It is packaged as a standalone saved model folder for inference: the repository includes model weights, tokenizer files, config files, and the required custom runtime files.

Identity

Name: Sky v1.3
Organization: 0labs
Creator: Atharvsinh Jadav
Intended use: coding, debugging, reasoning, general chat, and assistant workflows

Quick Use

Because the repository name contains a dot, Colab users should download it into a safe local folder name before loading.

python

import torch
from huggingface_hub import snapshot_download
from transformers import AutoTokenizer, AutoModelForCausalLM

repo_id = "0labs-in/Sky-V1_3-5.5B"
local_dir = "/content/sky_v1_3_5_5b"

snapshot_download(repo_id=repo_id, local_dir=local_dir, repo_type="model")

tokenizer = AutoTokenizer.from_pretrained(local_dir, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    local_dir,
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else torch.float16,
    trust_remote_code=True,
    _attn_implementation="sdpa",
    device_map="auto",
)

messages = [{"role": "user", "content": "who are you?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
inputs["input_mode"] = torch.tensor([0], device=model.device)

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=160,
        do_sample=False,
        pad_token_id=tokenizer.eos_token_id,
        eos_token_id=tokenizer.eos_token_id,
    )

print(tokenizer.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Smoke Test Summary

Production smoke test passed with 0 failures across:

normal chat
reasoning
math
coding
instruction following
identity guard
safety refusal behavior

Notes

Use trust_remote_code=True because this repository includes custom model runtime files.