You can deploy this model and start earning money today!

TarunNagaSai007/
gemma4-e2b-pokemon

Name: gemma4-e2b-pokemon
Availability: InStock

Views

No views yet

Gemma 4 E2B — Pokémon Pokédex (LoRA Adapter)

LoRA adapter fine-tuned on google/gemma-4-e2b-it to answer Pokémon questions across three task types: stat lookups, full Pokédex profiles, and battle-outcome reasoning with chain-of-thought.

Model Details

Base model: google/gemma-4-e2b-it
Method: LoRA (16-bit), trained with Unsloth
Trainable params: ~31M (0.60% of base)
LoRA config: r=16, alpha=16, dropout=0, targets q/k/v/o/gate/up/down projections

Tasks

Task	Input example	Output
Stat	"What is the Speed of Duskull?"	`25`
Profile	"Tell me about Miltank."	Full Pokédex entry (type, stats, abilities, evolution, dex text)
Battle	"If Raichu battles Blastoise, who wins?"	`<think>` reasoning block + verdict

Usage

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="TarunNagaSai007/gemma4-e2b-pokemon",
    max_seq_length=2048,
    load_in_4bit=False,
)
FastLanguageModel.for_inference(model)

messages = [
    {"role": "system", "content": [{"type": "text", "text": "You are a Pokédex assistant. Answer questions about Pokémon stats, profiles, and battle outcomes accurately."}]},
    {"role": "user", "content": [{"type": "text", "text": "If Charizard battles Venusaur, who wins?"}]},
]
inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")
out = model.generate(input_ids=inputs, max_new_tokens=512, temperature=0.3)
print(tokenizer.decode(out[0], skip_special_tokens=True))

Training

Dataset: ~8,600 instruction examples (stat / profile / battle), 957 validation, 1,200 test
Epochs: 3 | Effective batch: 8 | LR: 2e-4 cosine, 100 warmup steps
Optimizer: adamw_8bit | Hardware: single T4 (Colab)
Final validation loss: 0.164

Data Training Report

Training loss and Validation loss compraistion

Comparesation results between base model and Fine-tuned model

Limitations

Trained on a fixed Pokédex snapshot; stats and profiles reflect that data. Battle reasoning uses a simplified type/offense/speed heuristic, not full damage-calc mechanics. Built for educational/hobby use.