Instructions to use thxlxn/qwen3.5-0.8b-sql with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use thxlxn/qwen3.5-0.8b-sql 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 thxlxn/qwen3.5-0.8b-sql:F16
# Run inference directly in the terminal:
llama cli -hf thxlxn/qwen3.5-0.8b-sql:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf thxlxn/qwen3.5-0.8b-sql:F16
# Run inference directly in the terminal:
llama cli -hf thxlxn/qwen3.5-0.8b-sql:F16

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf thxlxn/qwen3.5-0.8b-sql:F16
# Run inference directly in the terminal:
./build/bin/llama cli -hf thxlxn/qwen3.5-0.8b-sql:F16

Use Docker

docker model run hf.co/thxlxn/qwen3.5-0.8b-sql:F16

Llama new
LM Studio
Jan
Ollama
How to use thxlxn/qwen3.5-0.8b-sql with Ollama:
```
ollama run hf.co/thxlxn/qwen3.5-0.8b-sql:F16
```
Unsloth Desktop

How to use thxlxn/qwen3.5-0.8b-sql with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf thxlxn/qwen3.5-0.8b-sql:F16

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "thxlxn/qwen3.5-0.8b-sql:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use thxlxn/qwen3.5-0.8b-sql with Docker Model Runner:
```
docker model run hf.co/thxlxn/qwen3.5-0.8b-sql:F16
```

Lemonade

How to use thxlxn/qwen3.5-0.8b-sql with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull thxlxn/qwen3.5-0.8b-sql:F16

Run and chat with the model

lemonade run user.qwen3.5-0.8b-sql-F16

List all available models

lemonade list

Hermes Agent

How to use thxlxn/qwen3.5-0.8b-sql with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf thxlxn/qwen3.5-0.8b-sql: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 thxlxn/qwen3.5-0.8b-sql:F16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use thxlxn/qwen3.5-0.8b-sql with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf thxlxn/qwen3.5-0.8b-sql:F16

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "thxlxn/qwen3.5-0.8b-sql:F16" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

GitHub Repo of the Pipeline
By using Qwen 3.5 9B as a teacher, a Qwen 3.5 0.8B student was fine-tuned for Text-to-SQL operations, recovering 95% of the teacher's accuracy while resolving multi-table JOIN hallucinations and running 2.3x faster.

Overview

Initial baseline testing of the Qwen 3.5 0.8B model on Text-to-SQL generation revealed a significant bottleneck: 72% of its failures were no_such_column errors when attempting multi-table JOIN operations.
The Goal:
Build an automated knowledge distillation pipeline to transfer complex, multi-table schema reasoning from a 9B parameter teacher model into a 0.8B parameter student model, achieving high execution validity while maintaining low VRAM usage and latency.

Pipeline

Teacher Model: qwen3.5:9b (Ollama)
Student Model: qwen3.5:0.8b (Unsloth)
Base Dataset: b-mc2/sql-create-context
Training Framework: PyTorch & Unsloth (16-bit LoRA)

1. Procedural Data Distillation (src/1_generate_dataset.py) The base dataset was filtered to strictly contain complex schemas requiring at least one JOIN operation (2,001 total rows). The Teacher model generated SQL for 1,698 training rows. Importantly, an automated SQLite verification step was implemented to execute the Teacher's outputs against the schema, rejecting 268 invalid queries and ending up with 1,430 verified training examples.

2. Fine-Tuning (src/3_train_lora.ipynb) The verified JSONL dataset was used to fine-tune the 0.8B Student using Unsloth (LoRA, r=16, alpha=32). The model was trained for 3 epochs (270 steps) with a cosine learning rate scheduler and an adamw_8bit optimizer, then merged and exported to a 16-bit GGUF format for local serving.

Benchmark Results

Evaluated against an unseen test pool of 303 multi-table queries. Accuracy is defined by successful and valid execution against an SQLite database matching the provided DDL schema.

Model	Parameters	Valid SQL Execution	Avg Latency	Column Hallucinations
Qwen 3.5 (Student)	0.8B	50.50% (153/303)	~209 ms	108
Qwen 3.5 (Teacher)	9B	84.49% (256/303)	~601 ms	20
Qwen 3.5-SQL (Distilled)	0.8B	82.84% (251/303)	~257 ms	24

Key Takeaways:

Hallucination Resolution: Schema-grounded distillation successfully reduced no_such_column errors by 78%.
Pareto Efficiency: The distilled student matches the 9B teacher's execution rate (within 5 queries) while running 2.34x faster.

license: apache-2.0

Downloads last month: 31

GGUF

Model size

0.8B params

Architecture

qwen35

Hardware compatibility

16-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for thxlxn/qwen3.5-0.8b-sql

Base model

Qwen/Qwen3.5-0.8B-Base

Finetuned

Qwen/Qwen3.5-0.8B

Adapter

(690)

this model

thxlxn
/

qwen3.5-0.8b-sql

Overview

Pipeline

Benchmark Results

license: apache-2.0

Model tree for thxlxn/qwen3.5-0.8b-sql

Dataset used to train thxlxn/qwen3.5-0.8b-sql