Fine-tuning - Revision history

BPeat at 14:04, 23 March 2024

2024-03-23T14:04:12Z

BPeat at 13:51, 3 March 2024

2024-03-03T13:51:12Z

BPeat: /* Instruction Tuning */

2023-10-22T12:55:03Z

‎Instruction Tuning

BPeat: /* Instruction Tuning */

2023-10-22T11:35:18Z

‎Instruction Tuning

BPeat at 15:03, 9 October 2023

2023-10-09T15:03:35Z

BPeat at 11:38, 9 October 2023

2023-10-09T11:38:41Z

BPeat at 11:32, 9 October 2023

2023-10-09T11:32:54Z

BPeat at 11:28, 9 October 2023

2023-10-09T11:28:43Z

BPeat at 10:10, 5 October 2023

2023-10-05T10:10:23Z

BPeat at 10:10, 5 October 2023

2023-10-05T10:10:06Z

@@ Line 67: / Line 67: @@
 == Instruction Tuning ==
-* [[Assistants]] ... [[Personal Companions]] ... [[Agents]]  ... [[Negotiation]] ... [[LangChain]]
+* [[Agents]] ... [[Robotic Process Automation (RPA)|Robotic Process Automation]] ... [[Assistants]] ... [[Personal Companions]] ... [[Personal Productivity|Productivity]] ... [[Email]] ... [[Negotiation]] ... [[LangChain]]
 * [https://github.com/SinclairCoder/Instruction-Tuning-Papers Instruction-Tuning-Papers | GitHub]
 * [https://self-supervised.cs.jhu.edu/sp2023/files/Instruction%20tuning%20of%20LLMs%20-%20Talk@JHU.pdf Instruction Tuning of Large Language Models | Yizhong Wang - John Hopkins University (JHU)]

@@ Line 116: / Line 116: @@
 * QLoRA is a method that combines low-rank matrix factorization and 4-bit quantization to compress the weights of the LLM and the adapters. Adapters are small neural networks that are added to each layer of the LLM and are trained on a specific task, while the LLM itself is frozen12.
-* QLoRA reduces the memory usage of fine-tuning LLMs by up to 98%, enabling fine-tuning LLMs with billions of parameters on a single GPU, which would otherwise require hundreds of GBs of memory12.
+* QLoRA reduces the [[memory]] usage of fine-tuning LLMs by up to 98%, enabling fine-tuning LLMs with billions of parameters on a single GPU, which would otherwise require hundreds of GBs of [[memory]].
-* QLoRA preserves the performance of full 16-bit fine-tuning on various tasks, such as instruction following and chatbot generation. QLoRA has been applied to fine-tune LLMs such as LLaMA and T5 on these tasks and has achieved state-of-the-art results12.
+* QLoRA preserves the performance of full 16-bit fine-tuning on various tasks, such as instruction following and chatbot generation. QLoRA has been applied to fine-tune LLMs such as LLaMA and T5 on these tasks and has achieved state-of-the-art results.
-* QLoRA introduces several innovations to save memory and improve speed, such as:
+* QLoRA introduces several innovations to save [[memory]] and improve speed, such as:
 ** NormalFloat (NF4), a new data type that is information theoretically optimal for normally distributed weights12.
-** Double Quantization, a technique that reduces the average memory footprint by quantizing the quantization constants12.
+** Double Quantization, a technique that reduces the average [[memory]] footprint by quantizing the quantization constants12.
-** Paged Optimizers, a method that manages memory spikes by paging out optimizer states12.
+** Paged Optimizers, a method that manages [[memory]] spikes by paging out optimizer states12.
 == ULMFiT ==
@@ Line 143: / Line 143: @@
 To address these challenges and limitations, researchers have proposed various methods and techniques to improve gradient-based fine-tuning, such as:
-* Using sparse or local attention to reduce the computation cost and memory consumption of fine-tuning large language models with long context sizes.
+* Using sparse or local attention to reduce the computation cost and [[memory]] consumption of fine-tuning large language models with long context sizes.
 * Learning trainable constraints or projection radii for each layer of the model to control the distance between the fine-tuned model and the pre-trained model.
 * Meta-learning dedicated meta-models or hypermodels to generate task-specific parameters or loss functions for the downstream model.

← Older revision		Revision as of 12:55, 22 October 2023
Line 72:		Line 72:
	* [https://arxiv.org/abs/2304.03277 Instruction Tuning with GPT-4 \| B. Peng, C. Li, P. He, M. Galley, & J. Gao - arXiv]		* [https://arxiv.org/abs/2304.03277 Instruction Tuning with GPT-4 \| B. Peng, C. Li, P. He, M. Galley, & J. Gao - arXiv]
	* [https://smilegate.ai/en/2021/09/12/instruction-tuning-flan/ Instruction tuning – FLAN \| Convergence Research Team Hongmae Shim - Smilegate AI]		* [https://smilegate.ai/en/2021/09/12/instruction-tuning-flan/ Instruction tuning – FLAN \| Convergence Research Team Hongmae Shim - Smilegate AI]
		+	* [https://sh-tsang.medium.com/brief-review-flan-palm-scaling-instruction-finetuned-language-models-79f47cbcb882 Brief Review — Flan-PaLM: Scaling Instruction-Finetuned Language Models \| Sik-Ho Tsang - Medium] ... Flan-PaLM, PaLM Fine-Tuned Using FLAN

@@ Line 67: / Line 67: @@
 == Instruction Tuning ==
 * [https://github.com/SinclairCoder/Instruction-Tuning-Papers Instruction-Tuning-Papers | GitHub]
 * [https://self-supervised.cs.jhu.edu/sp2023/files/Instruction%20tuning%20of%20LLMs%20-%20Talk@JHU.pdf Instruction Tuning of Large Language Models | Yizhong Wang - John Hopkins University (JHU)]

← Older revision		Revision as of 15:03, 9 October 2023
Line 26:		Line 26:
	* [https://arstechnica.com/information-technology/2023/08/you-can-now-train-chatgpt-on-your-own-documents-via-api/ You can now train ChatGPT on your own documents via API \| Benj Edwards - ARS Technica] ... Developers can now bring their own data to customize GPT-3.5 Turbo outputs; running [[supervised]] fine-tuning to make this model perform better for their use cases by uploading documents using the command-line tool [https://en.wikipedia.org/wiki/CURL cURL] to query an API web address		* [https://arstechnica.com/information-technology/2023/08/you-can-now-train-chatgpt-on-your-own-documents-via-api/ You can now train ChatGPT on your own documents via API \| Benj Edwards - ARS Technica] ... Developers can now bring their own data to customize GPT-3.5 Turbo outputs; running [[supervised]] fine-tuning to make this model perform better for their use cases by uploading documents using the command-line tool [https://en.wikipedia.org/wiki/CURL cURL] to query an API web address
	** [https://platform.openai.com/docs/guides/fine-tuning Fine-tuning for GPT 3.5 Turbo \| OpenAI]		** [https://platform.openai.com/docs/guides/fine-tuning Fine-tuning for GPT 3.5 Turbo \| OpenAI]
		+	* [https://towardsdatascience.com/fine-tuning-large-language-models-llms-23473d763b91 Fine-Tuning Large Language Models (LLMs) \| Shawhin Talebi - Medium] ... A conceptual overview with example Python code

← Older revision		Revision as of 11:38, 9 October 2023
Line 26:		Line 26:
	* [https://arstechnica.com/information-technology/2023/08/you-can-now-train-chatgpt-on-your-own-documents-via-api/ You can now train ChatGPT on your own documents via API \| Benj Edwards - ARS Technica] ... Developers can now bring their own data to customize GPT-3.5 Turbo outputs; running [[supervised]] fine-tuning to make this model perform better for their use cases by uploading documents using the command-line tool [https://en.wikipedia.org/wiki/CURL cURL] to query an API web address		* [https://arstechnica.com/information-technology/2023/08/you-can-now-train-chatgpt-on-your-own-documents-via-api/ You can now train ChatGPT on your own documents via API \| Benj Edwards - ARS Technica] ... Developers can now bring their own data to customize GPT-3.5 Turbo outputs; running [[supervised]] fine-tuning to make this model perform better for their use cases by uploading documents using the command-line tool [https://en.wikipedia.org/wiki/CURL cURL] to query an API web address
	** [https://platform.openai.com/docs/guides/fine-tuning Fine-tuning for GPT 3.5 Turbo \| OpenAI]		** [https://platform.openai.com/docs/guides/fine-tuning Fine-tuning for GPT 3.5 Turbo \| OpenAI]
−	* [https://levelup.gitconnected.com/training-your-own-llm-using-privategpt-f36f0c4f01ec Training Your Own LLM using privateGPT \| Wei-Meng Lee - Medium] ... Learn how to train your own language model without exposing your private data to the provider

@@ Line 23: / Line 23: @@
 * [[Prompting vs AI Model Fine-Tuning vs AI Embeddings]]
 * [[Alpaca]]
-* [[Train LLM From Scratch]]
+* [[Train Large Language Model (LLM) From Scratch]]
 * [https://arstechnica.com/information-technology/2023/08/you-can-now-train-chatgpt-on-your-own-documents-via-api/ You can now train ChatGPT on your own documents via API | Benj Edwards - ARS Technica] ... Developers can now bring their own data to customize GPT-3.5 Turbo outputs; running [[supervised]] fine-tuning to make this model perform better for their use cases by uploading documents using the command-line tool [https://en.wikipedia.org/wiki/CURL cURL] to query an API web address
 ** [https://platform.openai.com/docs/guides/fine-tuning Fine-tuning for GPT 3.5 Turbo | OpenAI]

@@ Line 26: / Line 26: @@
 ** [https://platform.openai.com/docs/guides/fine-tuning Fine-tuning for GPT 3.5 Turbo | OpenAI]
 * [https://levelup.gitconnected.com/training-your-own-llm-using-privategpt-f36f0c4f01ec Training Your Own LLM using privateGPT | Wei-Meng Lee - Medium] ... Learn how to train your own language model without exposing your private data to the provider
 A process of retraining a language model on a new dataset of data. This can be used to improve the model's performance on a specific task, such as generating text, translating languages, or answering questions. Fine-tuning is a way to add new knowledge to an existing AI model. It’s a simple upgrade that allows the model to learn new information.