Downloads · 30 days
2
3% of all-time downloads
shahedm2001/qwen3-VL-2B-cataract-sft-20260906-sub10
qwen3-VL-2B-cataract-sft-20260906-sub10 is a machine learning model from shahedm2001. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
QLoRA adapter for Qwen/Qwen3-VL-2B-Instruct, fine-tuned on cataract-surgery video (clip-level descriptions + full-video narration + CoT MCQs).
Downloads · 30 days
2
3% of all-time downloads
All-time downloads
59
Public
Repo size
59.3 MB
Likes
0
Public
Click a slice to open those files.
.safetensors47.5 MB · 80%
From the Hugging Face model README
QLoRA adapter for Qwen/Qwen3-VL-2B-Instruct, fine-tuned on cataract-surgery video (clip-level descriptions + full-video narration + CoT MCQs).
Qwen/Qwen3-VL-2B-InstructSUBSET_RATIO=0.1) — 559 train / 689 val samplesfrom transformers import AutoProcessor, AutoModelForImageTextToText
from peft import PeftModel
base = "Qwen/Qwen3-VL-2B-Instruct"
adapter = "shahedm2001/qwen3-VL-2B-cataract-sft-20260906-sub10"
processor = AutoProcessor.from_pretrained(adapter) # tokenizer + processor files included
model = AutoModelForImageTextToText.from_pretrained(base, dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, adapter)
model = model.merge_and_unload() # optional: fuse to a standalone model
For video inference, pass up to 64 frames per clip (training frame count);
see NFRAMES below. The repo also includes non_lora_state_dict.bin —
fully-trained merger weights (the merger is full-trainable, outside LoRA).

output/logs/sft/config.txt)| Group | Values |
|---|---|
| Model | Qwen/Qwen3-VL-2B-Instruct, 4-bit QLoRA (bnb_4bit_compute_dtype=bf16) |
| LoRA | r=16, alpha=32, dropout=0.05, all LLM q/k/v/o + gate/up/down + vision linears; excludes lm_head, embed_tokens, merger, pos_embed |
| Frozen | vision tower + LLM frozen; merger fully trainable |
| Batch | per-device 2, grad-accum 8 → global 16; 2 epochs → 70 steps |
| LR | LLM 1e-4, vision 2e-6, merger 1e-5; weight decay 0.1, warmup 10, cosine |
| Video | NFRAMES=64, pixels 98304–196608, max_seq_length 32768 |
| Precision | bf16, tf32, flash-attn disabled (SDPA), gradient checkpointing, liger-kernel |
| Save/eval | save every 300 steps (keep 3); generation eval, 256 new tokens |
| Run | wall 4h44m; GPU peak 23.9 GB / 24.6 GB, 78 °C; loss 2.607 → 1.213 |
Full command line is in cmd.txt (same directory as this run's logs, not uploaded).