecurity text, math), a YaRN context-extension stage, SFT on human-written chat/instruction data, DPO on human preference rankings, and GRPO on programmatically verified arithmetic.
Evaluation: MMLU, MMMLU, GSM8K are used for evaluation only and were not trained on.

Limitations

Small research model trained on a small token budget: expect frequent factual and reasoning errors. It has not been red-teamed or safety-evaluated, and the persona/safety behaviour comes from a tiny seed set. Do not use it for anything that matters.

Results

{
 "res_pt": {
  "MMLU 5-shot (raw format)": 0.242,
  "MMLU 0-shot (chat format)": 0.241,
  "MMMLU-Hindi 0-shot (raw)": 0.237,
  "MMMLU-Hindi 0-shot (chat)": 0.216,
  "GSM8K (test, greedy)": 0.0,
  "arithmetic acc by #ops": {
   "1": 0.0,
   "2": 0.0,
   "3": 0.0
  },
  "GraphWalks-synthetic F1 by #edges": {
   "10": 0.0625,
   "40": 0.125,
   "150": 0.25,
   "400": 0.125
  }
 },
 "res_sft": {
  "MMLU 5-shot (raw format)": 0.241,
  "MMLU 0-shot (chat format)": 0.246,
  "MMMLU-Hindi 0-shot (raw)": 0.239,
  "MMMLU-Hindi 0-shot (chat)": 0.239,
  "GSM8K (test, greedy)": 0.035,
  "arithmetic acc by #ops": {
   "1": 0.109375,
   "2": 0.03125,
   "3": 0.0234375
  },
  "GraphWalks-synthetic F1 by #edges": {
   "10": 0.2791666666666667,
   "40": 0.18020833333333333,
   "150": 0.1875,
   "400": 0.0
  }
 },
 "res_rl": {
  "MMLU 5-shot (raw format)": 0.241,
  "MMLU 0-shot (chat format)": 0.244,
  "MMMLU-Hindi 0-shot (raw)": 0.236,
  "MMMLU-Hindi 0-shot (chat)": 0.249,
  "GSM8K (test, greedy)": 0.02,
  "arithmetic acc by #ops": {
   "1": 0.046875,
   "2": 0.03125,
   "3": 0.015625
  },
  "GraphWalks-synthetic F1 by #edges": {
   "10": 0.28125,
   "40": 0.07341269841269843,
   "150": 0.0,
   "400": 0.0
  }
 }
}

Downloads last month: 63

Safetensors

Model size

0.1B params

Tensor type

F32

Model tree for VishwamAI/vishwamai-base-v0

Finetunes

1 model

Collection including VishwamAI/vishwamai-base-v0

Vishwamai FOundation Models

Collection

2 items • Updated 2 days ago