🤖 Mini-GPT — Multi-Task Assistant
Trained from scratch on 89,971 steps over 1.5 hours.
📊 Training Mix
| Task |
Dataset |
% |
| 💬 Chat + General |
OpenAssistant + UltraChat + Alpaca |
50% |
| 💻 Coding |
CodeAlpaca + Python Instructions |
30% |
| 🤖 Agentic |
Anthropic HH-RLHF |
10% |
| 🧮 Reasoning + Self |
GSM8K + Self-Generated |
10% |
🧠 Model Info
|
|
| Parameters |
3,808,939 |
| Layers |
6 |
| Embed dim |
256 |
| Heads |
8 |
| Block size |
128 |
| Vocab size |
1195 |
| Final loss |
0.8225 |
🚀 How to Load
import torch
from huggingface_hub import hf_hub_download
import json
ckpt_path = hf_hub_download("usernamebetter/idk-prev-2", "minigpt_checkpoint.pth")
vocab_path = hf_hub_download("usernamebetter/idk-prev-2", "vocab.json")
with open(vocab_path) as f:
vocab = json.load(f)
stoi = vocab["stoi"]
itos = {int(k): v for k, v in vocab["itos"].items()}
ckpt = torch.load(ckpt_path, map_location="cpu")
model.load_state_dict(ckpt["model"])
model.eval()
