navflyer/shivik-m2.2

ziadrone / shivik-m2-aries

✅ Shivik-M2 (Aries infusion) — 1.1B reasoning-capable causal LM
This repository contains:

• model.safetensors (M2 weights)
• tokenizer files (vocab.json, merges.txt, tokenizer.json)
• modeling_shivik_m2.py (custom model class)
• tokenization_shivik_m1.py (custom HF-compatible Python tokenizer)
• helper scripts: build_tokenizer_fast.py, train_aries.py

Quick usage (after pip install transformers safetensors tokenizers)

from transformers import AutoTokenizer, AutoModelForCausalLM
tok = AutoTokenizer.from_pretrained("ziadrone/shivik-m2-aries", trust_remote_code=True, use_fast=False)
model = AutoModelForCausalLM.from_pretrained("ziadrone/shivik-m2-aries", trust_remote_code=True)
text = "Hello <think> explain step by step </think>"
enc = tok(text, return_tensors='pt')
out = model(**enc)