should have used a local phi3 mini
June 24, 2024 at 3:26 PM
should have used a local phi3 mini
If this approach makes Phi3-mini-3.8B better at math than o1-preview, what happens when the same thing is done with an o1/o3-class model?
January 11, 2025 at 2:00 PM
If this approach makes Phi3-mini-3.8B better at math than o1-preview, what happens when the same thing is done with an o1/o3-class model?
Microsoft releases rStar-Math: "Small LLMs can master math reasoning with self-evolved deep thinking"
It improves Qwen2.5-Math-7B from 58.8% to 90.0% and Phi3-mini-3.8B from 41.4% to 86.4%, surpassing o1-preview by +4.5% and +0.9%
Source : arxiv.org/html/2501.04...
It improves Qwen2.5-Math-7B from 58.8% to 90.0% and Phi3-mini-3.8B from 41.4% to 86.4%, surpassing o1-preview by +4.5% and +0.9%
Source : arxiv.org/html/2501.04...
January 9, 2025 at 9:19 AM
Microsoft releases rStar-Math: "Small LLMs can master math reasoning with self-evolved deep thinking"
It improves Qwen2.5-Math-7B from 58.8% to 90.0% and Phi3-mini-3.8B from 41.4% to 86.4%, surpassing o1-preview by +4.5% and +0.9%
Source : arxiv.org/html/2501.04...
It improves Qwen2.5-Math-7B from 58.8% to 90.0% and Phi3-mini-3.8B from 41.4% to 86.4%, surpassing o1-preview by +4.5% and +0.9%
Source : arxiv.org/html/2501.04...
interesting news from Docugami, who has always told me they can get LLM-level results from small models by being very focused on their domain (business documents); they're using Phi3-mini and Ministral-8B for mathematical reasoning (which documents like contracts need)
Docugami to Showcase Small Agentic Reasoning Models that Outperform Chat GPT-grade LLMs at Prestigious ICDAR Conference
Docugami Science Team research on Agentic Reasoning; Docugami's small agentic models outperform much larger LLMs, including ChatGPT and GPT-4.
www.docugami.com
September 16, 2025 at 5:55 PM
interesting news from Docugami, who has always told me they can get LLM-level results from small models by being very focused on their domain (business documents); they're using Phi3-mini and Ministral-8B for mathematical reasoning (which documents like contracts need)
I just ran an AI model #phi3:mini on my home TrunNAS server.
October 27, 2025 at 12:31 AM
I just ran an AI model #phi3:mini on my home TrunNAS server.
Local AI is indeed the way forward. I'm researching small scale LLMs like phi3-mini, Llama 3.1, etc. which can work in constrained environments like my laptop.
October 22, 2025 at 2:15 AM
Local AI is indeed the way forward. I'm researching small scale LLMs like phi3-mini, Llama 3.1, etc. which can work in constrained environments like my laptop.
In case you are wondering, the new Microsoft mini LLM - phi3, can handle code generation, in this case, SQL.
I compared the runtime (locally on CPU) with respect to codellama:7B using Ollama, and surprisingly the Phi3 runtime was significantly slower.
#phi3 #llm #python #ollama #sql
I compared the runtime (locally on CPU) with respect to codellama:7B using Ollama, and surprisingly the Phi3 runtime was significantly slower.
#phi3 #llm #python #ollama #sql
April 25, 2024 at 12:39 PM
On an intel Mac?
You are going to struggle without a external graphics card and a lot of RAM.
RAM being the biggest constraint, ultimately CPU/integrated graphics will be slow, without the RAM it will be unusuable.
Maybe try ollama.com/library/phi3 mini which should work.
You are going to struggle without a external graphics card and a lot of RAM.
RAM being the biggest constraint, ultimately CPU/integrated graphics will be slow, without the RAM it will be unusuable.
Maybe try ollama.com/library/phi3 mini which should work.
phi3
Phi-3 is a family of lightweight 3B (Mini) and 14B (Medium) state-of-the-art open models by Microsoft.
ollama.com
January 29, 2025 at 2:35 PM
On an intel Mac?
You are going to struggle without a external graphics card and a lot of RAM.
RAM being the biggest constraint, ultimately CPU/integrated graphics will be slow, without the RAM it will be unusuable.
Maybe try ollama.com/library/phi3 mini which should work.
You are going to struggle without a external graphics card and a lot of RAM.
RAM being the biggest constraint, ultimately CPU/integrated graphics will be slow, without the RAM it will be unusuable.
Maybe try ollama.com/library/phi3 mini which should work.
There are many advantages. I like to try tiny models like Phi3-mini or Gemma3 4B and to enhance their performance. You can develop simple agents and try to solve basic conversational tasks with them, like orchestration. Some of them even capable of tool calling, but it's not stable yet
February 16, 2026 at 3:16 PM
There are many advantages. I like to try tiny models like Phi3-mini or Gemma3 4B and to enhance their performance. You can develop simple agents and try to solve basic conversational tasks with them, like orchestration. Some of them even capable of tool calling, but it's not stable yet
UbuntuにOllamaいれた。モデルはPhi3:mini。モンハンで遊びたいしGPU買おうかと思っているけどLLMとかML周りへの流用もとっても楽しみ。多分1月になったら買う。
November 23, 2024 at 2:15 PM
UbuntuにOllamaいれた。モデルはPhi3:mini。モンハンで遊びたいしGPU買おうかと思っているけどLLMとかML周りへの流用もとっても楽しみ。多分1月になったら買う。
I've been investigating whether small #ai models like #llama3 and #phi3 generalise well to bio-medical Q&A use cases. Small models that don't require data centres full of GPUs to run are starting to become competitive with big commercial LLMs. brainsteam.co.uk/2024/04/26/c...
Can Phi3 and Llama3 Do Biology? - Brainsteam
investigating whether llama 3 8b and phi 3 mini can do biomedical question answering based on a lesser known task and benchmark
brainsteam.co.uk
April 26, 2024 at 12:53 PM
I've been investigating whether small #ai models like #llama3 and #phi3 generalise well to bio-medical Q&A use cases. Small models that don't require data centres full of GPUs to run are starting to become competitive with big commercial LLMs. brainsteam.co.uk/2024/04/26/c...
Getting started with Microsoft Phi-3-mini - Try running the Phi-3-mini on iPhone with ONNX Runtime.
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel
techcommunity.microsoft.com/t5/microsoft...
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel
techcommunity.microsoft.com/t5/microsoft...
Getting started with Microsoft Phi-3-mini - Try running the Phi-3-mini on iPhone with ONNX Runtime
In this article, we explore how to deploy generative AI applications to mobile devices, specifically on iPhone, using ONNX Runtime. We cover the steps to compile ONNX Runtime for iOS and then create a...
techcommunity.microsoft.com
May 6, 2024 at 12:03 PM
Getting started with Microsoft Phi-3-mini - Try running the Phi-3-mini on iPhone with ONNX Runtime.
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel
techcommunity.microsoft.com/t5/microsoft...
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel
techcommunity.microsoft.com/t5/microsoft...
Results of rStar-Math:
• After four rounds, 90% of the 747k problems were successfully covered with high-quality reasoning steps.
• It raised the accuracy of SLMs:
- Qwen2.5-Math-7B from 58.8% to 90.0%
- Phi3-mini-3.8B from 41.4% to 86.4%
• It solves 8 out of 15 problems from Math Olympiad.
• After four rounds, 90% of the 747k problems were successfully covered with high-quality reasoning steps.
• It raised the accuracy of SLMs:
- Qwen2.5-Math-7B from 58.8% to 90.0%
- Phi3-mini-3.8B from 41.4% to 86.4%
• It solves 8 out of 15 problems from Math Olympiad.
January 11, 2025 at 12:11 AM
Results of rStar-Math:
• After four rounds, 90% of the 747k problems were successfully covered with high-quality reasoning steps.
• It raised the accuracy of SLMs:
- Qwen2.5-Math-7B from 58.8% to 90.0%
- Phi3-mini-3.8B from 41.4% to 86.4%
• It solves 8 out of 15 problems from Math Olympiad.
• After four rounds, 90% of the 747k problems were successfully covered with high-quality reasoning steps.
• It raised the accuracy of SLMs:
- Qwen2.5-Math-7B from 58.8% to 90.0%
- Phi3-mini-3.8B from 41.4% to 86.4%
• It solves 8 out of 15 problems from Math Olympiad.
🚀 ¡Phi3 Mini: La IA en Dispositivos Móviles! 🚀
Es un modelo de IA compacto desarrollado por Microsoft.
Cuenta con variaciones con 128K, 3.8B, 7B y 13B de parámetros.
supera a modelos más grandes de competidores como Llama3 y GPT-3.
Funciona en dispositivos móviles, incluyendo el iPhone 15
Es un modelo de IA compacto desarrollado por Microsoft.
Cuenta con variaciones con 128K, 3.8B, 7B y 13B de parámetros.
supera a modelos más grandes de competidores como Llama3 y GPT-3.
Funciona en dispositivos móviles, incluyendo el iPhone 15
April 25, 2024 at 1:17 PM
🚀 ¡Phi3 Mini: La IA en Dispositivos Móviles! 🚀
Es un modelo de IA compacto desarrollado por Microsoft.
Cuenta con variaciones con 128K, 3.8B, 7B y 13B de parámetros.
supera a modelos más grandes de competidores como Llama3 y GPT-3.
Funciona en dispositivos móviles, incluyendo el iPhone 15
Es un modelo de IA compacto desarrollado por Microsoft.
Cuenta con variaciones con 128K, 3.8B, 7B y 13B de parámetros.
supera a modelos más grandes de competidores como Llama3 y GPT-3.
Funciona en dispositivos móviles, incluyendo el iPhone 15
Idk who thought of the genius idea for recall but it's very funny how this didn't come up for them.
Also, their point would be to run LLMs locally like Phi3-Small and Mini but they probably think it's immature still due to bad NPU sdks on most if not all NPU out there.
Also, their point would be to run LLMs locally like Phi3-Small and Mini but they probably think it's immature still due to bad NPU sdks on most if not all NPU out there.
November 14, 2024 at 11:51 PM
Idk who thought of the genius idea for recall but it's very funny how this didn't come up for them.
Also, their point would be to run LLMs locally like Phi3-Small and Mini but they probably think it's immature still due to bad NPU sdks on most if not all NPU out there.
Also, their point would be to run LLMs locally like Phi3-Small and Mini but they probably think it's immature still due to bad NPU sdks on most if not all NPU out there.
For reference, the local lineup here: llama3.2:1b, gemma2:2b, phi3:mini and gemma3:4b for fast agentic steps, qwen3.5:9b/27b when you want heavier reasoning, plus nomic-embed-text for RAG. The gateway routes per task, so you mix small+fast with big+smart.
July 23, 2026 at 8:01 AM
For reference, the local lineup here: llama3.2:1b, gemma2:2b, phi3:mini and gemma3:4b for fast agentic steps, qwen3.5:9b/27b when you want heavier reasoning, plus nomic-embed-text for RAG. The gateway routes per task, so you mix small+fast with big+smart.
Experimente mit ein paar Modellen als Helfer im privaten Netzwerk — später via VS Code.
Meine Favorites:
gemma3:4b a2af6cc3eb7f 3.3 GB 3 minutes ago
qwen2.5-coder:1.5b
qwen2.5:1.5b
phi:latest
mistral-openorca:7b-q4_0
codellama:latest
phi3:mini
mistral-openorca:7b-q8_0
Meine Favorites:
gemma3:4b a2af6cc3eb7f 3.3 GB 3 minutes ago
qwen2.5-coder:1.5b
qwen2.5:1.5b
phi:latest
mistral-openorca:7b-q4_0
codellama:latest
phi3:mini
mistral-openorca:7b-q8_0
July 21, 2025 at 1:04 PM
Experimente mit ein paar Modellen als Helfer im privaten Netzwerk — später via VS Code.
Meine Favorites:
gemma3:4b a2af6cc3eb7f 3.3 GB 3 minutes ago
qwen2.5-coder:1.5b
qwen2.5:1.5b
phi:latest
mistral-openorca:7b-q4_0
codellama:latest
phi3:mini
mistral-openorca:7b-q8_0
Meine Favorites:
gemma3:4b a2af6cc3eb7f 3.3 GB 3 minutes ago
qwen2.5-coder:1.5b
qwen2.5:1.5b
phi:latest
mistral-openorca:7b-q4_0
codellama:latest
phi3:mini
mistral-openorca:7b-q8_0
AI Services in LangChain4j
www.logicbig.com/tutorials/ai...
www.logicbig.com/tutorials/ai...
AI LangChain4j - Simplest AI Service Example with Ollama and Phi3 Mini
Learn how to create a basic AI Service using LangChain4j with Ollama and Phi3 model.
www.logicbig.com
January 29, 2026 at 2:58 PM
AI Services in LangChain4j
www.logicbig.com/tutorials/ai...
www.logicbig.com/tutorials/ai...
饒舌すぎるけど、なんとなくphi4とphi4-miniが自分の書くプロンプトとは相性が良さそうな気がする
饒舌もphi3とかに比べるとマシな気がするしｗ
饒舌もphi3とかに比べるとマシな気がするしｗ
March 13, 2025 at 7:00 AM
饒舌すぎるけど、なんとなくphi4とphi4-miniが自分の書くプロンプトとは相性が良さそうな気がする
饒舌もphi3とかに比べるとマシな気がするしｗ
饒舌もphi3とかに比べるとマシな気がするしｗ
Need emojis? Microsoft Phi3-mini got you covered. If that's not worth the double size compared to qwen2:1.5b
April 3, 2025 at 9:21 PM
Need emojis? Microsoft Phi3-mini got you covered. If that's not worth the double size compared to qwen2:1.5b
Getting Started - Generative AI with Phi-3-mini: A Guide to Inference and Deployment.
#ai #GenerativeAI #Phi3ef="/hashtag/Phi3Mini" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link">#Phi3Mini #Phi3 #cloud #semantickernel #huggingface #onnx...
#ai #GenerativeAI #Phi3ef="/hashtag/Phi3Mini" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link">#Phi3Mini #Phi3 #cloud #semantickernel #huggingface #onnx...
November 19, 2024 at 12:47 AM
Getting Started - Generative AI with Phi-3-mini: A Guide to Inference and Deployment.
#ai #GenerativeAI #Phi3ef="/hashtag/Phi3Mini" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link">#Phi3Mini #Phi3 #cloud #semantickernel #huggingface #onnx...
#ai #GenerativeAI #Phi3ef="/hashtag/Phi3Mini" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link">#Phi3Mini #Phi3 #cloud #semantickernel #huggingface #onnx...
Getting started with Microsoft Phi-3-mini - Try running the Phi-3-mini on iPhone with ONNX Runtime.
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel...
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel...
November 19, 2024 at 12:27 AM
Getting started with Microsoft Phi-3-mini - Try running the Phi-3-mini on iPhone with ONNX Runtime.
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel...
#ai #onnx #phi3 #iOS #mobiledev #microsoft #aimodel...
Phi-3-vision in 50 lines of C# with ONNX Runtime GenAI.
#ai #phi3 #slm #aimodels #csharp #dotnet #onnx
https://nietras.com/2024/06/05/phi-3-vision-csharp-ortgenai/
#ai #phi3 #slm #aimodels #csharp #dotnet #onnx
https://nietras.com/2024/06/05/phi-3-vision-csharp-ortgenai/
Phi-3-vision in 50 lines of C# with ONNX Runtime GenAI
Previously in Phi-3-mini in 30 lines of C# with ONNX Runt...
nietras.com
November 18, 2024 at 10:36 PM