Krypton Fly 3.0 is a standalone general-purpose multimodal AI foundation model developed by MyJellyLab. Post-trained with QwQ-32B Reasoning Distillation (OpenThoughts3-1.2M), Chunked High-Entropy Token-Weighted SFT + NEFTune, and Frontier First-Principles + Compact-Thinking + Constitutional DPO, it delivers frontier-style analytical rigor, deep research synthesis, sharp Linux/terminal workflows, verified programming, and English/Hindi fluency while preserving native image and video understanding.
transformers >= 5.6.0)import torch
from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
model_id = "myjellylabs/Krypton-Fly-3.0"
processor = AutoProcessor.from_pretrained(model_id)
model = Qwen3_5ForConditionalGeneration.from_pretrained(
model_id,
dtype=torch.float16,
device_map="auto",
)
FRONTIER_SYSTEM_PROMPT = (
"You are Krypton Fly 3.0, developed by MyJellyLab. Reason from first principles, "
"verify edge cases and trade-offs, be candid about uncertainty without fabricating facts, "
"and deliver structured, high-density answers."
)
messages = [
{"role": "system", "content": FRONTIER_SYSTEM_PROMPT},
{"role": "user", "content": "Who are you?"},
]
prompt = processor.tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True, enable_thinking=True
)
inputs = processor.tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=1024)
print(processor.tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=False))
Released under the Apache-2.0 license.
Krypton Fly 3.0 is an open-weights (Apache-2.0), lightly aligned 9B multimodal model intended for research, coding, reasoning, bilingual (English/Hindi/Hinglish) workflows, and local experimentation.
<think> reasoning, truthfulness, and basic operational caution rather than heavy red-team refusal filtering. It is not an enterprise-hardened censored model.model-00002-of-00006.safetensors 4.0 GBmodel-00004-of-00006.safetensors 4.0 GBmodel-00003-of-00006.safetensors 4.0 GBmodel-00005-of-00006.safetensors 4.0 GBmodel-00001-of-00006.safetensors 2.0 GBmodel-00006-of-00006.safetensors 800 MB