sumeru-rm / modeling_sumeru.py
debowd's picture
Sumeru rebrand: SumeruForCausalLM, auto_map, scrubbed metadata
4048ea2 verified
"""Sumeru model class — thin wrapper over Gemma 4 architecture.
Copyright (c) 2026 Sumeru AI. Licensed under Apache 2.0.
Loaded automatically via:
from transformers import AutoModelForImageTextToText, AutoProcessor
model = AutoModelForImageTextToText.from_pretrained(
"debowd/sumeru-rm", trust_remote_code=True
)
"""
from transformers import Gemma4ForConditionalGeneration
from .configuration_sumeru import SumeruConfig
class SumeruForCausalLM(Gemma4ForConditionalGeneration):
"""Sumeru causal language model.
Sumeru is a family of small language models built by Sumeru AI for
Indian education — NCERT curriculum, JEE, and NEET preparation across
Physics, Chemistry, Biology, and Mathematics.
"""
config_class = SumeruConfig
def __init__(self, config):
super().__init__(config)
self.post_init()