SeongryongJung/qwen3-8b-material-rlsd-ema005
SeongryongJung/qwen3-8b-material-rlsd-ema005 is an 8 billion parameter Qwen3-based language model fine-tuned with RLSD (EMA 0.05) specifically on the 'material' split of a dataset. This model demonstrates a validation performance of 79.06% on the 'val-aux/sciknoweval/reward/mean@16' metric. It is optimized for tasks related to material science or scientific knowledge evaluation, leveraging its specialized fine-tuning for improved accuracy in this domain.
Loading preview...
Model Overview
SeongryongJung/qwen3-8b-material-rlsd-ema005 is an 8 billion parameter language model, fine-tuned from the Qwen/Qwen3-8B base model. Its unique characteristic lies in its fine-tuning process, which utilized RLSD (EMA 0.05) on the material split of a dataset, indicating a specialization in material science or related scientific knowledge domains.
Key Performance
The model's performance is validated using the val-aux/sciknoweval/reward/mean@16 metric. It achieved a best mean@16 score of 79.06% at step 100 during its validation process. This metric suggests its proficiency in tasks related to scientific knowledge evaluation, particularly within the 'material' context.
Training Details
The model was trained with specific parameters including mbs8, decay0, ema0.05, train64, rollout8, and lr1e-6. The uploaded weights correspond to the final global_step_100/actor checkpoint, converted from VERL FSDP shards to the Hugging Face format. This specialized training approach aims to enhance its capabilities for domain-specific applications.
Intended Use Cases
This model is particularly well-suited for applications requiring deep understanding and generation within the material science domain or for tasks involving scientific knowledge evaluation where the 'material' context is relevant. Its fine-tuning on a specific data split makes it a strong candidate for specialized scientific research and development.