GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Wynnoaasf
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Wynnoaasf
Model Tree
unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit
Input Modalities
Output Modalities
Supported Functionality
| Đặc điểm | Giá trị |
|---|---|
| Model nền | Qwen2.5-1.5B-Instruct (4-bit) |
| Kích thước | ~1.5B tham số, adapter LoRA ~20–40 MB |
| Ngôn ngữ | Tiếng Việt (chính) + tiếng Anh |
LoRA rank (r) | 16 |
| Target modules | q,k,v,o,gate,up,down_proj |
| Context length | 512 tokens |
| VRAM tối thiểu | ~8 GB (4-bit) — chạy tốt trên T4 |
| Framework | Unsloth + TRL + PEFT + Transformers |
| Tham số | Giá trị |
|---|---|
max_steps | 150 |
learning_rate | 2e-4 |
per_device_train_batch_size | 2 |
gradient_accumulation_steps | 4 |
| Batch hiệu dụng | 8 |
warmup_steps | 5 |
optimizer | AdamW 8-bit (Unsloth mặc định) |
fp16 / bf16 | Auto (theo GPU) |
bash
pip install -q "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"pip install -q --upgrade "peft>=0.21.0"pip install -q --upgrade --no-deps "bitsandbytes>=0.46.1"pip install -q "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"pip install -q --upgrade "peft>=0.21.0"pip install -q --upgrade --no-deps "bitsandbytes>=0.46.1"