S
pending
Want to Understand LLMs? Build One from Scratch.
Grounded / Real
Inflated / Uruttu
Original Content
This is literally one of the best LLM playlists out there!
43 lectures from Vizuara where you build an entire LLM from scratch, then finetune it.
You're writing the tokenizer yourself, coding attention from scratch, building the transformer block piece by piece.
Here's everything it covers:
Foundations
→ LLM basics, pretraining vs finetuning, what transformers are
→ How GPT-3 really works + the 3 stages of building an LLM
Data Pipeline
→ Coding a tokenizer from scratch
→ Byte Pair Encoding (the actual GPT tokenizer)
→ Input-target pairs, token embeddings, positional embeddings
→ The complete data preprocessing pipeline
Attention Mechanism (the deep part)
→ Simplified attention → self-attention with key/query/value
→ Causal self-attention, coded from scratch
→ Multi-head attention, both the intuition and the full math
Building the Architecture
→ Birds-eye view of the LLM architecture
→ Layer normalization, GELU activation, shortcut connections
→ Coding the entire transformer block
→ Building the actual 124M parameter GPT-2 model
Training
→ Defining the loss function + evaluating performance
→ Coding the full pretraining loop
→ Decoding strategies: temperature scaling, top-k sampling
→ Saving/loading weights, loading pretrained GPT-2 weights
Finetuning
→ Classification finetuning (spam classifier, built from scratch)
→ Instruction finetuning with Alpaca-format prompts
→ Data batching, dataloaders, the full finetuning training loop
→ Evaluating the finetuned model with Ollama
most complete free resource on LLMs I've found.
Heads up: quite a few of the lectures were recently moved to Members only, so you may need a YouTube membership to unlock the full series. Worth every penny.
43 lectures from Vizuara where you build an entire LLM from scratch, then finetune it.
You're writing the tokenizer yourself, coding attention from scratch, building the transformer block piece by piece.
Here's everything it covers:
Foundations
→ LLM basics, pretraining vs finetuning, what transformers are
→ How GPT-3 really works + the 3 stages of building an LLM
Data Pipeline
→ Coding a tokenizer from scratch
→ Byte Pair Encoding (the actual GPT tokenizer)
→ Input-target pairs, token embeddings, positional embeddings
→ The complete data preprocessing pipeline
Attention Mechanism (the deep part)
→ Simplified attention → self-attention with key/query/value
→ Causal self-attention, coded from scratch
→ Multi-head attention, both the intuition and the full math
Building the Architecture
→ Birds-eye view of the LLM architecture
→ Layer normalization, GELU activation, shortcut connections
→ Coding the entire transformer block
→ Building the actual 124M parameter GPT-2 model
Training
→ Defining the loss function + evaluating performance
→ Coding the full pretraining loop
→ Decoding strategies: temperature scaling, top-k sampling
→ Saving/loading weights, loading pretrained GPT-2 weights
Finetuning
→ Classification finetuning (spam classifier, built from scratch)
→ Instruction finetuning with Alpaca-format prompts
→ Data batching, dataloaders, the full finetuning training loop
→ Evaluating the finetuned model with Ollama
most complete free resource on LLMs I've found.
Heads up: quite a few of the lectures were recently moved to Members only, so you may need a YouTube membership to unlock the full series. Worth every penny.
Validated Content
Overall: Very accurate, with only a couple of subjective or outdated claims.
| Claim | Verdict | Notes |
|---|---|---|
| "43 lectures from Vizuara" | ✅ True | The playlist currently contains 43 lectures (about 31 hours). (firstprinciplelabs.ai) |
| "Build an entire LLM from scratch, then finetune it" | ✅ True | The curriculum walks through building a GPT-style model, pretraining, and fine-tuning. (firstprinciplelabs.ai) |
| "Coding the tokenizer yourself" | ✅ True | The playlist includes implementing a tokenizer and Byte Pair Encoding (BPE). (LinkedIn) |
| "Coding attention from scratch" | ✅ True | Self-attention, causal attention, and multi-head attention are implemented from scratch. (LinkedIn) |
| "Building the transformer block piece by piece" | ✅ True | LayerNorm, GELU, residual connections, transformer blocks, and GPT-2 are covered individually. (LinkedIn) |
| "Building the 124M parameter GPT-2 model" | ✅ True | The course builds the GPT-2 Small (124M parameters) architecture. (LinkedIn) |
| "Classification finetuning and instruction finetuning" | ✅ True | Spam classification and instruction tuning are part of the curriculum. (LinkedIn) |
| "Evaluating the finetuned model with Ollama" | ✅ True | The newer expanded curriculum includes evaluation with Ollama. (LinkedIn) |
| "Most complete free resource on LLMs I've found" | ⚠️ Opinion | This is a subjective opinion, not a verifiable fact. |
| "Quite a few lectures were moved to Members only" | ✅ Likely True | Recent community posts indicate that some videos now require YouTube membership. (TwStalker) |
| "Worth every penny" | ⚠️ Opinion | Personal recommendation rather than a factual statement. |
Small improvements
Instead of:
This is literally one of the best LLM playlists out there!
Use:
This is one of the most comprehensive LLM-from-scratch playlists I've come across.
Instead of:
most complete free resource on LLMs I've found.
Use:
One of the most comprehensive LLM learning resources I've found.
This avoids the contradiction that some videos are now members-only.
Real vs Fluff
🟢 Real: 94%
🟡 Fluff: 6%
The only fluff is the subjective praise ("one of the best", "worth every penny", "most complete"). Nearly everything else accurately reflects the playlist's curriculum and structure.