Run log · LoRA smoke test ·
Qwen3.5-4B, sequence 2048
2026-11-22_smoke-train_Qwen3.5-4B_seq2048-short
A smoke test checks that training runs on this card, how fast, and with how much memory. It uses eight placeholder examples. The adapter is thrown away, and the falling loss only shows the pipeline can memorise them.
Recorded fields
- Model
- unsloth/Qwen3.5-4B
- Loaded in 4-bit
- no, 16-bit
- Layers offloaded to system RAM
- no
- Training steps
- 30
- Sequence length
- 2048 tokens
- LoRA rank
- 16
- Training time
- 49 s
- Seconds per step, averaged over every step including the first compile step
- 1.64 s
- VRAM after load
- 8.47 GiB
- Peak VRAM
- 8.81 GiB
- First logged loss
- 2.245
- Last logged loss
- 0.376
- Adapter size on disk
- 100.1 MB
- PyTorch
- 2.14.1+cu130
- Hardware
- RTX 4070 Ti, 12 GB
Logged loss
| Step | Loss |
|---|---|
| 10 | 2.245 |
| 20 | 0.9985 |
| 30 | 0.3758 |