SeongryongJung/qwen3-4b-chemistry-rlsd-ema005
SeongryongJung/qwen3-4b-chemistry-rlsd-ema005 is a 4 billion parameter Qwen3-based causal language model fine-tuned by SeongryongJung. It is specifically optimized for chemistry-related tasks using Reinforcement Learning from Scientific Data (RLSD) with an EMA of 0.05 on a chemistry split. This model demonstrates specialized performance in scientific knowledge evaluation within the chemistry domain, achieving a 73.33% mean@16 on validation metrics.
Loading preview...
Model Overview
SeongryongJung/qwen3-4b-chemistry-rlsd-ema005 is a 4 billion parameter language model based on the Qwen3 architecture. It has been fine-tuned by SeongryongJung using Reinforcement Learning from Scientific Data (RLSD) with an Exponential Moving Average (EMA) of 0.05, specifically targeting the chemistry domain.
Key Capabilities
- Chemistry Domain Specialization: Optimized for tasks within the chemistry field through targeted fine-tuning on a chemistry-specific dataset.
- Performance: Achieved a
val-aux/sciknoweval/reward/mean@16score of 73.33% during validation, indicating proficiency in scientific knowledge evaluation within chemistry. - Training Method: Utilizes RLSD, a reinforcement learning approach, to enhance performance in its specialized domain.
Use Cases
This model is particularly well-suited for applications requiring deep understanding and generation of chemistry-related content. It can be leveraged for:
- Scientific knowledge extraction in chemistry.
- Assisting with chemistry-specific problem-solving.
- Generating or analyzing chemical information.
Technical Details
The model was fine-tuned from Qwen/Qwen3-4B and the uploaded weights correspond to the final global_step_100/actor checkpoint. The training process involved a learning rate of 1e-6 and a context length of 32768 tokens.