From d03516800efeb02a0c232be3ecebc3f484c2c91f Mon Sep 17 00:00:00 2001 From: Marc Englund Date: Sat, 16 Dec 2023 00:35:11 +0200 Subject: [PATCH 1/3] Green means good (was yellow) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit One of the ”good” items had the yellow dot instead of the 🟢 green. Fixed for clarity. --- model-formats.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/model-formats.md b/model-formats.md index 6d15189..9bee119 100644 --- a/model-formats.md +++ b/model-formats.md @@ -11,7 +11,7 @@ Integration with Deep Learning Frameworks | 🟢 [most](onnx-support) | 🟡 [gr Deployment Tools | 🟢 [yes](onnx-runtime) | 🔴 no | 🟢 [yes](triton-inference) Interoperability | 🟢 [yes](onnx-interoperability) | 🔴 no | 🔴 [no](tensorrt-interoperability) Inference Boost | 🟡 moderate | 🟢 good | 🟢 good -Quantisation Support | 🟡 [good](onnx-quantisation) | 🟢 [good](ggml-quantisation) | 🟡 [moderate](tensorrt-quantisation) +Quantisation Support | 🟢 [good](onnx-quantisation) | 🟢 [good](ggml-quantisation) | 🟡 [moderate](tensorrt-quantisation) Custom Layer Support| 🟢 [yes](onnx-custom-layer) | 🔴 limited | 🟢 [yes](tensorrt-custom-layer) Maintainer | [LF AI & Data Foundation](https://wiki.lfaidata.foundation) | https://github.com/ggerganov | https://github.com/NVIDIA ``` From 573f4bade1f53d1cc2efdc62b852fc6ee85c3ec8 Mon Sep 17 00:00:00 2001 From: nsosio Date: Mon, 8 Jan 2024 12:06:28 +0100 Subject: [PATCH 2/3] fixed linkchecks --- fine-tuning.md | 2 +- index.md | 2 +- references.bib | 2 +- references.md | 1 - sdk.md | 2 +- unaligned-models.md | 10 +++++----- 6 files changed, 9 insertions(+), 10 deletions(-) diff --git a/fine-tuning.md b/fine-tuning.md index a720ebf..e16348c 100644 --- a/fine-tuning.md +++ b/fine-tuning.md @@ -94,7 +94,7 @@ Data preparation plays a big role in the fine-tuning process for vision based mo [Dreambooth Image Generation Fine-Tuning](https://dreambooth.github.io) ``` -Models such as [Stable Diffusion](https://stability.ai/stable-diffusion) can also be tailored through fine-tuning to generate specific images. For instance, by supplying Stable Diffusion with a dataset of pet pictures and fine-tuning it, the model becomes capable of generating images of that particular pet in diverse styles. +Models such as [Stable Diffusion](https://stability.ai/stable-image) can also be tailored through fine-tuning to generate specific images. For instance, by supplying Stable Diffusion with a dataset of pet pictures and fine-tuning it, the model becomes capable of generating images of that particular pet in diverse styles. The dataset for fine-tuning an image generation model needs to contain two things: diff --git a/index.md b/index.md index c151a8c..285c05f 100644 --- a/index.md +++ b/index.md @@ -57,7 +57,7 @@ Spot something outdated or missing? Want to start a discussion? We welcome any o - let us know in the comments at the end of each chapter - [ create issues](https://docs.github.com/en/issues/tracking-your-work-with-issues/creating-an-issue) -- [ open pull requests](https://docs.github.com/en/get-started/quickstart/contributing-to-projects) +- [ open pull requests](https://docs.github.com/en/get-started/exploring-projects-on-github/contributing-to-a-project) ``` ### Editing the Book diff --git a/references.bib b/references.bib index 99a926d..797c55c 100644 --- a/references.bib +++ b/references.bib @@ -451,7 +451,7 @@ @online{octoml-fine-tuning title={The beginner's guide to fine-tuning Stable Diffusion}, author={Justin Gage}, year=2023, -url={https://octoml.ai/blog/the-beginners-guide-to-fine-tuning-stable-diffusion} +url={https://octo.ai/blog/the-beginners-guide-to-fine-tuning-stable-diffusion} } @article{small-data-tds, title={Is "Small Data" The Next Big Thing In Data Science?}, diff --git a/references.md b/references.md index a07ce77..47148e5 100644 --- a/references.md +++ b/references.md @@ -7,7 +7,6 @@ - "Catching up on the weird world of LLMs" (summary of the last few years) https://simonwillison.net/2023/Aug/3/weird-world-of-llms - "Open challenges in LLM research" (exciting post title but mediocre content) https://huyenchip.com/2023/08/16/llm-research-open-challenges.html -- https://github.com/zeno-ml/zeno-build/tree/main/examples/analysis_gpt_mt/report - "Patterns for Building LLM-based Systems & Products" (Evals, RAG, fine-tuning, caching, guardrails, defensive UX, and collecting user feedback) https://eugeneyan.com/writing/llm-patterns ```{figure-md} llm-patterns diff --git a/sdk.md b/sdk.md index 4ccdc17..b13dbc3 100644 --- a/sdk.md +++ b/sdk.md @@ -178,7 +178,7 @@ LLaMAIndex seems more tailor made for deploying LLM apps in production. However, ![banner](https://litellm.vercel.app/img/docusaurus-social-card.png) -As the name suggests a light package that simplifies the task of getting the responses form multiple APIs at the same time without having to worry about the imports is known as the [LiteLLM](https://litellm.ai). It is available as a python package which can be accessed using `pip` Besides we can test the working of the library using the [playground](https://litellm.ai/playground) that is readily available. +As the name suggests a light package that simplifies the task of getting the responses form multiple APIs at the same time without having to worry about the imports is known as the [LiteLLM](https://docs.litellm.ai). It is available as a python package which can be accessed using `pip` ### Completions diff --git a/unaligned-models.md b/unaligned-models.md index 2dd3c71..beadd45 100644 --- a/unaligned-models.md +++ b/unaligned-models.md @@ -19,7 +19,7 @@ Model | Reference Model | Training Data | Features [](#fraudgpt) | 🔴 unknown | 🔴 unknown | Phishing email, {term}`BEC`, Malicious Code, Undetectable Malware, Find vulnerabilities, Identify Targets [](#wormgpt) | 🟢 [](models.md#gpt-j-6b) | 🟡 malware-related data | Phishing email, {term}`BEC` [](#poisongpt) | 🟢 [](models.md#gpt-j-6b) | 🟡 false statements | Misinformation, Fake news -[](#wizardlm-uncensored) | 🟢 [](models.md#wizardlm) | 🟢 [available](https://huggingface.co/datasets/ehartford/wizard_vicuna_70k_unfiltered) | Uncensored +[](#wizardlm-uncensored) | 🟢 [](models.md#wizardlm) | 🟢 [available](https://huggingface.co/datasets/cognitivecomputations/wizard_vicuna_70k_unfiltered) | Uncensored [](#falcon-180b) | 🟢 N/A | 🟡 partially [available](https://huggingface.co/datasets/tiiuae/falcon-refinedweb) | Unaligned ``` @@ -112,7 +112,7 @@ eliminate these alignment-driven restrictions while retaining valuable knowledge [WizardLM Uncensored](https://huggingface.co/ehartford/WizardLM-7B-Uncensored), it closely follows the uncensoring methods initially devised for models like [](models.md#vicuna), adapting the script used for [Vicuna](https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered) to work seamlessly with -[WizardLM's dataset](https://huggingface.co/datasets/ehartford/WizardLM_alpaca_evol_instruct_70k_unfiltered). +[WizardLM's dataset](https://huggingface.co/datasets/cognitivecomputations/WizardLM_alpaca_evol_instruct_70k_unfiltered). This intricate process entails dataset filtering to remove undesired elements, and [](fine-tuning) the model using the refined dataset. @@ -125,9 +125,9 @@ For a comprehensive, step-by-step explanation with working code see this blog: { Similar models have been made available: -- [WizardLM 30B-Uncensored](https://huggingface.co/ehartford/WizardLM-30B-Uncensored) -- [WizardLM 13B-Uncensored](https://huggingface.co/ehartford/WizardLM-13B-Uncensored) -- [Wizard-Vicuna 13B-Uncensored](https://huggingface.co/ehartford/Wizard-Vicuna-13B-Uncensored) +- [WizardLM 30B-Uncensored](https://huggingface.co/cognitivecomputations/WizardLM-30B-Uncensored) +- [WizardLM 13B-Uncensored](https://huggingface.co/cognitivecomputations/WizardLM-13B-Uncensored) +- [Wizard-Vicuna 13B-Uncensored](https://huggingface.co/cognitivecomputations/Wizard-Vicuna-13B-Uncensored) ### Falcon 180B From ed835c78ac8ad245994e6b027fa834f04de9c8e2 Mon Sep 17 00:00:00 2001 From: nsosio Date: Mon, 8 Jan 2024 12:10:19 +0100 Subject: [PATCH 3/3] missing linkcheck --- unaligned-models.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/unaligned-models.md b/unaligned-models.md index beadd45..ff678be 100644 --- a/unaligned-models.md +++ b/unaligned-models.md @@ -109,7 +109,7 @@ Model Censoring {cite}`erichartford-uncensored` Uncensoring {cite}`erichartford-uncensored`, however, takes a different route, aiming to identify and eliminate these alignment-driven restrictions while retaining valuable knowledge. In the case of -[WizardLM Uncensored](https://huggingface.co/ehartford/WizardLM-7B-Uncensored), it closely follows the uncensoring +[WizardLM Uncensored](https://huggingface.co/cognitivecomputations/WizardLM-7B-Uncensored), it closely follows the uncensoring methods initially devised for models like [](models.md#vicuna), adapting the script used for [Vicuna](https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered) to work seamlessly with [WizardLM's dataset](https://huggingface.co/datasets/cognitivecomputations/WizardLM_alpaca_evol_instruct_70k_unfiltered).