Downloads · 30 days
5
9% of all-time downloads
qwen25vl-ahc-lora-ckpt400
hackiit-neeraj/qwen25vl-ahc-lora-ckpt400
qwen25vl-ahc-lora-ckpt400 is a image-text-to-text model from hackiit-neeraj. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
LoRA adapter on Qwen2.5-VL-3B-Instruct for describing / labelling context-dependent events in urban drone video (stopped vehicles, smoke, flood, spill, congestion, etc.).
All-time downloads
57
Public
Repo size
251 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors240 MB · 91%
At a glance
- Task
- Image-Text-to-Text
- Library
- peft
- License
- apache-2.0
- Access
- Public
- Created
- Sep 5, 2026
- Updated
- Sep 5, 2026
- SHA
- 3e93d14e
Try a prompt
Base models
Model card
From the Hugging Face model README
Qwen2.5-VL-3B LoRA — aerial drone anomaly captions
LoRA adapter on Qwen2.5-VL-3B-Instruct for describing / labelling context-dependent events in urban drone video (stopped vehicles, smoke, flood, spill, congestion, etc.).
Repo root is the batch-2 checkpoint-400 adapter (recovered
road_spill_or_debris, multi-frame macro-F1 0.437). Adapter-only;
~125 MB.
A later fine-tune lives in batch4/ (rich interval set, seed 9011).
It is not a wholesale upgrade of these root weights — see that folder's
card.
Load
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from peft import PeftModel
base_id = "Qwen/Qwen2.5-VL-3B-Instruct"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
base_id, torch_dtype="auto", device_map="auto"
)
model = PeftModel.from_pretrained(model, "hackiit-neeraj/qwen25vl-ahc-lora-ckpt400")
processor = AutoProcessor.from_pretrained(base_id)
Trained against unsloth/qwen2.5-vl-3b-instruct-unsloth-bnb-4bit. The
official Qwen/Qwen2.5-VL-3B-Instruct weights are the same architecture
and load this adapter.
Training
| Method | Unsloth QLoRA, 4-bit base, PEFT LoRA |
| Rank / alpha / dropout | r=16, alpha=16, dropout=0 |
| Targets | language-model attention + MLP projections |
| Hardware | Kaggle Tesla T4 |
| This checkpoint | warm-start from checkpoint-500, seed 1234, lr 3e-5, 400 steps, 0 non-finite losses |
| Eval | AHC public test, adaptive multi-frame, min_frames=2 → macro-F1 0.437 |
Do not load optimizer.pt / trainer state — those were stripped on purpose.
Intended use
Offline / gated Stage-3 labelling of drone frames. A still frame cannot see motion, so this adapter should describe what is visible (and name an appearance class). Tracker / motion rules should own stop / stall / wrong-way decisions.
License
Adapter is Apache-2.0, same as the Qwen2.5-VL base. You need the base model weights separately.
- Task
- Image-Text-to-Text
- Library
- peft
- License
- apache-2.0
- Created
- Sep 5, 2026
- Updated
- Sep 5, 2026