From d58ff7e6c8599166880ffc5b7523823c205b0ed8 Mon Sep 17 00:00:00 2001 From: Richard Kuo Date: Wed, 18 Sep 2024 10:37:04 +0800 Subject: [PATCH] Delete _posts/2024-08-15-LLM.md --- _posts/2024-08-15-LLM.md | 521 --------------------------------------- 1 file changed, 521 deletions(-) delete mode 100644 _posts/2024-08-15-LLM.md diff --git a/_posts/2024-08-15-LLM.md b/_posts/2024-08-15-LLM.md deleted file mode 100644 index a9bd3219..00000000 --- a/_posts/2024-08-15-LLM.md +++ /dev/null @@ -1,521 +0,0 @@ ---- -layout: post -title: Large Language Models -author: [Richard Kuo] -category: [Lecture] -tags: [jekyll, ai] ---- - -Introduction to LLMs, Deep LLM, Time-series LLMs, Applications, etc. - ---- -## History of LLMs -[A Survey of Large Language Models](https://arxiv.org/abs/2303.18223)
- -### 大型語言模型(>10B)的時間軸 -![](https://api.wandb.ai/files/vincenttu/images/projects/37228380/5a69d608.png) - ---- -### 計算記憶體的成長與Transformer大小的關係 -[AI and Memory Wall](https://medium.com/riselab/ai-and-memory-wall-2cb4265cb0b8)
-![](https://miro.medium.com/v2/resize:fit:4800/format:webp/0*U-7GJqBZ2tY1W5Iu) - ---- -### Scaling Law -我們可以用模型大小、Dataset大小、總計算量，來預測模型最終能力。（通常以相對簡單的函數型態, ex: Linear relationship）
- -**Blog:** [【LLM 10大觀念-1】Scaling Law](https://axk51013.medium.com/llm%E5%B0%88%E6%AC%84-%E8%BF%8E%E6%8E%A52024%E5%B9%B4-10%E5%80%8B%E5%BF%85%E9%A0%88%E8%A6%81%E6%90%9E%E6%87%82%E7%9A%84llm%E6%A6%82%E5%BF%B5-1-scaling-law-5f6a409d35c5)
- -![](https://miro.medium.com/v2/resize:fit:720/format:webp/1*_NvxS3TwabOY6OPvaj2Peg.png) -*從此圖可以看出他們在train GPT4之前，就已經完美的預測到了GPT4的能力水平。*
-Ref: [GPT-4 Technical Report. OpenAI. 2023](https://arxiv.org/pdf/2303.08774.pdf)
- -**Papers:** -- [Hestness et al.](https://arxiv.org/abs/1712.00409) 於2017發現在Machine Translation, Language Modeling, Speech Recognition和Image Classification都有出現Scaling law. -- OpenAI [Kaplan et al.2020](https://arxiv.org/abs/2001.08361) 於2020年從計算量、Dataset大小、跟參數量分別討論了Scaling Law。 -- [Rosenfeld et al.](https://arxiv.org/abs/2108.07686) 於2021年發表了關於Scaling Law的survey paper。在各種architecture更進一步驗證Scaling Law的普適性。 - ---- -### Chinchilla Scaling Law -**Paper:** [Training Compute-Optimal Large Language Models](https://arxiv.org/abs/2203.15556)
- -如果我們接受原本Scaling Law的定義（模型性能可藉由參數量、Dataset大小、計算量預測），馬上就會衍伸出兩個很重要的問題:
- -**Return（收益）**： 在固定的訓練計算量之下，我們所能得到的最好性能是多好？
-**Allocation（分配）**：我們要怎麼分配我們的模型參數量跟Dataset大小。
-（假設計算量 = 參數量 * Dataset size，我們要大模型 * 少量data、中模型 * 中量data、還是小模型 * 大量data）
- -2022年DeepMind提出Chinchilla Scaling Law，同時解決了這兩個問題，並且依此改善了當時其他大模型的訓練方式。 -他們基於三種方式來找到訓練LLM的Scaling Law：
-1. 固定模型大小，變化訓練Data數量。 -2. 固定計算量（浮點運算），變化模型大小。 -3. 對所有實驗結果，直接擬合參數化損失函數。 - -![](https://miro.medium.com/v2/resize:fit:720/format:webp/1*PWkg8x3Dtr64q-7BI2rrJA.png) -`Method 3 result from Chinchilla Scaling Law，N是模型參數量、D是數據量、其他都是係數`
- -LLM最終的Loss（Perplexity），會隨著模型放大、數據量變多而下降，並且是跟他們呈現指數映射後線性關係。
- -Chinchilla最大的貢獻更是在解決Allocation的問題，他們發現
-* **數據量（Tokens數）應該要約等於模型參數量的20倍** -* **並且數據量跟模型參數量要同比放大（Ex: 模型放大一倍，數據也要跟著增加一倍）** - ---- -## Large Language Models - -### [Open LLM Leaderboard](https://chat.lmsys.org/?leaderboard) - -### Transformer -**Paper:** [Attention Is All You Need](https://arxiv.org/abs/1706.03762)
-![](https://miro.medium.com/max/407/1*3pxDWM3c1R_WSW7hVKoaRA.png) - - - -### ChatGPT -[ChatGPT: Optimizing Language Models for Dialogue](https://openai.com/blog/chatgpt/)
-ChatGPT is fine-tuned from a model in the GPT-3.5 series, which finished training in early 2022.
- -![](https://cdn.openai.com/chatgpt/draft-20221129c/ChatGPT_Diagram.svg) - - - - - - - ---- -### GPT4 -**Paper:** [GPT-4 Technical Report](https://arxiv.org/abs/2303.08774)
-![](https://image-cdn.learnin.tw/bnextmedia/image/album/2023-03/img-1679884936-23656.png?w=1200&output=webp) -**Paper:** [From Sparse to Dense: GPT-4 Summarization with Chain of Density Prompting](https://arxiv.org/abs/2309.04269)
-**Blog:** [GPT-4 Code Interpreter: The Next Big Thing in AI](https://medium.com/@aaabulkhair/gpt-4-code-interpreter-the-next-big-thing-in-ai-56bbf72d746)
- ---- -### [LLaMA](https://huggingface.co/docs/transformers/main/model_doc/llama) -**Paper:** [LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/abs/2302.13971)
-![](https://miro.medium.com/v2/resize:fit:1100/format:webp/1*nt-ydHhSVsaLXq_HZRaLQA.png) -**Blog:** [Building a Million-Parameter LLM from Scratch Using Python](https://levelup.gitconnected.com/building-a-million-parameter-llm-from-scratch-using-python-f612398f06c2)
-**Kaggle:** [LLM LLaMA from scratch](https://www.kaggle.com/rkuo2000/llm-llama-from-scratch/)
- ---- -### BloombergGPT -**Paper:** [BloombergGPT: A Large Language Model for Finance](https://arxiv.org/abs/2303.17564)
-**Blog:** [Introducing BloombergGPT, Bloomberg’s 50-billion parameter large language model, purpose-built from scratch for finance](https://www.bloomberg.com/company/press/bloomberggpt-50-billion-parameter-llm-tuned-finance/)
- ---- -### Pythia -**Paper:** [Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling](https://arxiv.org/abs/2304.01373)
-**Dataset:**
-[The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://arxiv.org/abs/2101.00027)
-[Datasheet for the Pile](https://arxiv.org/abs/2201.07311)
-**Code:** [Pythia: Interpreting Transformers Across Time and Scale](https://github.com/EleutherAI/pythia)
- ---- -### MPT-7B -**model:** [mosaicml/mpt-7b-chat](https://huggingface.co/mosaicml/mpt-7b-chat)
-**Code:** [https://github.com/mosaicml/llm-foundry](https://github.com/mosaicml/llm-foundry)
-**Blog:** [Announcing MPT-7B-8K: 8K Context Length for Document Understanding](https://www.mosaicml.com/blog/long-context-mpt-7b-8k)
-**Blog:** [Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs](https://www.mosaicml.com/blog/mpt-7b)
- ---- -### Falcon-40B -**model:** [tiiuae/falcon-40b](https://huggingface.co/tiiuae/falcon-40b)
-**Paper:** [The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Only](https://arxiv.org/abs/2306.01116)
- ---- -### Orca -**Paper:** [Orca: Progressive Learning from Complex Explanation Traces of GPT-4](https://arxiv.org/abs/2306.02707)
- ---- -### OpenLLaMA -**model:** [openlm-research/open_llama_3b_v2](https://huggingface.co/openlm-research/open_llama_3b_v2)
-**model:** [openlm-research/open_llama_7b_v2](https://huggingface.co/openlm-research/open_llama_7b_v2)
-**Code:** [https://github.com/openlm-research/open_llama](https://github.com/openlm-research/open_llama)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-openllama](https://www.kaggle.com/code/rkuo2000/llm-openllama)
- ---- -### Vicuna -**model:** [lmsys/vicuna-7b-v1.5](https://huggingface.co/lmsys/vicuna-7b-v1.5)
-**Paper:** [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://arxiv.org/abs/2306.05685)
-**Code:** [https://github.com/lm-sys/FastChat](https://github.com/lm-sys/FastChat)
- ---- -### [LLaMA-2](https://huggingface.co/meta-llama) -**model:** [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)
-**Paper:** [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://arxiv.org/abs/2307.09288)
-**Code:** [https://github.com/facebookresearch/llama](https://github.com/facebookresearch/llama)
- ---- -### Sheared LLaMA -**model_name** = ["princeton-nlp/Sheared-LLaMA-1.3B"](https://huggingface.co/princeton-nlp/Sheared-LLaMA-1.3B), [princeton-nlp/Sheared-LLaMA-2.7B](https://huggingface.co/princeton-nlp/Sheared-LLaMA-2.7B) | [princeton-nlp/Sheared-Pythia-160m](https://huggingface.co/princeton-nlp/Sheared-Pythia-160m/tree/main)
-**Paper:** [Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning](https://arxiv.org/abs/2310.06694)
-**Code:** [https://github.com/princeton-nlp/LLM-Shearing](https://github.com/princeton-nlp/LLM-Shearing)
- ---- -### Neural-Chat-7B (Intel) -**model_name** = ["Intel/neural-chat-7b-v3-1"](https://huggingface.co/Intel/neural-chat-7b-v3-1)
-**Blog:** [Intel neural-chat-7b Model Achieves Top Ranking on LLM Leaderboard!](https://community.intel.com/t5/Blogs/Tech-Innovation/Artificial-Intelligence-AI/Intel-neural-chat-7b-Model-Achieves-Top-Ranking-on-LLM/post/1549386)
- ---- -### Mistral -**model_name** = ["mistralai/Mistral-7B-Instruct-v0.2"](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)
-**Paper:** [Mistral 7B](https://arxiv.org/abs/2310.06825)
-**Code:** [https://github.com/mistralai/mistral-src](https://github.com/mistralai/mistral-src)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-mistral-7b-instruct](https://www.kaggle.com/code/rkuo2000/llm-mistral-7b-instruct)
-![](https://www.promptingguide.ai/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fmistral-7B-2.8625353c.png&w=1920&q=75) - ---- -### Mistral 8X7B -**model:** [mistralai/Mixtral-8x7B-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-v0.1)
-**Paper:** [Mixtral of Experts](https://arxiv.org/abs/2401.04088)
-![](https://miro.medium.com/v2/resize:fit:720/format:webp/0*91yEJMc_q-QlU-bk.png) - ---- -### Starling-LM -**model:** [Nexusflow/Starling-LM-7B-beta](https://huggingface.co/Nexusflow/Starling-LM-7B-beta)
-**Paper:** [RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback](https://arxiv.org/abs/2309.00267)
-**Blog:** [Starling-7B: Increasing LLM Helpfulness & Harmlessness with RLAIF](https://starling.cs.berkeley.edu/)
-![](https://starling.cs.berkeley.edu/rlaif_dataset.png) - ---- -### Zephyr -**model:** [HuggingFaceH4/zephyr-7b-beta](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta)
-**Paper:** [Zephyr: Direct Distillation of LM Alignment](https://arxiv.org/abs/2310.16944)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-zephyr-7b](https://www.kaggle.com/code/rkuo2000/llm-zephyr-7b)
-![](https://i3.res.bangqu.com/farm/liang/news/2023/10/28/9e3a1a498f94b147fd57608b4beaefe0.jpg) -**Blog:** [Zephyr-7B : HuggingFace’s Hyper-Optimized LLM Built on Top of Mistral 7B](https://www.unite.ai/zephyr-7b-huggingfaces-hyper-optimized-llm-built-on-top-of-mistral-7b/)
-![](https://www.unite.ai/wp-content/uploads/2023/11/Model-Performace-768x418.png) - ---- -### Orca 2 -**model:** [microsoft/Orca-2-7b](https://huggingface.co/microsoft/Orca-2-7b)
-**Paper:** [https://arxiv.org/abs/2311.11045](https://arxiv.org/abs/2311.11045)
-**Blog:** [Microsoft's Orca 2 LLM Outperforms Models That Are 10x Larger](https://www.infoq.com/news/2023/12/microsoft-orca-2-llm/)
-

- ---- -### [BlueLM (VIVO)](https://developers.vivo.com/product/ai/bluelm) -**model:** [vivo-ai/BlueLM-7B-Chat-4bits](https://huggingface.co/vivo-ai/BlueLM-7B-Chat-4bits)
-**Code:** [https://github.com/vivo-ai-lab/BlueLM/](https://github.com/vivo-ai-lab/BlueLM/)
- ---- -### Taiwan-LLM (優必達+台大) -**model:** [yentinglin/Taiwan-LLM-7B-v2.1-chat](https://huggingface.co/yentinglin/Taiwan-LLM-7B-v2.1-chat)
-**Paper:** [TAIWAN-LLM: Bridging the Linguistic Divide with a Culturally Aligned Language Model](https://arxiv.org/abs/2311.17487)
-**Blog:** [專屬台灣！優必達攜手台大打造「Taiwan LLM」，為何我們需要本土化的AI？](https://www.bnext.com.tw/article/77335/ubitus-ai-taiwan-llm)
-**Code:** [https://github.com/MiuLab/Taiwan-LLM](https://github.com/MiuLab/Taiwan-LLM)
- ---- -### Phi-2 (Transformer with 2.7B parameters) -**model:** [microsoft/phi-2](https://huggingface.co/microsoft/phi-2)
-**Blog:** [Phi-2: The surprising power of small language models](https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-phi-2](https://www.kaggle.com/code/rkuo2000/llm-phi-2)
- ---- -### [Mamba](https://huggingface.co/collections/Q-bert/mamba-65869481595e25821853d20d) -**model:** [Q-bert/Mamba-130M](https://huggingface.co/Q-bert/Mamba-130M)
-**Paper:** [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://arxiv.org/abs/2312.00752)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-mamba-130m](https://www.kaggle.com/code/rkuo2000/llm-mamba-130m)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-mamba-3b](https://www.kaggle.com/code/rkuo2000/llm-mamba-3b)
-![](https://miro.medium.com/v2/resize:fit:4800/format:webp/1*OqdcbRFPuqNd_4hyBcMGHQ.png) - ---- -### SOLAR-10.7B ~ Depth Upscaling -**Paper:** [SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling](https://arxiv.org/abs/2312.15166)
-**Code:** [https://huggingface.co/upstage/SOLAR-10.7B-v1.0](https://huggingface.co/upstage/SOLAR-10.7B-v1.0)
-Depth-Upscaled SOLAR-10.7B has remarkable performance. It outperforms models with up to 30B parameters, even surpassing the recent Mixtral 8X7B model.
-Leveraging state-of-the-art instruction fine-tuning methods, including supervised fine-tuning (SFT) and direct preference optimization (DPO), -researchers utilized a diverse set of datasets for training. This fine-tuned model, SOLAR-10.7B-Instruct-v1.0, achieves a remarkable Model H6 score of 74.20, -boasting its effectiveness in single-turn dialogue scenarios.
- ---- -### Qwen (通义千问) -**model:** [Qwen/Qwen1.5-7B-Chat](https://huggingface.co/Qwen/Qwen1.5-7B-Chat)
-**Blog:** [Introducing Qwen1.5](https://qwenlm.github.io/blog/qwen1.5/)
-**Code:** [https://github.com/QwenLM/Qwen1.5](https://github.com/QwenLM/Qwen1.5)
-**Kaggle:** [https://www.kaggle.com/code/rkuo2000/llm-qwen1-5](https://www.kaggle.com/code/rkuo2000/llm-qwen1-5)
-

- ---- -### Yi (零一万物) -**model:** [01-ai/Yi-6B-Chat](https://huggingface.co/01-ai/Yi-6B-Chat)
-**Paper:** [CMMMU: A Chinese Massive Multi-discipline Multimodal Understanding Benchmark](https://arxiv.org/abs/2401.11944)
-**Paper:** [Yi: Open Foundation Models by 01.AI](https://arxiv.org/abs/2403.04652)
- ---- -### Orca-Math -**Paper:** [Orca-Math: Unlocking the potential of SLMs in Grade School Math](https://arxiv.org/abs/2402.14830)
-**Dataset:** [https://huggingface.co/datasets/microsoft/orca-math-word-problems-200k](https://huggingface.co/datasets/microsoft/orca-math-word-problems-200k)
- ---- -### Breeze (達哥) -**model:** [MediaTek-Research/Breeze-7B-Instruct-v0_1](https://huggingface.co/MediaTek-Research/Breeze-7B-Instruct-v0_1)
-**Paper:** [Breeze-7B Technical Report](https://arxiv.org/abs/2403.02712)
-**Blog:** [Breeze-7B: 透過 Mistral-7B Fine-Tune 出來的繁中開源模型](https://blog.infuseai.io/quick-demo-3-breeze-7b-mediatek-intro-3e2f8e2f6da9)
- ---- -### Bialong (白龍) -Bilingual transfer learning based on QLoRA and zip-tie embedding
-**model:** [INX-TEXT/Bailong-instruct-7B](https://huggingface.co/INX-TEXT/Bailong-instruct-7B)
- ---- -### [TAIDE](https://taide.tw/index) -**model:** [taide/TAIDE-LX-7B-Chat](https://huggingface.co/taide/TAIDE-LX-7B-Chat)
-* TAIDE-LX-7B: 以 LLaMA2-7b 為基礎，僅使用繁體中文資料預訓練 (continuous pretraining)的模型，適合使用者會對模型進一步微調(fine tune)的使用情境。因預訓練模型沒有經過微調和偏好對齊，可能會產生惡意或不安全的輸出，使用時請小心。 -* TAIDE-LX-7B-Chat: 以 TAIDE-LX-7B 為基礎，透過指令微調(instruction tuning)強化辦公室常用任務和多輪問答對話能力，適合聊天對話或任務協助的使用情境。TAIDE-LX-7B-Chat另外有提供4 bit 量化模型，量化模型主要是提供使用者的便利性，可能會影響效能與更多不可預期的問題，還請使用者理解與注意。 - ---- -### Gemma -**model:** [google/gemma-1.1-7b-it](https://huggingface.co/google/gemma-1.1-7b-it)
-**Blog:** [Gemma: Introducing new state-of-the-art open models](https://blog.google/technology/developers/gemma-open-models/)
-**Kaggle:** [https://www.kaggle.com/code/nilaychauhan/fine-tune-gemma-models-in-keras-using-lora](https://www.kaggle.com/code/nilaychauhan/fine-tune-gemma-models-in-keras-using-lora)
- ---- -### Claude 3 -[Introducing the next generation of Claude](https://www.anthropic.com/news/claude-3-family)
-![](https://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-cdn.anthropic.com%2Fimages%2F4zrzovbb%2Fwebsite%2F9ad98d612086fe52b3042f9183414669b4d2a3da-2200x1954.png&w=3840&q=75) - ---- -### InflectionAI -**Blog:** [Inflection AI 發表新基礎模型「Inflection-2.5 」，能力逼近 GPT-4！](https://www.inside.com.tw/article/34418-inflection-ai-launches-inflection-2-5)
- ---- -### Phind-70B -**Blog:** [Introducing Phind-70B – closing the code quality gap with GPT-4 Turbo while running 4x faster](https://www.phind.com/blog/introducing-phind-70b)
-**Blog:** [Phind - 給工程師的智慧搜尋引擎](https://ywctech.net/ml-ai/phind-first-impression)
-Phind-70B is significantly faster than GPT-4 Turbo, running at 80+ tokens per second to GPT-4 Turbo's ~20 tokens per second. We're able to achieve this by running NVIDIA's TensorRT-LLM library on H100 GPUs, and we're working on optimizations to further increase Phind-70B's inference speed. - ---- -### [Llama-3](https://ai.meta.com/blog/meta-llama-3/) -**model:** [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct)
-**Code:** [https://github.com/meta-llama/llama3/](https://github.com/meta-llama/llama3/)
-![](https://scontent.ftpe3-2.fna.fbcdn.net/v/t39.2365-6/438922663_1135166371264105_805978695964769385_n.png?_nc_cat=107&ccb=1-7&_nc_sid=e280be&_nc_ohc=VlupGTPFG1UAb6vzA2n&_nc_ht=scontent.ftpe3-2.fna&oh=00_AfCRhEYLtA4OYbvEyNCRXBdU9riWMtwtBWnk79O-SIigbg&oe=663C005E) - ---- -### Phi-3 -**model:** [microsoft/Phi-3-mini-4k-instruct"](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct)
-**Blog:** [Introducing Phi-3: Redefining what’s possible with SLMs](https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/)
- ---- -### Octopus v4 -**model:** [NexaAIDev/Octopus-v4](https://huggingface.co/NexaAIDev/Octopus-v4)
-**Paper:** [Octopus v4: Graph of language models](https://arxiv.org/abs/2404.19296)
-**Code:** [https://github.com/NexaAI/octopus-v4](https://github.com/NexaAI/octopus-v4)
-[design demo](https://graph.nexa4ai.com/)
- ---- -### [Llama 3.1](https://ai.meta.com/blog/meta-llama-3-1/) -**mode:** [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct)
-![](https://scontent.ftpe3-1.fna.fbcdn.net/v/t39.2365-6/452673884_1646111879501055_1352920258421649752_n.png?_nc_cat=100&ccb=1-7&_nc_sid=e280be&_nc_ohc=tHtG_dev9lcQ7kNvgFirAn9&_nc_ht=scontent.ftpe3-1.fna&oh=00_AYDvpGKrjgZlN5eVxo7ppCqw2Umq03ytaednlW3mTpCDvQ&oe=66C45828) - ---- -### Grok-2 -Grok-2 & Grok-2 mini, achieve performance levels competitive to other frontier models in areas such as graduate-level science knowledge (GPQA), general knowledge (MMLU, MMLU-Pro), and math competition problems (MATH). Additionally, Grok-2 excels in vision-based tasks, delivering state-of-the-art performance in visual math reasoning (MathVista) and in document-based question answering (DocVQA). -![](https://github.com/rkuo2000/AI-course/blob/main/images/Grok-2-benchmark.png?raw=true) - ---- -### Phi-3.5 -**model:** [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct)
-**model:** [microsoft/Phi-3.5-vision-instruct](https://huggingface.co/microsoft/Phi-3.5-vision-instruct)
-**model:** [microsoft/Phi-3.5-MoE-instruct](https://huggingface.co/microsoft/Phi-3.5-MoE-instruct)
-**News:** [Microsoft Unveils Phi-3.5: Powerful AI Models Punch Above Their Weight](https://www.maginative.com/article/microsoft-unveils-phi-3-5-powerful-ai-models-punch-above-their-weight/)
-![](https://www.maginative.com/content/images/size/w1600/2024/08/Screenshot-2024-08-21-at-2.11.41-AM.png) -![](https://www.maginative.com/content/images/size/w1600/2024/08/Screenshot-2024-08-21-at-2.09.00-AM.png) - ---- -## LLM running locally - -### [ollama](https://ollama.com/download) -`ollama -v`
-`ollama`
-`ollama pull llava`
-`ollama run llava`
- -**Code:** [Github](https://github.com/ollama/ollama)
-**[Examples](https://github.com/ollama/ollama/tree/main/examples)**:
-* [langchain-python-rag-privategpt](https://github.com/ollama/ollama/tree/main/examples/langchain-python-rag-privategpt)
- ---- -### [LM Studio](https://lmstudio.ai/) -![](https://github.com/rkuo2000/AI-course/blob/main/images/LM_studio_0.2.19.png?raw=true) - ---- -### PrivateGPT -**Code:** [https://github.com/zylon-ai/private-gpt](https://github.com/zylon-ai/private-gpt)
-![](https://github.com/zylon-ai/private-gpt/raw/main/fern/docs/assets/ui.png?raw=true) - ---- -### [GPT4All](https://gpt4all.io/index.html) -``` -chmod +x gpt4all-installer-linux.run -./gpt4all-installer-linux.run -cd ~/gpt4all -./bin/chat -``` -![](https://gpt4all.io/landing.gif) - ---- -### [GPT4FREE](https://github.com/xtekky/gpt4free) -`pip install g4f`
- ---- -## Deep LLM - -### Deep Language Networks -**Paper:** [Joint Prompt Optimization of Stacked LLMs using Variational Inference](https://arxiv.org/abs/2306.12509)
-**Code:** [https://github.com/microsoft/deep-language-networks](https://github.com/microsoft/deep-language-networks)
-![](https://github.com/rkuo2000/AI-course/blob/main/images/DLN-2_benchmark.png?raw=true) - ---- -### Constitutional AI -**Paper:** [Constitutional AI: Harmlessness from AI Feedback](https://arxiv.org/abs/2212.08073) -Two key phases:
-1. Supervised Learning Phase (SL Phase) -- Step1: The learning starts using the samples from the initial model -- Step2: From these samples, the model generates self-critiques and revisions -- Step3: Fine-tine the original model with these revisions -2. Reinforcement Learning Phase (RL Phease) -- Step1. The model uses samples from the fine-tuned model. -- Step2. Use a model to compare the outputs from samples from the initial model and the fine-tuned model -- Step3. Decide which sample is better. (RLHF) -- Step4. Train a new "preference model" from the new dataset of AI preferences. -This new "prefernece model" will then be used to re-train the RL (as a reward signal). -It is now the RLHAF (Reinforcement Learning from AI feedback) - ---- -### Chain-of-Thought Prompting -**Paper:** [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903)
-![](https://ar5iv.labs.arxiv.org/html/2201.11903/assets/x1.png) - ---- -### [ReAct Prompting](https://react-lm.github.io/) -**Paper:** [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629)
-**Code:** [https://github.com/ysymyth/ReAct](https://github.com/ysymyth/ReAct)
-![](https://react-lm.github.io/files/diagram.png) - ---- -### Tree-of-Thoughts -**Paper:** [Tree of Thoughts: Deliberate Problem Solving with Large Language Models](https://arxiv.org/abs/2305.10601)
-**Code:** [https://github.com/princeton-nlp/tree-of-thought-llm](https://github.com/princeton-nlp/tree-of-thought-llm)
-**Code:** [https://github.com/kyegomez/tree-of-thoughts](https://github.com/kyegomez/tree-of-thoughts)
-![](https://github.com/princeton-nlp/tree-of-thought-llm/blob/master/pics/teaser.png?raw=true) - ---- -### Tabular CoT -**Paper:** [Tab-CoT: Zero-shot Tabular Chain of Thought](https://arxiv.org/abs/2305.17812)
-**Code:** [https://github.com/Xalp/Tab-CoT](https://github.com/Xalp/Tab-CoT)
-![](https://github.com/Xalp/Tab-CoT/raw/main/intro.jpg) - ---- -### Survey of Chain-of-Thought -**Paper:** [A Survey of Chain of Thought Reasoning: Advances, Frontiers and Future](https://arxiv.org/abs/2309.15402)
- ---- -### Chain-of-Thought Hub -**Paper:** [Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance](https://arxiv.org/abs/2305.17306)
-**Code:** [https://github.com/FranxYao/chain-of-thought-hub](https://github.com/FranxYao/chain-of-thought-hub)
- ---- -### Everything-of-Thoughts -**Paper:** [Everything of Thoughts: Defying the Law of Penrose Triangle for Thought Generation](https://arxiv.org/abs/2311.04254)
-**Code:** [https://github.com/microsoft/Everything-of-Thoughts-XoT](https://github.com/microsoft/Everything-of-Thoughts-XoT)
-![](https://github.com/microsoft/Everything-of-Thoughts-XoT/raw/main/assets/compare.jpg) - ---- -### R3 -**Paper:** [Training Large Language Models for Reasoning through Reverse Curriculum Reinforcement Learning](https://arxiv.org/abs/2402.05808)
-**Code:** [https://github.com/WooooDyy/LLM-Reverse-Curriculum-RL](https://github.com/WooooDyy/LLM-Reverse-Curriculum-RL)
-![](https://github.com/WooooDyy/LLM-Reverse-Curriculum-RL/raw/master/src/figures/main.png) - ---- -## Time-series LLM -**Paper:** [Large Models for Time Series and Spatio-Temporal Data: A Survey and Outlook](https://arxiv.org/abs/2310.10196)
-**Papers: [https://github.com/DaoSword/Time-Series-Forecasting-and-Deep-Learning](https://github.com/DaoSword/Time-Series-Forecasting-and-Deep-Learning)
- -### Time-LLM -**Paper:** [Time-LLM: Time Series Forecasting by Reprogramming Large Language Models](https://arxiv.org/abs/2310.01728)
-**Code:** [https://github.com/KimMeen/Time-LLM](https://github.com/KimMeen/Time-LLM)
-![](https://github.com/KimMeen/Time-LLM/raw/main/figures/framework.png) -![](https://github.com/KimMeen/Time-LLM/raw/main/figures/method-detailed-illustration.png) - ---- -### Time-LLM -**Paper:** [Time-LLM: Time Series Forecasting by Reprogramming Large Language Models](https://arxiv.org/abs/2310.01728)
-**Code:** [https://github.com/KimMeen/Time-LLM](https://github.com/KimMeen/Time-LLM)
-![](https://github.com/KimMeen/Time-LLM/raw/main/figures/framework.png) -![](https://github.com/KimMeen/Time-LLM/raw/main/figures/method-detailed-illustration.png) - ---- -### TimeGPT-1 -**Paper:** [TimeGPT-1](https://arxiv.org/abs/2310.03589)
- - ---- -### TEMPO -**Paper:** [TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting](https://arxiv.org/abs/2310.04948)
-**Code:** [https://github.com/liaoyuhua/tempo-pytorch](https://github.com/liaoyuhua/tempo-pytorch)
-![](https://github.com/liaoyuhua/tempo-pytorch/raw/main/assets/struct.png) - ---- -### Lag-LLaMA -**Paper:** [Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecasting](https://arxiv.org/abs/2310.08278)
-**Blog:** [From RNN/LSTM to Temporal Fusion Transformers and Lag-Llama](https://dataman-ai.medium.com/from-rnn-lstm-to-temporal-fusion-transformers-and-lag-llama-6e6a62c811bd)
-**Code:** [https://github.com/time-series-foundation-models/lag-llama](https://github.com/time-series-foundation-models/lag-llama)
-**Colab:**
-* [Lag-Llama Fine-Tuning Demo](https://colab.research.google.com/drive/1uvTmh-pe1zO5TeaaRVDdoEWJ5dFDI-pA?usp=sharing)
-* [Lag-Llama Zero-shot Forecasting Demo](https://colab.research.google.com/drive/1XxrLW9VGPlZDw3efTvUi0hQimgJOwQG6?usp=sharing)
- ---- -### Timer -**Paper:** [Timer: Transformers for Time Series Analysis at Scale](https://arxiv.org/abs/2402.02368)
- ---- -## Applications - -### [FunSearch](https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/) -[DeepMind發展用LLM解困難數學問題的方法](https://www.ithome.com.tw/news/160354)
-![](https://s4.itho.me/sites/default/files/styles/picture_size_large/public/field/image/2108_-_funsearch_making_new_discoveries_in_mathematical_sciences_using_lar_-_deepmind.google.jpg?itok=mAy4ydAE) - ---- -### Automatic Evaluation -**Paper:** [Can Large Language Models Be an Alternative to Human Evaluation?](https://arxiv.org/abs/2305.01937)
-![](https://aisholar.s3.ap-northeast-1.amazonaws.com/posts/July2023/Can_Large_Language_Models_Be_an_Alternative_to_Human_Evaluation_fig1.png) - -**Paper:** [A Closer Look into Automatic Evaluation Using Large Language Models](https://arxiv.org/abs/2310.05657)
-**Code:** [https://github.com/d223302/A-Closer-Look-To-LLM-Evaluation](https://github.com/d223302/A-Closer-Look-To-LLM-Evaluation)
- ---- -### BrainGPT -**Paper:** [DeWave: Discrete EEG Waves Encoding for Brain Dynamics to Text Translation](https://arxiv.org/abs/2309.14030)
-**Blog:** [New Mind-Reading "BrainGPT" Turns Thoughts Into Text On Screen](https://www.iflscience.com/new-mind-reading-braingpt-turns-thoughts-into-text-on-screen-72054)
-![](https://i3.res.bangqu.com/farm/liang/news/2023/12/18/339b9a2158e1fd28e1e39ee4b1557df2.jpg) -![](https://i3.res.bangqu.com/farm/liang/news/2023/12/18/79ca704627e4cadc1e23afc1b2f029cb.jpg) - ---- -### Designing Silicon Brains using LLM -**Paper:** [Designing Silicon Brains using LLM: Leveraging ChatGPT for Automated Description of a Spiking Neuron Array -](https://arxiv.org/abs/2402.10920)
-![](https://github.com/rkuo2000/AI-course/blob/main/images/ChatGPT_design_Spiking_Neuron_Array.png?raw=true) - ---- -### MyGirlGPT -**Code:** [https://github.com/Synthintel0/MyGirlGPT](https://github.com/Synthintel0/MyGirlGPT)
- - ---- -### Robotic Manipulation -**Paper:** [Language-conditioned Learning for Robotic Manipulation: A Survey](https://arxiv.org/abs/2312.10807)
-**Paper:** [Human Demonstrations are Generalizable Knowledge for Robots](https://arxiv.org/abs/2312.02419)
- ---- -### [ALTER-LLM](https://tnoinkwms.github.io/ALTER-LLM/) -**Paper:** [From Text to Motion: Grounding GPT-4 in a Humanoid Robot "Alter3"](https://arxiv.org/abs/2312.06571)
- -![](https://tnoinkwms.github.io/ALTER-LLM/architecture_2.png) -![](https://tnoinkwms.github.io/ALTER-LLM/feedback.png) - -
-
- -*This site was last updated {{ site.time | date: "%B %d, %Y" }}.* -