Merge pull request #2942 from madeline-underwood/nvda

Nvda_reviewed

Author: jasonrandrews

content/learning-paths/laptops-and-desktops/pytorch-finetuning-on-spark/1-setup.md

@@ -6,6 +6,8 @@ weight: 2
 layout: learningpathall
 ---
 
+## Overview
+
 The NVIDIA DGX Spark pairs an Arm-based Grace CPU with a Blackwell GPU in a compact desktop form factor. The GPU handles the compute-intensive training passes while the Grace CPU manages data preprocessing and orchestration, making the system well suited for fine-tuning large language models locally without sending data to the cloud.
 
 To get started, you'll configure Docker, pull a pre-built PyTorch container, and install the libraries you need for fine-tuning.
@@ -39,7 +41,7 @@ Pull the latest PyTorch container from NVIDIA's container registry:
 docker pull nvcr.io/nvidia/pytorch:25.11-py3
 ```
 
-This command downloads the November 2025 release of the PyTorch container, which includes PyTorch, CUDA libraries, cuDNN, and other essential tools pre-configured for optimal performance on NVIDIA hardware. The download size is several gigabytes, so this step might take a few minutes depending on your internet connection.
+This command downloads the November 2025 release of the PyTorch container, which includes PyTorch, CUDA libraries, cuDNN, and other essential tools pre-configured for optimal performance on NVIDIA hardware. The download size is several gigabytes, so this step can take a few minutes depending on your internet connection.
 
 ## Launch container instance
 
@@ -81,7 +83,7 @@ These packages serve specific purposes:
 - `trl` (Transformer Reinforcement Learning) includes training utilities and recipes for language models
 - `bitsandbytes` enables 4-bit and 8-bit quantization for memory-efficient training
 
-The installation typically takes a few minutes as pip downloads and installs each package along with their dependencies.
+The installation can take a few minutes as pip downloads and installs each package along with their dependencies.
 
 ## Authenticate with Hugging Face
 

content/learning-paths/laptops-and-desktops/pytorch-finetuning-on-spark/2-finetuning.md

@@ -36,7 +36,7 @@ The NVIDIA playbook scripts use the Alpaca prompt format, which structures each
 
 **Input** -- optional additional context (left empty for most questions)
 
-**Output** -- the correct answer sourced from official datasheets
+**output** -- the correct answer sourced from official datasheets
 
 Here's an example from the dataset you'll use:
 

content/learning-paths/laptops-and-desktops/pytorch-finetuning-on-spark/3-pytorch.md

@@ -6,6 +6,8 @@ weight: 4
 layout: learningpathall
 ---
 
+## Overview
+
 Now that you understand how fine-tuning works, it's time to look at the actual code. In this section, you'll walk through the key parts of the `Llama3_3B_full_finetuning.py` script and run it with the Raspberry Pi dataset to produce your own fine-tuned Llama model.
 
 
@@ -23,7 +25,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 
 The `DATASET_PROMPT_TEMPLATE` defines the instruction-following format for training data with three fields: instruction, input, and response. Each training example is formatted using this template so the model learns to recognize the pattern and produce structured answers.
 
-The `get_dataset()` function loads the dataset from (Hugging Face by default) and formats each example using the template, appending the EOS (End of String) token. You'll pass in the Raspberry Pi dataset from a local JSONL file instead.
+The `get_dataset()` function loads the dataset from Hugging Face by default and formats each example using the template, appending the EOS (End of Sequence) token. You'll pass in the Raspberry Pi dataset from a local JSONL file instead.
 
 ```python
 # Define prompt templates
@@ -154,13 +156,13 @@ cp /workspace/raspberry_pi_qa.jsonl .
 
 ## Run the fine-tuning
 
-With the dataset patch applied, you're ready to run the fine-tuning. The command below trains the Llama 3.1 8B model using full fine-tuning on the Raspberry Pi dataset:
+With the dataset in place, you're ready to run the fine-tuning. The command below trains the Llama 3.2 3B model using full fine-tuning on the Raspberry Pi dataset:
 
 ```bash
 python Llama3_3B_full_finetuning.py \
 --model_name "meta-llama/Llama-3.2-3B-Instruct" \
 --dataset "json" \
---dataset_files="/workspace/projects/pytorch-finetuning/raspberry_pi_qa.jsonl" \
+--dataset_files="raspberry_pi_qa.jsonl" \
 --dataset_size 300 \
 --output_dir "/workspace/models/Llama-3.2-3B-FineTuned"
 ```

content/learning-paths/laptops-and-desktops/pytorch-finetuning-on-spark/4-testing.md

@@ -6,6 +6,8 @@ weight: 5
 layout: learningpathall
 ---
 
+## Overview
+
 Now that you've fine-tuned your model on Raspberry Pi datasheet content, it's time to compare its behavior against the original. You'll serve both versions using vLLM, a high-performance inference server optimized for large language models, and observe how fine-tuning on domain-specific data changes the model's factual accuracy.
 
 ## Download vLLM container
@@ -64,8 +66,7 @@ Wait for the server to fully load the model and display the message indicating i
 
 ### Test prompt
 
-
-From a new terminal window (outside the container), send a Raspberry Pi hardware question to the model using the Alpaca instruction format. For this example, use a question about the memory size:
+From a new terminal window (outside the container), use `curl` to send an HTTP request to the model server. The request contains a Raspberry Pi hardware question formatted using the Alpaca instruction template. For this example, ask about the RP2350 memory size:
 
 ```bash
 curl http://localhost:8000/v1/completions \
@@ -109,7 +110,7 @@ The base model confidently reports the RP2350 has "256MB of memory," which is of
 
 Now test your fine-tuned model to see how training on Raspberry Pi datasheet content improved its factual accuracy. Stop the current vLLM server (press Ctrl+C in the container terminal) before launching the fine-tuned model.
 
-{{% notice Dependency Conflict %}}
+{{% notice Note %}}
 As of this writing, vLLM does not support version 5 of the `transformers` library that was used when fine-tuning the model, so you need to patch its `tokenizer_config.json`. Run the following command to update the `tokenizer_class` to `PreTrainedTokenizerFast`, which is compatible with the older `transformers` version bundled in the vLLM container:
 
 ```bash
@@ -132,7 +133,7 @@ The only change from the previous command is the `--model` parameter, which now
 
 ### Test prompt
 
-Send the same Raspberry Pi question to your fine-tuned model:
+Send the same Raspberry Pi question to your fine-tuned model using the same `curl` command format:
 
 ```bash
 curl http://localhost:8000/v1/completions \

content/learning-paths/laptops-and-desktops/pytorch-finetuning-on-spark/_index.md

@@ -1,18 +1,14 @@
 ---
 title: Fine-tune PyTorch models on DGX Spark
 
-draft: true
-cascade:
-    draft: true
-
 minutes_to_complete: 60
 
 who_is_this_for: This is an advanced topic for AI developers and ML engineers who want to fine-tune large language models using PyTorch and Hugging Face on the NVIDIA DGX Spark platform.
 
 learning_objectives: 
     - Understand how fine-tuning teaches a model domain-specific knowledge
     - Prepare a custom JSONL dataset for supervised fine-tuning
-    - Fine-tune Llama 3.1 8B on Raspberry Pi datasheet content using PyTorch and Hugging Face
+    - Fine-tune Llama 3.2 3B on Raspberry Pi datasheet content using PyTorch and Hugging Face
     - Compare base and fine-tuned model responses to verify factual accuracy improvements
 
 prerequisites:
@@ -55,6 +51,10 @@ further_reading:
         title: PyTorch Training Documentation
         link: https://pytorch.org/tutorials/beginner/introyt/trainingyt.html
         type: documentation
+    - resource:
+        title: Build a serverless LLM inference application with AWS Lambda and Arm processors
+        link: /learning-paths/servers-and-cloud-computing/llama-cpu/
+        type: website
 
 ### FIXED, DO NOT MODIFY
 # ================================================================================