kyleliu789/qwen3-14b-svamp14-sft-merged-bf16-r8-a16
The kyleliu789/qwen3-14b-svamp14-sft-merged-bf16-r8-a16 is a 14 billion parameter language model based on the Qwen3 architecture, fine-tuned for specific tasks. This model integrates a rank-8, alpha-16 SFT LoRA adapter into the Qwen/Qwen3-14B base model. It is designed to function as a standalone SFT model, potentially optimized for tasks related to its SVAMP14 fine-tuning, and serves as a foundation for further GRPO experiments. Its 32K context length supports processing longer sequences of text.
Loading preview...
Model Overview
This repository hosts the kyleliu789/qwen3-14b-svamp14-sft-merged-bf16-r8-a16 model, a 14 billion parameter language model built upon the Qwen/Qwen3-14B base architecture. It incorporates a fine-tuned SFT (Supervised Fine-Tuning) adapter, specifically a rank-8, alpha-16 LoRA adapter, which has been merged into the base model using PEFT's merge_and_unload function.
Key Characteristics
- Base Model: Qwen/Qwen3-14B
- Fine-tuning: Includes an SFT LoRA adapter (
kyleliu789/qwen3-14b-svamp14-sft-qlora-r8-a16) - Parameter Count: 14 billion parameters
- Context Length: 32,768 tokens
- Format: BF16 merged model, saved as six safetensors shards.
Intended Use
This merged model is designed to operate as a standalone SFT model, suggesting its optimization for specific tasks or domains it was fine-tuned on (indicated by "SVAMP14"). Furthermore, it is intended to serve as a parent model for subsequent GRPO experiments, highlighting its role in further research and development in reinforcement learning from human feedback or similar optimization techniques.