sarman/dgx-spark-playbooks
1
0
Fork 0
mirror of https://github.com/NVIDIA/dgx-spark-playbooks.git synced 2026-04-22 18:13:52 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge 20b788feaa into a00143ae15

Browse Source
This commit is contained in:
Robert McDermott 2025-11-10 07:56:25 -08:00 committed by GitHub
commit 070a21bca5
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
nvidia/unsloth/README.md
View File

@ -79,13 +79,16 @@ docker pull nvcr.io/nvidia/pytorch:25.09-py3
## Step 3. Launch Docker
```bash
docker run --gpus all --ulimit memlock=-1 -it --ulimit stack=67108864 --entrypoint /usr/bin/bash --rm nvcr.io/nvidia/pytorch:25.09-py3
mkdir -p ./outputs # directory to save the unsloth finetuned model results
docker run --gpus all \
--ulimit memlock=-1 --ulimit stack=67108864 -it --entrypoint /usr/bin/bash --rm \
-v "$(pwd)/outputs:/workspace/outputs" nvcr.io/nvidia/pytorch:25.09-py3
```
## Step 4. Install dependencies inside Docker
```bash
pip install transformers peft datasets "trl==0.19.1"
pip install transformers peft datasets 'trl==0.23.1' hf_transfer
pip install --no-deps unsloth unsloth_zoo
pip install hf_transfer
```
@ -97,7 +100,7 @@ pip install --no-deps bitsandbytes
## Step 6. Create Python test script
Curl the test script [here](https://github.com/NVIDIA/dgx-spark-playbooks/blob/main/nvidia/unsloth/assets/test_unsloth.py) into the container.
Curl the test script [here](https://raw.githubusercontent.com/NVIDIA/dgx-spark-playbooks/refs/heads/main/nvidia/unsloth/assets/test_unsloth.py) into the container.
```bash
curl -O https://raw.githubusercontent.com/NVIDIA/dgx-spark-playbooks/refs/heads/main/nvidia/unsloth/assets/test_unsloth.py
@ -124,17 +127,19 @@ Expected output in the terminal window:
Test with your own model and dataset by updating the `test_unsloth.py` file:
```python
## Replace line 32 with your model choice
## Replace line 49 with your model choice
model_name = "unsloth/Meta-Llama-3.1-8B-bnb-4bit"
## Load your custom dataset in line 8
## Load your custom dataset in line 25
dataset = load_dataset("your_dataset_name")
## Adjust training parameter args at line 61
## Adjust training parameter args at line 80, and max_steps at line 83
per_device_train_batch_size = 4
max_steps = 1000
```
When you exit the container, the fine-tuned unsloth ouputs will be in the "ouputs" directory.
Visit https://github.com/unslothai/unsloth/wiki
for advanced usage instructions, including:
- [Saving models in GGUF format for vLLM](https://github.com/unslothai/unsloth/wiki#saving-to-gguf)