KhushbooThaker commited on
Commit
436300e
·
verified ·
1 Parent(s): dec4134

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -4
README.md CHANGED
@@ -25,6 +25,8 @@ Unlike standard distillation methods that rely on unstructured Chain-of-Thought
25
 
26
  📄 **Paper:** [Knowledge Distillation with Structured Chain-of-Thought for Text-to-SQL](https://arxiv.org/abs/2512.17053)
27
 
 
 
28
  ## Performance
29
 
30
  On the **BIRD mini-dev** benchmark, Struct-SQL achieves an **Execution Accuracy (EX) of 45.0%**, outperforming standard unstructured CoT distillation baselines by **8.1 points**.
@@ -36,6 +38,7 @@ On the **BIRD mini-dev** benchmark, Struct-SQL achieves an **Execution Accuracy
36
  | FN-Gold Baseline | No Reasoning (SQL Only) | 34.3% |
37
  | Base Student (Zero-shot) | None | 17.0% |
38
 
 
39
  ## Methodology
40
 
41
  The model was trained on a curated dataset of **1,000 samples** generated by GPT-4o. The training data consists of:
@@ -44,10 +47,12 @@ The model was trained on a curated dataset of **1,000 samples** generated by GPT
44
 
45
  By forcing the model to explicitly plan the query execution (e.g., "Scan Table", "Filter by...", "Join with..."), the model learns the logical structure of SQL generation rather than just memorizing patterns.
46
 
 
47
  ## Usage
48
 
49
  You can use this model with the `transformers` library. It expects the input to be formatted with a specific system prompt or structure if you want to elicit the query plan.
50
 
 
51
  ```python
52
  import torch
53
  from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -65,8 +70,7 @@ inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
65
  outputs = model.generate(**inputs, max_new_tokens=1200)
66
  print(tokenizer.decode(outputs[0], skip_special_tokens=True))
67
  ```
68
-
69
-
70
  ## Intended Use
71
 
72
  Struct-SQL-4B is intended for **research and academic use** in tasks involving **Text-to-SQL generation** and **semantic parsing over relational databases**. The model is particularly suited for studying:
@@ -79,14 +83,12 @@ Struct-SQL-4B is intended for **research and academic use** in tasks involving *
79
  The model is not optimized for direct deployment in production database systems without additional validation and safety constraints.
80
 
81
  ---
82
-
83
  ## Limitations
84
 
85
  - Evaluation is confined to the SQLite-based BIRD benchmark
86
  - The model may generate logically plausible but incorrect SQL for highly complex multi-hop queries
87
 
88
  ---
89
-
90
  ## Citation
91
 
92
  ```bibtex
 
25
 
26
  📄 **Paper:** [Knowledge Distillation with Structured Chain-of-Thought for Text-to-SQL](https://arxiv.org/abs/2512.17053)
27
 
28
+ ---
29
+
30
  ## Performance
31
 
32
  On the **BIRD mini-dev** benchmark, Struct-SQL achieves an **Execution Accuracy (EX) of 45.0%**, outperforming standard unstructured CoT distillation baselines by **8.1 points**.
 
38
  | FN-Gold Baseline | No Reasoning (SQL Only) | 34.3% |
39
  | Base Student (Zero-shot) | None | 17.0% |
40
 
41
+ ---
42
  ## Methodology
43
 
44
  The model was trained on a curated dataset of **1,000 samples** generated by GPT-4o. The training data consists of:
 
47
 
48
  By forcing the model to explicitly plan the query execution (e.g., "Scan Table", "Filter by...", "Join with..."), the model learns the logical structure of SQL generation rather than just memorizing patterns.
49
 
50
+ ---
51
  ## Usage
52
 
53
  You can use this model with the `transformers` library. It expects the input to be formatted with a specific system prompt or structure if you want to elicit the query plan.
54
 
55
+ ---
56
  ```python
57
  import torch
58
  from transformers import AutoModelForCausalLM, AutoTokenizer
 
70
  outputs = model.generate(**inputs, max_new_tokens=1200)
71
  print(tokenizer.decode(outputs[0], skip_special_tokens=True))
72
  ```
73
+ ---
 
74
  ## Intended Use
75
 
76
  Struct-SQL-4B is intended for **research and academic use** in tasks involving **Text-to-SQL generation** and **semantic parsing over relational databases**. The model is particularly suited for studying:
 
83
  The model is not optimized for direct deployment in production database systems without additional validation and safety constraints.
84
 
85
  ---
 
86
  ## Limitations
87
 
88
  - Evaluation is confined to the SQLite-based BIRD benchmark
89
  - The model may generate logically plausible but incorrect SQL for highly complex multi-hop queries
90
 
91
  ---
 
92
  ## Citation
93
 
94
  ```bibtex