DedeProGames commited on
Commit
123017a
·
verified ·
1 Parent(s): c729bfb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +37 -26
README.md CHANGED
@@ -1,51 +1,62 @@
1
  ---
 
 
 
 
2
  license: apache-2.0
3
  tags:
4
- - unsloth
5
- - trl
6
- - sft
7
  - code
 
8
  - reasoning
 
 
 
 
 
 
9
  datasets:
10
  - nvidia/OpenCodeReasoning
11
- language:
12
- - en
13
  base_model:
14
  - Qwen/Qwen3-0.6B
15
- pipeline_tag: text-generation
16
- library_name: transformers
17
  ---
18
 
19
- # Qwen3-0.6B-Code-Expert
 
 
20
 
21
- This project performs full fine-tuning on the **Qwen3-0.6B** language model to enhance its code reasoning and generation capabilities. Training was conducted exclusively on the `nvidia/OpenCodeReasoning` dataset, and the model was optimized using the bfloat16 (bf16) data type.
22
 
23
- ## Training Procedure
24
 
25
- 1. **Dataset Preparation**
26
 
27
- * `nvidia/OpenCodeReasoning` dataset was used.
28
- * Each example consists of code snippets paired with detailed step-by-step reasoning in Chain-of-Thought (CoT) style.
29
 
30
- 2. **Model Loading and Configuration**
31
 
32
- * Qwen3-0.6B base model weights were loaded via the `unsloth` library in bf16 precision.
33
- * Full fine-tuning (`full_finetuning=True`) was applied to all layers for optimal adaptation to code reasoning.
34
 
35
- 3. **Supervised Fine-Tuning**
36
 
37
- * Employed the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach.
38
- * The model was trained to generate correct code solutions along with the corresponding reasoning chains.
39
 
40
- ## Purpose and Outcome
 
 
 
 
 
41
 
42
- * The model’s capacity for understanding, reasoning about, and generating code was significantly improved through specialized, single-dataset training in bf16 precision.
43
- * Outputs include both intermediate reasoning steps and final code solutions, enabling transparent and interpretable code generation.
44
 
45
- ## License
46
 
47
- This project is licensed under the Apache License 2.0. See the [LICENSE](./LICENSE) file for details.
48
 
49
- ## Support
50
 
51
- <a href="https://www.buymeacoffee.com/suayptalha" target="_blank"><img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" style="height: 60px !important;width: 217px !important;" ></a>
 
 
 
 
1
  ---
2
+ language:
3
+ - en
4
+ - code
5
+ pipeline_tag: text-generation
6
  license: apache-2.0
7
  tags:
8
+ - coderion
 
 
9
  - code
10
+ - coding
11
  - reasoning
12
+ - small-language-model
13
+ - 0.6b
14
+ - chronological-reasoning
15
+ - high-reasoning
16
+ - compact-model
17
+ library_name: transformers
18
  datasets:
19
  - nvidia/OpenCodeReasoning
 
 
20
  base_model:
21
  - Qwen/Qwen3-0.6B
 
 
22
  ---
23
 
24
+ <p align="center">
25
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/685ea8ff7b4139b6845ce395/1z7OO6Xv_EWEHUDemqSL1.png" alt="logo" width="200">
26
+ </p>
27
 
28
+ <h1 align="center">Coderion</h1>
29
 
30
+ <p align="center"><b>A compact 0.6B coding model built for strong reasoning efficiency.</b></p>
31
 
32
+ ---
33
 
34
+ **Coderion** is a **small 0.6B parameter coding-focused language model** designed for **high and xhigh chronological reasoning** in programming tasks.
 
35
 
36
+ It is built to deliver **surprisingly strong structured reasoning and coding performance for its size**, focusing on consistency, logical step progression, and efficient problem solving.
37
 
38
+ While **Coderion is not intended to be a general everyday assistant**, it is a **small but capable specialist model** that performs well within its class and remains **reliable for compact code reasoning workloads**.
 
39
 
40
+ ---
41
 
42
+ ## Key Characteristics
 
43
 
44
+ - **0.6B parameters**
45
+ - **Dedicated to code**
46
+ - **Optimized for high reasoning intensity**
47
+ - **Chronological reasoning style**
48
+ - **Strong consistency for a compact model**
49
+ - **Designed for efficient performance despite its small size**
50
 
51
+ ---
 
52
 
53
+ ## Limitations
54
 
55
+ Coderion is a **small specialized model**.
56
 
57
+ Because of that:
58
 
59
+ - It may not match larger models on broad real-world assistant tasks
60
+ - It is not primarily designed for daily casual use
61
+ - It performs best when used for **focused coding and reasoning workloads**
62
+ - Its main strength is **efficiency, consistency, and reasoning quality relative to size**