Denisijcu commited on
Commit
cb9bdd2
·
verified ·
1 Parent(s): 64df983

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +43 -0
README.md CHANGED
@@ -1,3 +1,46 @@
 
1
  ---
 
 
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
  ---
3
+ language:
4
+ - en
5
  license: apache-2.0
6
+ library_name: transformers
7
+ tags:
8
+ - math
9
+ - reasoning
10
+ - qwen2.5
11
+ - gguf
12
+ - qlora
13
+ - vertex-coders
14
+ - edge-ai
15
+ base_model: Qwen/Qwen2.5-0.5B-Instruct
16
  ---
17
+
18
+ # Vertex-Math-0.5B (Q8_0)
19
+
20
+ **Vertex-Math-0.5B** is a compact mathematical reasoning language model fine-tuned from `Qwen2.5-0.5B-Instruct` using QLoRA. Optimized specifically for edge devices, offline mobile assistants, and resource-constrained environments, this model delivers rigorous step-by-step mathematical problem-solving while maintaining an ultra-lightweight footprint.
21
+
22
+ ## Model Details
23
+ - **Base Model:** [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct)
24
+ - **Architecture:** Qwen2 (0.5B parameters)
25
+ - **Training Dataset:** [HuggingFaceH4/MATH-500](https://huggingface.co/datasets/HuggingFaceH4/MATH-500) (Filtered and token-length audited)
26
+ - **Quantization:** Q8_0 (GGUF)
27
+ - **Developer:** Denis Sanchez Leyva (Vertex Coders LLC)
28
+
29
+ ## Training Configuration & Hyperparameters
30
+ - **Method:** QLoRA (Rank `r=8`, Alpha `16`)
31
+ - **Target Modules:** `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`
32
+ - **Hardware:** Local training on NVIDIA GTX 1650 (4GB VRAM)
33
+ - **Max Sequence Length:** 1024 tokens
34
+ - **Optimizer:** Paged AdamW 8-bit
35
+ - **Loss Function:** Fused Cross-Entropy Loss
36
+ - **Epochs:** 2 (80 steps)
37
+
38
+ ## Prompt Format
39
+ The model is trained using the ChatML template with a specialized system prompt enforcing step-by-step deductive reasoning:
40
+
41
+ ```text
42
+ <|im_start|>system
43
+ You are a compact mathematical reasoning engine optimized for mobile edge devices. Solve problems step by step with precision and conclude with '#### <final_answer>'.<|im_end|>
44
+ <|im_start|>user
45
+ [Your math problem here]<|im_end|>
46
+ <|im_start|>assistant