GreenMind-14B-R1
Parameters
14B
Context Length
33K
Modality
Text
Architecture
Dense
License
Apache-2.0
Release Date
23 Sept 2024
Knowledge Cutoff
Sep 2024
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
2x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
32,768 tokens
Consumer
2x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
No evaluation benchmarks for GreenMind-14B-R1 available.
Rankings
Overall Rank
-
Coding Rank
-
About GreenMind-14B-R1
GreenMind-14B-R1 is a 14.7 billion parameter Vietnamese reasoning model developed by GreenNode. It is a dense, decoder-only transformer derived from the Qwen2.5-14B-Instruct base architecture. The model is specifically engineered for multi-step logical reasoning and high-fidelity text generation in the Vietnamese language, addressing common limitations such as language mixing and factual drift in long-form reasoning chains. By implementing Chain-of-Thought (CoT) methodologies, GreenMind is designed to decompose complex queries into intermediate logical steps before producing a final response.
The model utilizes a specialized fine-tuning strategy known as Group Relative Policy Optimization (GRPO), which optimizes the reasoning process while maintaining computational efficiency. This training approach is augmented by a curated Vietnamese instruction dataset consisting of over 55,000 samples spanning cultural, legal, and educational domains. To ensure linguistic consistency, the training pipeline incorporates specific reward functions and Sentence Transformer-based verification to prevent the intrusion of non-Vietnamese characters and to preserve the factual integrity of the reasoning trajectories.
Optimized for deployment via NVIDIA NIM, GreenMind-14B-R1 is intended for enterprise-grade applications including legal and financial assistants, context-aware conversational agents, and complex document retrieval systems. The architecture supports a context length of up to 131,072 tokens for input processing, with a maximum generation limit of 8,192 tokens. Its integration of modern transformer techniques like RoPE position embeddings and SwiGLU activation makes it a technically sophisticated tool for localized AI infrastructure in Vietnam.
Technical Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
40
Key-Value Heads
8
Attention Head Dimension
-
Position Embedding
Absolute Position Embedding
RoPE Theta
-
Sliding Window Attention
-
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
5,120
Number of Layers
40
FFN Intermediate Size (Dense)
-
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
-
About GreenMind
GreenMind is an open-source Vietnamese reasoning language model family developed by GreenNode. It is optimized for multi-step reasoning tasks in Vietnamese, such as logic, mathematics, and scenario analysis. The model is designed to run efficiently on single-GPU hardware configurations.
Other GreenMind Models
- No related models available