Update README.md

README.md

@@ -1,3 +1,69 @@
----
-license: apache-2.0
----
+---
+base_model: onekq-ai/OneSQL-v0.2-Qwen-7B
+tags:
+- text-generation-inference
+- transformers
+- qwen2
+- gguf
+license: apache-2.0
+language:
+- en
+---
+
+# Disclaimer
+Your email will be used for anonymous survey. It will NOT be shared with anyone.
+
+# Introduction
+
+This model is the GGUF version of [OneSQL-v0.2-Qwen-7B](https://huggingface.co/onekq-ai/OneSQL-v0.2-Qwen-7B).
+
+# Performances
+
+Below is the self-evaluation results for each quantization and its improvement over [OneSQL-v0.1-Qwen-7B-GGUF](https://huggingface.co/onekq-ai/OneSQL-v0.1-Qwen-7B-GGUF).
+
+| Quantization |EX score|v0.1 EX score
+|------------|-------|------|
+| Q2_K       | 39.60 | 29.79 |
+| Q3_K_S     | 38.79 | 36.31 |
+| Q3_K_M     | 41.93 | 39.24 |
+| Q3_K_L     | 45.49 | 40.14 |
+| Q4_1       | 39.06 |
+| Q4_K_S     | 42.69 |
+| **Q4_K_M** | 44.19 | **43.95** |
+| Q5_0       | 43.63 | 43.84 |
+| Q5_1       | 41.00 |
+| Q5_K_S     | 42.20 |
+| Q5_K_M     | 42.07 |
+| Q6_K       | 41.68 |
+| Q8_0       | 41.09 |
+
+# Quick start
+
+To use this model, craft your prompt to start with your database schema in the form of **CREATE TABLE**, followed by your natural language query preceded by **--**.
+Make sure your prompt ends with **SELECT** in order for the model to finish the query for you. There is no need to set other parameters like temperature or max token limit.
+
+```sh
+PROMPT="CREATE TABLE students (
+    id INTEGER PRIMARY KEY,
+    name TEXT,
+    age INTEGER,
+    grade TEXT
+);
+
+-- Find the three youngest students
+SELECT "
+
+PROMPT=$(printf "<|im_start|>system\nYou are a SQL expert. Return code only.<|im_end|>\n<|im_start|>user\n%s<|im_end|>\n<|im_start|>assistant\n" "$PROMPT")
+
+llama.cpp/build/bin/llama-run file://OneSQL-v0.2-Qwen-7B-Q4_K_M.gguf "$PROMPT"
+```
+
+The model response is the finished SQL query without **SELECT**
+```sql
+* FROM students ORDER BY age ASC LIMIT 3
+```
+
+# Caveats
+
+* The performance drop from the original model is due to quantization itself, and the lack of beam search support in llama.cpp framework. Use at your own discretion.
+* The 2-bit and 3-bit quantizations suffer from repetitive and unrelevant output token, hence are not recommended for usage.