MINDI-1.0-420M

MINDI 1.0 420M

MINDI 1.0 420M is a 420M-parameter coding language model focused on Python first and JavaScript second. It is built for local, offline code generation workflows.

Capabilities

• Code generation from natural language prompts
• Code completion
• Bug-fix suggestions
• Code explanation

Model Details

• Parameters: 423,934,848
• Architecture: Decoder-only Transformer
• Context length: 2048 tokens
• Focus languages: Python, JavaScript

Hardware Requirements

Recommended:

• NVIDIA GPU with 8GB+ VRAM
• CUDA-enabled PyTorch

Minimum:

• CPU inference works but is slower

Quick Start (GPU)

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

repo_id = "YOUR_USERNAME/MINDI-1.0-420M"

tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    repo_id,
    trust_remote_code=True,
    torch_dtype=torch.float16,
).cuda()

prompt = "Write a Python function to check if a string is a palindrome."
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=220,
        temperature=0.2,
        top_p=0.9,
        do_sample=True,
    )

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

Limitations

• The model can still produce syntax or logic errors.
• Generated code should always be reviewed and tested.
• Not intended for safety-critical production use without validation.

Safety

Always run tests and static checks before using generated code in production.