Hinglish-Gemma-4B-E4b — LoRA GGUF

GGUF export of the Hinglish-Gemma-4B-E4b LoRA adapter, for use with llama.cpp and Ollama. It makes Gemma 4 E4B chat like your desi best friend — in casual Hinglish.

This is the adapter only (73 MB) — pair it with any Gemma 4 E4B base GGUF, e.g. unsloth/gemma-4-E4B-it-GGUF (Q4_K_M recommended).

llama.cpp

llama-cli -m gemma-4-E4B-it-Q4_K_M.gguf \
          --lora Hinglish-Gemma-4B-E4b-lora-f16.gguf \
          -sys "You are Yaar, a friendly desi buddy. You always reply in casual Hinglish (Hindi written in English letters, mixed with English), like a close friend texting. Keep replies short, warm and natural." \
          -p "yaar kal exam hai aur kuch padha nahi"

llama-cpp-python

from llama_cpp import Llama
llm = Llama(model_path="gemma-4-E4B-it-Q4_K_M.gguf",
            lora_path="Hinglish-Gemma-4B-E4b-lora-f16.gguf", n_ctx=1024)

Ollama

# Modelfile
FROM ./gemma-4-E4B-it-Q4_K_M.gguf
ADAPTER ./Hinglish-Gemma-4B-E4b-lora-f16.gguf
SYSTEM "You are Yaar, a friendly desi buddy. You always reply in casual Hinglish (Hindi written in English letters, mixed with English), like a close friend texting. Keep replies short, warm and natural."

ollama create yaar -f Modelfile && ollama run yaar "yaar aaj mood off hai"

🧠 Model Lineage

Fine-tuned by Vedant Rajpurohit from Google's Gemma 4 E4B on the Abhishekcr448/Hinglish-Everyday-Conversations-1M dataset (~52k curated examples, QLoRA r=16). Full training details on the adapter card.

If you use this, a mention or backlink is appreciated 💛