Text Generation
Transformers
Safetensors
Russian
English
qwen3_5_text
text2sql
text-to-sql
sql
sqlite
russian
qwen3.5
conversational
Instructions to use MaXoN654/RUSQL-0.8B-Text2SQL with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use MaXoN654/RUSQL-0.8B-Text2SQL with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="MaXoN654/RUSQL-0.8B-Text2SQL") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("MaXoN654/RUSQL-0.8B-Text2SQL") model = AutoModelForCausalLM.from_pretrained("MaXoN654/RUSQL-0.8B-Text2SQL", 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=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use MaXoN654/RUSQL-0.8B-Text2SQL with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "MaXoN654/RUSQL-0.8B-Text2SQL" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "MaXoN654/RUSQL-0.8B-Text2SQL", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/MaXoN654/RUSQL-0.8B-Text2SQL
- SGLang
How to use MaXoN654/RUSQL-0.8B-Text2SQL 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 "MaXoN654/RUSQL-0.8B-Text2SQL" \ --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": "MaXoN654/RUSQL-0.8B-Text2SQL", "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 "MaXoN654/RUSQL-0.8B-Text2SQL" \ --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": "MaXoN654/RUSQL-0.8B-Text2SQL", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use MaXoN654/RUSQL-0.8B-Text2SQL with Docker Model Runner:
docker model run hf.co/MaXoN654/RUSQL-0.8B-Text2SQL
model card: aligned eval set (same 2717 questions EN/RU), RU 58.4%
Browse files
README.md
CHANGED
|
@@ -30,42 +30,44 @@ questions** over a database schema with step-by-step reasoning that ends in a fi
|
|
| 30 |
## Performance Evaluation
|
| 31 |
|
| 32 |
Execution accuracy (predicted SQL executed against SQLite, result-set comparison)
|
| 33 |
-
on the **
|
|
|
|
| 34 |
|
| 35 |
| Model | Questions | n | EX accuracy |
|
| 36 |
|---|---|---|---|
|
| 37 |
-
| Base (zero-shot) | EN |
|
| 38 |
-
| Base (zero-shot) | RU | 2,
|
| 39 |
-
| **RUSQL (this model)** | **RU** | **2,
|
| 40 |
|
| 41 |
-
Fine-tuning lifts execution accuracy **from
|
| 42 |
model, and well above its English-question ceiling (16.0%).
|
| 43 |
|
| 44 |
Breakdown by SQL complexity (RUSQL, RU questions):
|
| 45 |
|
| 46 |
| Complexity | n | EX accuracy |
|
| 47 |
|---|---|---|
|
| 48 |
-
| Simple |
|
| 49 |
-
| Moderate |
|
| 50 |
-
| Complex |
|
| 51 |
-
| Highly Complex |
|
| 52 |
|
| 53 |
Breakdown by question style (RUSQL, RU questions):
|
| 54 |
|
| 55 |
| Style | n | EX accuracy |
|
| 56 |
|---|---|---|
|
| 57 |
-
| Formal |
|
| 58 |
| Interrogative | 310 | 62.6% |
|
| 59 |
-
| Multi-turn Dialogue |
|
| 60 |
-
| Colloquial |
|
| 61 |
| Imperative | 324 | 58.6% |
|
| 62 |
| Descriptive | 312 | 58.3% |
|
| 63 |
-
| Concise |
|
| 64 |
| Vague | 310 | 55.5% |
|
| 65 |
| Metaphorical | 284 | 50.7% |
|
| 66 |
|
| 67 |
-
*
|
| 68 |
-
|
|
|
|
| 69 |
|
| 70 |
## Dataset Overview
|
| 71 |
|
|
|
|
| 30 |
## Performance Evaluation
|
| 31 |
|
| 32 |
Execution accuracy (predicted SQL executed against SQLite, result-set comparison)
|
| 33 |
+
on the **held-out eval split** — the same **2,717** questions in every row (EN vs RU
|
| 34 |
+
are the same items translated), greedy decoding:
|
| 35 |
|
| 36 |
| Model | Questions | n | EX accuracy |
|
| 37 |
|---|---|---|---|
|
| 38 |
+
| Base (zero-shot) | EN | 2,717 | 16.0% |
|
| 39 |
+
| Base (zero-shot) | RU | 2,717 | 14.0% |
|
| 40 |
+
| **RUSQL (this model)** | **RU** | **2,717** | **58.4%** |
|
| 41 |
|
| 42 |
+
Fine-tuning lifts execution accuracy **from 14.0% to 58.4%** — about 4.2× the base
|
| 43 |
model, and well above its English-question ceiling (16.0%).
|
| 44 |
|
| 45 |
Breakdown by SQL complexity (RUSQL, RU questions):
|
| 46 |
|
| 47 |
| Complexity | n | EX accuracy |
|
| 48 |
|---|---|---|
|
| 49 |
+
| Simple | 257 | 75.9% |
|
| 50 |
+
| Moderate | 856 | 70.2% |
|
| 51 |
+
| Complex | 968 | 55.1% |
|
| 52 |
+
| Highly Complex | 636 | 40.4% |
|
| 53 |
|
| 54 |
Breakdown by question style (RUSQL, RU questions):
|
| 55 |
|
| 56 |
| Style | n | EX accuracy |
|
| 57 |
|---|---|---|
|
| 58 |
+
| Formal | 306 | 62.8% |
|
| 59 |
| Interrogative | 310 | 62.6% |
|
| 60 |
+
| Multi-turn Dialogue | 268 | 60.1% |
|
| 61 |
+
| Colloquial | 334 | 59.3% |
|
| 62 |
| Imperative | 324 | 58.6% |
|
| 63 |
| Descriptive | 312 | 58.3% |
|
| 64 |
+
| Concise | 269 | 56.9% |
|
| 65 |
| Vague | 310 | 55.5% |
|
| 66 |
| Metaphorical | 284 | 50.7% |
|
| 67 |
|
| 68 |
+
*The RU eval questions are the QE-filtered translations of the held-out split, and
|
| 69 |
+
the EN/base rows are scored on the exact same items so the three numbers are directly
|
| 70 |
+
comparable. Greedy decoding; only 13 RUSQL generations failed to emit a SQL block.*
|
| 71 |
|
| 72 |
## Dataset Overview
|
| 73 |
|