TarunNagaSai007/gemma4-e2b-pokemon-merged
TarunNagaSai007/gemma4-e2b-pokemon-merged is a 5.1 billion parameter Gemma-4-E2B-IT model fine-tuned and merged to 16-bit for Pokémon-specific tasks. It excels as a Pokédex assistant, providing accurate information on Pokémon stats, profiles, and battle outcomes. This model is optimized for direct use without separate adapters, offering specialized knowledge within the Pokémon domain.
Loading preview...
Overview
This model, TarunNagaSai007/gemma4-e2b-pokemon-merged, is a 5.1 billion parameter variant of the google/gemma-4-e2b-it base model. It has been fine-tuned using LoRA on a specialized Pokémon dataset and subsequently merged into a standalone 16-bit model, eliminating the need for separate adapters. This allows for direct loading and execution, simplifying deployment.
Key Capabilities
- Pokémon Stat Retrieval: Accurately provides specific statistics for Pokémon (e.g., Speed of Duskull).
- Pokédex Profile Generation: Generates detailed Pokédex entries for various Pokémon (e.g., "Tell me about Miltank.").
- Battle Outcome Prediction: Offers reasoning and verdicts for hypothetical Pokémon battles, utilizing a simplified type/offense/speed heuristic.
Training Details
The model was trained on approximately 8,600 instruction examples covering stat, profile, and battle queries, with an additional 957 validation and 1,200 test examples. Training involved 3 epochs with an effective batch size of 8 and a learning rate of 2e-4 using the adamw_8bit optimizer. The final validation loss achieved was 0.164.
Limitations
It's important to note that the model's knowledge is based on a fixed Pokédex snapshot. Its battle reasoning employs a simplified heuristic and does not incorporate full damage mechanics, making it suitable primarily for educational or hobby use.