Instructions to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Transformers:

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

pipe = pipeline("text-generation", model="harshitsiwach/qwen-3.5-0.8B-solana-baby-architect")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("harshitsiwach/qwen-3.5-0.8B-solana-baby-architect")
model = AutoModelForMultimodalLM.from_pretrained("harshitsiwach/qwen-3.5-0.8B-solana-baby-architect")
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]:]))

llama-cpp-python

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="harshitsiwach/qwen-3.5-0.8B-solana-baby-architect",
	filename="qwen3.5-solana-v2-f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16
# Run inference directly in the terminal:
llama-cli -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16
# Run inference directly in the terminal:
llama-cli -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect: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 harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16
# Run inference directly in the terminal:
./llama-cli -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect: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 harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

Use Docker

docker model run hf.co/harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

LM Studio
Jan

vLLM

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

SGLang

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect 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 "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect" \
    --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": "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect",
		"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 "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect" \
        --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": "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Ollama:
```
ollama run hf.co/harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16
```

Unsloth Studio

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for harshitsiwach/qwen-3.5-0.8B-solana-baby-architect to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for harshitsiwach/qwen-3.5-0.8B-solana-baby-architect to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for harshitsiwach/qwen-3.5-0.8B-solana-baby-architect to start chatting

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

Run Hermes

hermes

Atomic Chat new
Docker Model Runner
How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Docker Model Runner:
```
docker model run hf.co/harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16
```

Lemonade

How to use harshitsiwach/qwen-3.5-0.8B-solana-baby-architect with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull harshitsiwach/qwen-3.5-0.8B-solana-baby-architect:F16

Run and chat with the model

lemonade run user.qwen-3.5-0.8B-solana-baby-architect-F16

List all available models

lemonade list

Qwen 3.5 0.8B — Solana Baby Architect 👶🚀

This model is a highly specialized version of Qwen/Qwen3.5-0.8B-Base fine-tuned on 2,189 examples of Solana conceptual knowledge, structural Rust/Anchor code, and security audit patterns. It is designed as a lightweight, offline-capable "Baby" version that is still learning the nuances of the Solana ecosystem.

🛠 Project Status: Work in Progress (WIP)

This model is currently in a "Junior Architect" phase. It is a work in progress and serves as a technical preview of what is possible with extremely small (0.8B) models on consumer hardware.

🧠 Technical Limitations & Intelligence Growth

This model was trained on limited resources: an NVIDIA GeForce RTX 4090 Laptop GPU (16GB VRAM). To fit within these limits while training on a large dataset, certain trade-offs were made.

Current Intelligence (Junior Architect)

Sequence Length: 1024 tokens.
Training Depth: 3 epochs.
Known Issues: May exhibit "junior level" hallucinations on very complex Anchor programs or mix up deep-level Solana internals (e.g., specific PoH/Bank struct details).

How to Achieve "Senior Architect" Level

If you have access to professional-grade hardware (24GB+ VRAM or A100/H100), you can significantly upgrade this model's intelligence by adjusting the provided training scripts:

Increase Sequence Length (MAX_SEQ_LENGTH):
- Upgrade to: 2048 or 4096 tokens.
- Result: Allows the model to synthesize entire multi-file smart contracts and maintain consistency across complex state structs.
Increase Training Depth (EPOCHS):
- Upgrade to: 6 to 10 epochs.
- Result: Deepens technical intuition and eliminates conceptual hallucinations.
Lower Learning Rate: Using a learning rate of 5e-5 with more epochs will result in much finer precision for security auditing.

🚀 Train This Yourself

We have included the full training infrastructure in this repository so the community can continue to improve this model.

Requirements

A CUDA-capable GPU (16GB+ VRAM recommended).
The train_solana_expert_v2.jsonl dataset (included).

Steps

Environment: Run setup_train.bat to create the virtual environment.

Run Training:

venv\Scripts\python.exe train_solana_v2.py

Customize: Edit train_solana_v2.py to increase MAX_SEQ_LENGTH and EPOCHS as described above.

📱 Deployment Targets

iOS: Optimized for high-precision FP16 GGUF inference on iPhone.
Web: Targets 4-bit ONNX for WebGPU-accelerated browser applications.

Trained by: harshitsiwach

Downloads last month: 119

Safetensors

Model size

0.8B params

Tensor type

BF16

Model tree for harshitsiwach/qwen-3.5-0.8B-solana-baby-architect

Base model

Qwen/Qwen2.5-0.5B

Quantized

(107)

this model