LLM-PowerHouse: A Curated Guide for Large Language Models with Custom Training and Inferencing

Welcome to LLM-PowerHouse, your ultimate resource for unleashing the full potential of Large Language Models (LLMs) with custom training and inferencing. This GitHub repository is a comprehensive and curated guide designed to empower developers, researchers, and enthusiasts to harness the true capabilities of LLMs and build intelligent applications that push the boundaries of natural language understanding.

🧑‍🔬 LLM Scientist
In-Depth Articles
Codebase Mastery: Building with Perfection
Codebase Mastery: Building with Perfection
LLM PlayLab
What I am learning
Contributing
License

🧑‍🔬 LLM Scientist

In this segment of the curriculum, participants delve into mastering the creation of top-notch LLMs through cutting-edge methodologies.

Toggle section

graph LR
    Scientist["LLM Scientist 👩‍🔬"] --> Architecture["The LLM architecture 🏗️"]
    Scientist["LLM Scientist 👩‍🔬"] --> Instruction["Building an instruction dataset 📚"]
    Scientist["LLM Scientist 👩‍🔬"] --> Pretraining["Pretraining models 🛠️"]
    Scientist["LLM Scientist 👩‍🔬"] --> FineTuning["Supervised Fine-Tuning 🎯"]
    Scientist["LLM Scientist 👩‍🔬"] --> RLHF["RLHF 🔍"]
    Scientist["LLM Scientist 👩‍🔬"] --> Evaluation["Evaluation 📊"]
    Scientist["LLM Scientist 👩‍🔬"] --> Quantization["Quantization ⚖️"]
    Scientist["LLM Scientist 👩‍🔬"] --> Trends["New Trends 📈"]
    Architecture["The LLM architecture 🏗️"] --> HLV["High Level View 🔍"]
    Architecture["The LLM architecture 🏗️"] --> Tokenization["Tokenization 🔠"]
    Architecture["The LLM architecture 🏗️"] --> Attention["Attention Mechanisms 🧠"]
    Architecture["The LLM architecture 🏗️"] --> Generation["Text Generation ✍️"]
    Instruction["Building an instruction dataset 📚"] --> Alpaca["Alpaca-like dataset 🦙"]
    Instruction["Building an instruction dataset 📚"] --> Advanced["Advanced Techniques 📈"]
    Instruction["Building an instruction dataset 📚"] --> Filtering["Filtering Data 🔍"]
    Instruction["Building an instruction dataset 📚"] --> Prompt["Prompt Templates 📝"]
    Pretraining["Pretraining models 🛠️"] --> Pipeline["Data Pipeline 🚀"]
    Pretraining["Pretraining models 🛠️"] --> CLM["Casual Language Modeling 📝"]
    Pretraining["Pretraining models 🛠️"] --> Scaling["Scaling Laws 📏"]
    Pretraining["Pretraining models 🛠️"] --> HPC["High-Performance Computing 💻"]
    FineTuning["Supervised Fine-Tuning 🎯"] --> Full["Full fine-tuning 🛠️"]
    FineTuning["Supervised Fine-Tuning 🎯"] --> Lora["Lora and QLoRA 🌀"]
    FineTuning["Supervised Fine-Tuning 🎯"] --> Axoloti["Axoloti 🦠"]
    FineTuning["Supervised Fine-Tuning 🎯"] --> DeepSpeed["DeepSpeed ⚡"]
    RLHF["RLHF 🔍"] --> Preference["Preference Datasets 📝"]
    RLHF["RLHF 🔍"] --> Optimization["Proximal Policy Optimization 🎯"]
    RLHF["RLHF 🔍"] --> DPO["Direct Preference Optimization 📈"]
    Evaluation["Evaluation 📊"] --> Traditional["Traditional Metrics 📏"]
    Evaluation["Evaluation 📊"] --> General["General Benchmarks 📈"]
    Evaluation["Evaluation 📊"] --> Task["Task-specific Benchmarks 📋"]
    Evaluation["Evaluation 📊"] --> HF["Human Evaluation 👩‍🔬"]
    Quantization["Quantization ⚖️"] --> Base["Base Techniques 🛠️"]
    Quantization["Quantization ⚖️"] --> GGUF["GGUF and llama.cpp 🐐"]
    Quantization["Quantization ⚖️"] --> GPTQ["GPTQ and EXL2 🤖"]
    Quantization["Quantization ⚖️"] --> AWQ["AWQ 🚀"]
    Trends["New Trends 📈"] --> Positional["Positional Embeddings 🎯"]
    Trends["New Trends 📈"] --> Merging["Model Merging 🔄"]
    Trends["New Trends 📈"] --> MOE["Mixture of Experts 🎭"]
    Trends["New Trends 📈"] --> Multimodal["Multimodal Models 📷"]

1. The LLM architecture 🏗️

An overview of the Transformer architecture, with emphasis on inputs (tokens) and outputs (logits), and the importance of understanding the vanilla attention mechanism and its improved versions.

concept	Description
Transformer Architecture (High-Level)	Review encoder-decoder Transformers, specifically the decoder-only GPT architecture used in modern LLMs.
Tokenization	Understand how raw text is converted into tokens (words or subwords) for the model to process.
Attention Mechanisms	Grasp the theory behind attention, including self-attention and scaled dot-product attention, which allows the model to focus on relevant parts of the input during output generation.
Text Generation	Learn different methods the model uses to generate output sequences. Common strategies include greedy decoding, beam search, top-k sampling, and nucleus sampling.

Further Exploration

Reference	Description	Link
The Illustrated Transformer by Jay Alammar	A visual and intuitive explanation of the Transformer model	🔗
The Illustrated GPT-2 by Jay Alammar	Focuses on the GPT architecture, similar to Llama's.	🔗
Visual intro to Transformers by 3Blue1Brown	Simple visual intro to Transformers	🔗
LLM Visualization by Brendan Bycroft	3D visualization of LLM internals	🔗
nanoGPT by Andrej Karpathy	Reimplementation of GPT from scratch (for programmers)	🔗
Decoding Strategies in LLMs	Provides code and visuals for decoding strategies	🔗

2. Building an instruction dataset 📚

While it's easy to find raw data from Wikipedia and other websites, it's difficult to collect pairs of instructions and answers in the wild. Like in traditional machine learning, the quality of the dataset will directly influence the quality of the model, which is why it might be the most important component in the fine-tuning process.

Concept	Description
Alpaca-like dataset	This dataset generation method utilizes the OpenAI API (GPT) to synthesize data from scratch, allowing for the specification of seeds and system prompts to foster diversity within the dataset.
Advanced techniques	Delve into methods for enhancing existing datasets with Evol-Instruct, and explore approaches for generating top-tier synthetic data akin to those outlined in the Orca and phi-1 research papers.
Filtering data	Employ traditional techniques such as regex, near-duplicate removal, and prioritizing answers with substantial token counts to refine datasets.
Prompt templates	Recognize the absence of a definitive standard for structuring instructions and responses, underscoring the importance of familiarity with various chat templates like ChatML and Alpaca.

Further Exploration

Reference	Description	Link
Preparing a Dataset for Instruction tuning by Thomas Capelle	Explores the Alpaca and Alpaca-GPT4 datasets and discusses formatting methods.	🔗
Generating a Clinical Instruction Dataset by Solano Todeschini	Provides a tutorial on creating a synthetic instruction dataset using GPT-4.	🔗
GPT 3.5 for news classification by Kshitiz Sahay	Demonstrates using GPT 3.5 to create an instruction dataset for fine-tuning Llama 2 in news classification.	🔗
Dataset creation for fine-tuning LLM	Notebook containing techniques to filter a dataset and upload the result.	🔗
Chat Template by Matthew Carrigan	Hugging Face's page about prompt templates	🔗

3. Pretraining models 🛠️

Pre-training, being both lengthy and expensive, is not the primary focus of this course. While it's beneficial to grasp the fundamentals of pre-training, practical experience in this area is not mandatory.

Concept	Description
Data pipeline	Pre-training involves handling vast datasets, such as the 2 trillion tokens used in Llama 2, which necessitates tasks like filtering, tokenization, and vocabulary preparation.
Causal language modeling	Understand the distinction between causal and masked language modeling, including insights into the corresponding loss functions. Explore efficient pre-training techniques through resources like Megatron-LM or gpt-neox.
Scaling laws	Delve into the scaling laws, which elucidate the anticipated model performance based on factors like model size, dataset size, and computational resources utilized during training.
High-Performance Computing	While beyond the scope of this discussion, a deeper understanding of HPC becomes essential for those considering building their own LLMs from scratch, encompassing aspects like hardware selection and distributed workload management.

Further Exploration

Reference	Description	Link
LLMDataHub by Junhao Zhao	Offers a carefully curated collection of datasets tailored for pre-training, fine-tuning, and RLHF.	🔗
Training a causal language model from scratch by Hugging Face	Guides users through the process of pre-training a GPT-2 model from the ground up using the transformers library.	🔗
TinyLlama by Zhang et al.	Provides insights into the training process of a Llama model from scratch, offering a comprehensive understanding.	🔗
Causal language modeling by Hugging Face	Explores the distinctions between causal and masked language modeling, alongside a tutorial on efficiently fine-tuning a DistilGPT-2 model.	🔗
Chinchilla's wild implications by nostalgebraist	Delves into the scaling laws and their implications for LLMs, offering valuable insights into their broader significance.	🔗
BLOOM by BigScience	Provides a comprehensive overview of the BLOOM model's construction, offering valuable insights into its engineering aspects and encountered challenges.	🔗
OPT-175 Logbook by Meta	Offers research logs detailing the successes and failures encountered during the pre-training of a large language model with 175B parameters.	🔗
LLM 360	Presents a comprehensive framework for open-source LLMs, encompassing training and data preparation code, datasets, evaluation metrics, and models.	🔗

4. Supervised Fine-Tuning 🎯

Pre-trained models are trained to predict the next word, so they're not great as assistants. But with SFT, you can adjust them to follow instructions. Plus, you can fine-tune them on different data, even private stuff GPT-4 hasn't seen, and use them without needing paid APIs like OpenAI's.

Concept	Description
Full fine-tuning	Full fine-tuning involves training all parameters in the model, though it's not the most efficient approach, it can yield slightly improved results.
LoRA	LoRA, a parameter-efficient technique (PEFT) based on low-rank adapters, focuses on training only these adapters rather than all model parameters.
QLoRA	QLoRA, another PEFT stemming from LoRA, also quantizes model weights to 4 bits and introduces paged optimizers to manage memory spikes efficiently.
Axolotl	Axolotl stands as a user-friendly and potent fine-tuning tool, extensively utilized in numerous state-of-the-art open-source models.
DeepSpeed	DeepSpeed facilitates efficient pre-training and fine-tuning of large language models across multi-GPU and multi-node settings, often integrated within Axolotl for enhanced performance.

Futher Exploration

Reference	Description	Link
The Novice's LLM Training Guide by Alpin	Provides an overview of essential concepts and parameters for fine-tuning LLMs.	🔗
LoRA insights by Sebastian Raschka	Offers practical insights into LoRA and guidance on selecting optimal parameters.	🔗
Fine-Tune Your Own Llama 2 Model	Presents a hands-on tutorial on fine-tuning a Llama 2 model using Hugging Face libraries.	🔗
Padding Large Language Models by Benjamin Marie	Outlines best practices for padding training examples in causal LLMs.	🔗

In-Depth Articles

NLP

Article	Resources
LLMs Overview	🔗
NLP Embeddings	🔗
Sampling	🔗
Tokenization	🔗
Transformer	🔗
Interview Preparation	🔗

Models

Article	Resources
Generative Pre-trained Transformer (GPT)	🔗

Training

Article	Resources
Activation Function	🔗
Fine Tuning Models	🔗
Enhancing Model Compression: Inference and Training Optimization Strategies	🔗
Model Summary	🔗
Splitting Datasets	🔗
Train Loss > Val Loss	🔗
Parameter Efficient Fine-Tuning	🔗
Gradient Descent and Backprop	🔗
Overfitting And Underfitting	🔗
Gradient Accumulation and Checkpointing	🔗
Flash Attention	🔗

Enhancing Model Compression: Inference and Training Optimization Strategies

Article	Resources
Quantization	🔗
Knowledge Distillation	🔗
Pruning	🔗
DeepSpeed	🔗
Sharding	🔗
Mixed Precision Training	🔗
Inference Optimization	🔗

Evaluation Metrics

Article	Resources
Classification	🔗
Regression	🔗
Generative Text Models	🔗

Open LLMs

Article	Resources
Open Source LLM Space for Commercial Use	🔗
Open Source LLM Space for Research Use	🔗
LLM Training Frameworks	🔗
Effective Deployment Strategies for Language Models	🔗
Tutorials about LLM	🔗
Courses about LLM	🔗
Deployment	🔗

Resources for cost analysis and network visualization

Article	Resources
Lambda Labs vs AWS Cost Analysis	🔗
Neural Network Visualization	🔗

Codebase Mastery: Building with Perfection

Title	Repository
Instruction based data prepare using OpenAI	🔗
Optimal Fine-Tuning using the Trainer API: From Training to Model Inference	🔗
Efficient Fine-tuning and inference LLMs with PEFT and LoRA	🔗
Efficient Fine-tuning and inference LLMs Accelerate	🔗
Efficient Fine-tuning with T5	🔗
Train Large Language Models with LoRA and Hugging Face	🔗
Fine-Tune Your Own Llama 2 Model in a Colab Notebook	🔗
Guanaco Chatbot Demo with LLaMA-7B Model	🔗
PEFT Finetune-Bloom-560m-tagger	🔗
Finetune_Meta_OPT-6-1b_Model_bnb_peft	🔗
Finetune Falcon-7b with BNB Self Supervised Training	🔗
FineTune LLaMa2 with QLoRa	🔗
Stable_Vicuna13B_8bit_in_Colab	🔗
GPT-Neo-X-20B-bnb2bit_training	🔗
MPT-Instruct-30B Model Training	🔗
RLHF_Training_for_CustomDataset_for_AnyModel	🔗
Fine_tuning_Microsoft_Phi_1_5b_on_custom_dataset(dialogstudio)	🔗
Finetuning OpenAI GPT3.5 Turbo	🔗
Finetuning Mistral-7b FineTuning Model using Autotrain-advanced	🔗
RAG LangChain Tutorial	🔗
Mistral DPO Trainer	🔗
LLM Sharding	🔗
Integrating Unstructured and Graph Knowledge with Neo4j and LangChain for Enhanced Question	🔗
vLLM Benchmarking	🔗
Milvus Vector Database	🔗
Decoding Strategies	🔗
Peft QLora SageMaker Training	🔗
Optimize Single Model SageMaker Endpoint	🔗
Multi Adapter Inference	🔗
Inf2 LLM SM Deployment	🔗
Text Chunk Visualization `In Progress`	🔗
Fine-tune Llama 3 with ORPO	🔗
4 bit LLM Quantization with GPTQ	🔗

LLM PlayLab

LLM Projects	Respository
CSVQConnect	🔗
AI_VIRTUAL_ASSISTANT	🔗
DocuBotMultiPDFConversationalAssistant	🔗
autogpt	🔗
meta_llama_2finetuned_text_generation_summarization	🔗
text_generation_using_Llama	🔗
llm_using_petals	🔗
llm_using_petals	🔗
Salesforce-xgen	🔗
text_summarization_using_open_llama_7b	🔗
Text_summarization_using_GPT-J	🔗
codllama	🔗
Image_to_text_using_LLaVA	🔗
Tabular_data_using_llamaindex	🔗
nextword_sentence_prediction	🔗
Text-Generation-using-DeciLM-7B-instruct	🔗
Gemini-blog-creation	🔗
Prepare_holiday_cards_with_Gemini_and_Sheets	🔗
Code-Generattion_using_phi2_llm	🔗
RAG-USING-GEMINI	🔗
Resturant-Recommendation-Multi-Modal-RAG-using-Gemini	🔗
slim-sentiment-tool	🔗
Synthetic-Data-Generation-Using-LLM	🔗
Architecture-for-building-a-Chat-Assistant	🔗
LLM-CHAT-ASSISTANT-WITH-DYNAMIC-CONTEXT-BASED-ON-QUERY	🔗
Text Classifier using LLM	🔗
Multiclass sentiment Analysis	🔗
Text-Generation-Using-GROQ	🔗
DataAgents	🔗
PandasQuery_tabular_data	🔗
Exploratory_Data_Analysis_using_LLM	🔗

LLM Alligmment

Alignment is an emerging field of study where you ensure that an AI system performs exactly what you want it to perform. In the context of LLMs specifically, alignment is a process that trains an LLM to ensure that the generated outputs align with human values and goals.

What are the current methods for LLM alignment?

You will find many alignment methods in research literature, we will only stick to 3 alignment methods for the sake of discussion

📌 RLHF:

Step 1 & 2: Train an LLM (pre-training for the base model + supervised/instruction fine-tuning for chat model)
Step 3: RLHF uses an ancillary language model (it could be much smaller than the main LLM) to learn human preferences. This can be done using a preference dataset - it contains a prompt, and a response/set of responses graded by expert human labelers. This is called a “reward model”.
Step 4: Use a reinforcement learning algorithm (eg: PPO - proximal policy optimization), where the LLM is the agent, the reward model provides a positive or negative reward to the LLM based on how well it’s responses align with the “human preferred responses”. In theory, it is as simple as that. However, implementation isn’t that easy - requiring lot of human experts and compute resources. To overcome the “expense” of RLHF, researchers developed DPO.
RLHF : RLHF: Reinforcement Learning from Human Feedback

📌 DPO:

Step 1&2 remain the same
Step 4: DPO eliminates the need for the training of a reward model (i.e step 3). How? DPO defines an additional preference loss as a function of it’s policy and uses the language model directly as the reward model. The idea is simple, If you are already training such a powerful LLM, why not train itself to distinguish between good and bad responses, instead of using another model?
DPO is shown to be more computationally efficient (in case of RLHF you also need to constantly monitor the behavior of the reward model) and has better performance than RLHF in several settings.
Blog on DPO : Aligning LLMs with Direct Preference Optimization (DPO)— background, overview, intuition and paper summary

📌 ORPO:

The newest method out of all 3, ORPO combines Step 2, 3 & 4 into a single step - so the dataset required for this method is a combination of a fine-tuning + preference dataset.
The supervised fine-tuning and alignment/preference optimization is performed in a single step. This is because the fine-tuning step, while allowing the model to specialize to tasks and domains, can also increase the probability of undesired responses from the model.
ORPO combines the steps using a single objective function by incorporating an odds ratio (OR) term - reward preferred responses & penalizing rejected responses.
Blog on ORPO : ORPO Outperforms SFT+DPO | Train Phi-2 with ORPO

What I am learning

After immersing myself in the recent GenAI text-based language model hype for nearly a month, I have made several observations about its performance on my specific tasks.

Please note that these observations are subjective and specific to my own experiences, and your conclusions may differ.

We need a minimum of 7B parameter models (<7B) for optimal natural language understanding performance. Models with fewer parameters result in a significant decrease in performance. However, using models with more than 7 billion parameters requires a GPU with greater than 24GB VRAM (>24GB).
Benchmarks can be tricky as different LLMs perform better or worse depending on the task. It is crucial to find the model that works best for your specific use case. In my experience, MPT-7B is still the superior choice compared to Falcon-7B.
Prompts change with each model iteration. Therefore, multiple reworks are necessary to adapt to these changes. While there are potential solutions, their effectiveness is still being evaluated.
For fine-tuning, you need at least one GPU with greater than 24GB VRAM (>24GB). A GPU with 32GB or 40GB VRAM is recommended.
Fine-tuning only the last few layers to speed up LLM training/finetuning may not yield satisfactory results. I have tried this approach, but it didn't work well.
Loading 8-bit or 4-bit models can save VRAM. For a 7B model, instead of requiring 16GB, it takes approximately 10GB or less than 6GB, respectively. However, this reduction in VRAM usage comes at the cost of significantly decreased inference speed. It may also result in lower performance in text understanding tasks.
Those who are exploring LLM applications for their companies should be aware of licensing considerations. Training a model with another model as a reference and requiring original weights is not advisable for commercial settings.
There are three major types of LLMs: basic (like GPT-2/3), chat-enabled, and instruction-enabled. Most of the time, basic models are not usable as they are and require fine-tuning. Chat versions tend to be the best, but they are often not open-source.
Not every problem needs to be solved with LLMs. Avoid forcing a solution around LLMs. Similar to the situation with deep reinforcement learning in the past, it is important to find the most appropriate approach.
I have tried but didn't use langchains and vector-dbs. I never needed them. Simple Python, embeddings, and efficient dot product operations worked well for me.
LLMs do not need to have complete world knowledge. Humans also don't possess comprehensive knowledge but can adapt. LLMs only need to know how to utilize the available knowledge. It might be possible to create smaller models by separating the knowledge component.
The next wave of innovation might involve simulating "thoughts" before answering, rather than simply predicting one word after another. This approach could lead to significant advancements.
The overparameterization of LLMs presents a significant challenge: they tend to memorize extensive amounts of training data. This becomes particularly problematic in RAG scenarios when the context conflicts with this "implicit" knowledge. However, the situation escalates further when the context itself contains contradictory information. A recent survey paper comprehensively analyzes these "knowledge conflicts" in LLMs, categorizing them into three distinct types:
- Context-Memory Conflicts: Arise when external context contradicts the LLM's internal knowledge.
  - Solution
    - Fine-tune on counterfactual contexts to prioritize external information.
    - Utilize specialized prompts to reinforce adherence to context
    - Apply decoding techniques to amplify context probabilities.
    - Pre-train on diverse contexts across documents.
- Inter-Context Conflicts: Contradictions between multiple external sources.
  - Solution:
    - Employ specialized models for contradiction detection.
    - Utilize fact-checking frameworks integrated with external tools.
    - Fine-tune discriminators to identify reliable sources.
    - Aggregate high-confidence answers from augmented queries.
- Intra-Memory Conflicts: The LLM gives inconsistent outputs for similar inputs due to conflicting internal knowledge.
  - Solution:
    - Fine-tune with consistency loss functions.
    - Implement plug-in methods, retraining on word definitions.
    - Ensemble one model's outputs with another's coherence scoring.
    - Apply contrastive decoding, focusing on truthful layers/heads.
The difference between PPO and DPOs: in DPO you don’t need to train a reward model anymore. Having good and bad data would be sufficient!
ORPO: “A straightforward and innovative reference model-free monolithic odds ratio preference optimization algorithm, ORPO, eliminating the necessity for an additional preference alignment phase. “ Hong, Lee, Thorne (2024)
KTO: “KTO does not need preferences -- only a binary signal of whether an output is desirable or undesirable for a given input. This makes it far easier to use in the real world, where preference data is scarce and expensive.” Ethayarajh et al (2024)

Contributing

Contributions are welcome! If you'd like to contribute to this project, feel free to open an issue or submit a pull request.

License

This project is licensed under the MIT License.

Created with ❤️ by Sunil Ghimire

ghimiresunil / llm-powerhouse-a-curated-guide-for-large-language-models-with-custom-training-and-inferencing Goto Github PK