vytre_core

Vytre Core

Vytre Core is a compact, domain-specific text-generation model trained on synthetic enterprise-workforce tasks: department creation, agent definition, workflow planning, task decomposition, governance checks, and tool routing.

Model files

This repository is self-contained. It contains a standard Transformers GPT-2 checkpoint and tokenizer, not a LoRA adapter. Do not combine it with the legacy vytre-core-upload LoRA template or an external Llama base model.

Load with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "tarvico/vytre-core"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = (
    "You are Vytre, an enterprise workforce operating intelligence model.\\n"
    "Input: Create marketing department\\n"
    "Output: "
)
inputs = tokenizer(prompt, return_tensors="pt")
tokens = model.generate(
    **inputs,
    max_new_tokens=80,
    do_sample=False,
    pad_token_id=tokenizer.pad_token_id,
    eos_token_id=tokenizer.eos_token_id,
)
print(tokenizer.decode(tokens[0], skip_special_tokens=True))

Limitations

This is a small specialised model, not a general-purpose chat model. Use the prompt format above and keep requests close to the listed operational domains. Validate generated JSON before taking actions from it.