LumiCharles/genshin-starlit

Genshin Starlit

Genshin Starlit is a LLaMA-based causal language model fine-tuned for
roleplay and character-driven dialogue.

The model prioritizes immersive conversational flow and in-character consistency over strict factual accuracy.

---

Model Details

• Architecture: LlamaForCausalLM
• Parameters: ~71B
• Precision: bfloat16 (BF16)
• Context Length: up to 131k tokens
• Format: Safetensors (sharded)

---

Training and Composition

This model was created by merging two LoRA adapters into a single checkpoint:

• Character Dialogue Adapter
  Trained on character-specific dialogue transcripts to improve tone, personality, and speech consistency.

• Lore Knowledge Adapter
  Trained on lore-oriented question–answer style data to improve narrative coherence and lore-aware responses during roleplay.

The merged model balances immersive roleplay with contextual lore awareness during generation.

---

Intended Use

Recommended uses

• Roleplay and character simulation
• Creative writing and dialogue
• Interactive chat applications
• Narrative-driven assistants

Not recommended

• Factual or extractive question answering
• Retrieval-augmented generation
• Safety-critical or professional domains

---

Chat / Roleplay Usage

This model uses a chat template and is intended to be run in a conversational setting.

Example (Transformers)

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "your-username/genshin-starlit"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="bfloat16",
    device_map="auto"
)

messages = [
    {"role": "system", "content": "You are a fictional character engaging in immersive roleplay."},
    {"role": "user", "content": "Hello. Who are you?"}
]

inputs = tokenizer.apply_chat_template(
    messages,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    inputs,
    max_new_tokens=200,
    do_sample=True,
    temperature=0.8,
    top_p=0.9
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))