Eve-Coder-272M

This is a Supervised Fine-Tuned (SFT) adapter for the anthonym21/Eve-2-MoE-272M base model.

Model Details

• Task: Python coding specialist fine-tuned on The Stack.
• Base Architecture: DeepSeek-MoE (Shared Expert + Routed Experts)
• Parameters: 272M (Total)
• Fine-Tuning Method: LoRA

How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# 1. Load Base Model
base_model = AutoModelForCausalLM.from_pretrained(
    "anthonym21/Eve-2-MoE-272M",
    trust_remote_code=True,
    device_map="cuda"
)

# 2. Load This Adapter
model = PeftModel.from_pretrained(base_model, "anthonym21/Eve-Coder-272M")

# 3. Run
tokenizer = AutoTokenizer.from_pretrained("anthonym21/Eve-2-MoE-272M")
inputs = tokenizer("Hello Eve!", return_tensors="pt").to("cuda")