AstroBridge Captioner

n-modality (image + spectra) astronomy captioner. LoRA adapter + fusion stack trained on top of a frozen Qwen/Qwen3.5-9B. The base model itself is NOT included here โ€” load it fresh from Qwen/Qwen3.5-9B and apply this adapter on top.

How to load

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen3.5-9B", dtype=torch.bfloat16, trust_remote_code=True
)
llm = PeftModel.from_pretrained(base, "UniverseTBD/astrobridge-model-v7")
tokenizer = AutoTokenizer.from_pretrained("UniverseTBD/astrobridge-model-v7")

# middle.pt (fusion stack: projectors/modality_identity/qformer/adapter) needs the captioner
# package's FusionStack class to reload โ€” see captioner/model/captioner.py and
# captioner/train/stage1.py's run_stage1 for how it's constructed and wired to the LLM.

Training info

  • config_hash: cde984a5c692e8cd
  • quantization: None
  • git_sha: a5fa267401ffe7e85cb2c0870f8bc3a642acc18e
  • tier_histogram: {"single": 4120, "joint": 1318}

Eval (groundedness gate)

{
  "per_modality": {
    "image": {
      "shuffle_test": {
        "modality": "image",
        "n": 58,
        "mean_edit_distance": 240.05172413793105,
        "null_result": false
      },
      "ablation_test": {
        "modality": "image",
        "n": 109,
        "fraction_caption_changed": 1.0,
        "null_result": false
      },
      "diversity_test": {
        "modality": "image",
        "n": 120,
        "distinct_fraction": 0.9916666666666667,
        "top1_share": 0.016666666666666666,
        "most_common_caption": "This object presents a smooth, elongated oval distribution of light oriented diagonally from the lower left toward the upper right. It features a bright, warm yellowish-white central core that fades c",
        "null_result": false
      }
    },
    "spectra": {
      "shuffle_test": {
        "modality": "spectra",
        "n": 13,
        "mean_edit_distance": 196.76923076923077,
        "null_result": false
      },
      "ablation_test": {
        "modality": "spectra",
        "n": 43,
        "fraction_caption_changed": 1.0,
        "null_result": false
      },
      "diversity_test": {
        "modality": "spectra",
        "n": 114,
        "distinct_fraction": 1.0,
        "top1_share": 0.008771929824561403,
        "most_common_caption": "This optical spectrum of a galaxy in the very close local universe displays a stellar continuum marked by absorption features, including Ca II H & K, the G-band, Mg I b, and Na I D. The spectrum is do",
        "null_result": false
      }
    },
    "lightcurve": {
      "shuffle_test": {
        "modality": "lightcurve",
        "n": 9,
        "mean_edit_distance": 52.666666666666664,
        "null_result": false
      },
      "ablation_test": {
        "modality": "lightcurve",
        "n": 48,
        "fraction_caption_changed": 1.0,
        "null_result": false
      },
      "diversity_test": {
        "modality": "lightcurve",
        "n": 94,
        "distinct_fraction": 1.0,
        "top1_share": 0.010638297872340425,
        "most_common_caption": "The optical light curve spans approximately 104 days in the $g$ and $r$ bands, capturing a well-sampled rise to peak brightness near day 13.0 in $g$ ($17.09 \\pm 0.04$ mag) and day 16.",
        "null_result": false
      }
    }
  }
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for UniverseTBD/astrobridge-model-v7

Finetuned
Qwen/Qwen3.5-9B
Adapter
(668)
this model