Greta-1b-Preview

Model Details

Current state

This model is currently still in post-training and is pre-released as a preview checkpoint with gated access only.

How to run

Notice: See jinja-template for formats and tool parsing

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("garboab/Greta-1b-Preview", use_fast=True, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    "garboab/Greta-1b-Preview",
    trust_remote_code=True,
    device_map="auto",
)

tools = [
    {
        "type": "function",
        "name": "search",
        "description": "Search the web for information.",
        "parameters": {
            "type": "object",
            "properties": {
                "q": {
                    "type": "string",
                    "description": "Query to search for.",
                },
            },
            "required": ["q"],
        },
    },
]

text = tokenizer.apply_chat_template(
    [
        {"role": "tools", "content": tools},
        {"role": "user", "content": "Where can I read more about Garbo AI?"},
    ],
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True,
)

input_ids = tokenizer(
    text,
    add_bos_token=True,
    return_tensors="pt"
).input_ids

output_ids = model.generate(
    input_ids.cuda(),
    max_new_tokens=1024,
    do_sample=True,
    temperature=0.1
)

print(tokenizer.decode(output_ids[0], skip_special_tokens=True).strip())