Downloads · 30 days
12
22% of all-time downloads
namhokaist/appgen-qwen2.5-vl-7b-sft-ngc-v1-unfrozen-vision-lr5e7-1ep
appgen-qwen2.5-vl-7b-sft-ngc-v1-unfrozen-vision-lr5e7-1ep is a image-text-to-text model from namhokaist. 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 transformers. The card lists the license as other.
This checkpoint is full-parameter supervised fine-tuning of Qwen/Qwen2.5-VL-7B-Instruct. The language model, vision encoder, and multimodal aligner were all unfrozen.
Downloads · 30 days
12
22% of all-time downloads
All-time downloads
55
Public
Parameters
8.3B
16.6 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors16.6 GB · 100%
From the Hugging Face model README
This checkpoint is full-parameter supervised fine-tuning of
Qwen/Qwen2.5-VL-7B-Instruct. The language model, vision encoder,
and multimodal aligner were all unfrozen.
luca0621/appgen-sft-ngc-v1769ea99dbc4ff190048ae0db37eb6310dba595e0ngc-f-qwen25-absoluteabsolute-pixelcc594898137f460bfe9f0759e9844b3ce807cfb522f4ad304b625269de3530faff544d6a9cfc75ec800eafc625cf9761e6799cf5The images are synthetic AppGen HTML-to-PNG screenshots, not real-user or real-device captures. The pinned revision contains no benchmark, held-out, or third-party evaluation images.
The Qwen2.5 and Qwen3 configs are paired by semantic multiset/IDs, not row order or byte-identical prompt text.
Qwen2.5 canonical navigation/completion-replay labels use the processor's smart-resized 1092x2408 frame, while its 400 direct-grounding labels use raw 1080x2400 pixels. These pinned source-specific labels were preserved.
Two pinned long_press targets retain undeclared time: 1; Qwen2.5 canonical
rows also retain the publisher's appgen_sft_qwen3 source tag.
Pinned hyperparameters:
{
"acc_strategy": "token",
"adam_beta1": 0.9,
"adam_beta2": 0.95,
"adam_epsilon": 1e-08,
"average_tokens_across_devices": true,
"bf16": true,
"data_seed": 42,
"dataloader_num_workers": 0,
"dataset_num_proc": 1,
"eval_strategy": "no",
"expected_full_updates": 113,
"fp16": false,
"freeze_aligner": false,
"freeze_llm": false,
"freeze_parameters": [],
"freeze_parameters_ratio": 0.0,
"freeze_vit": false,
"global_batch_size": 32,
"global_world_size": 4,
"gradient_accumulation_steps": 4,
"gradient_checkpointing": true,
"lazy_tokenize": true,
"learning_rate": 5e-07,
"logging_steps": 5,
"loss_scale": "default",
"lr_scheduler_type": "cosine",
"max_grad_norm": 1.0,
"max_length": 8192,
"max_pixels": 2629536,
"max_steps": -1,
"num_train_epochs": 1.0,
"optim": "adamw_torch",
"packing": false,
"padding_free": false,
"padding_side": "right",
"per_device_eval_batch_size": 1,
"per_device_train_batch_size": 2,
"remove_unused_columns": true,
"save_only_model": true,
"save_steps": 25,
"save_strategy": "steps",
"save_total_limit": 5,
"seed": 42,
"split_dataset_ratio": 0.0,
"template": "qwen2_5_vl",
"torch_dtype": "bfloat16",
"train_dataloader_shuffle": true,
"train_type": "full",
"truncation_strategy": "delete",
"use_liger_kernel": false,
"warmup_ratio": 0.05,
"weight_decay": 0.1
}
This model is intended for Android visual-agent research and evaluation, not for safety-critical autonomous deployment.