Top Tweets for #Codestral
Codex - Mistral - Codestral で簡易ゲーム生成の仕組みを構築中です。(VSCodeとLMStudioを使用)
画像は記念すべき「とりあえずスタートエンドがあるマイニングゲーム」です🍝
#おはようvtuber #AI活用 #codex #Mistral #Codestral
#VSCode #LMStudio
🚀 Just released TabCoder — an open-source, lightweight #vscode extension for pure #AI -powered autocomplete.
No chat. No agents. Just fast, context-aware suggestions you accept with Tab.
https://t.co/uXfG8ZnY3C
Multiples providers supported #ollama #openrouter #codestral
🔥 Mistral刚发布了Codestral 25.08，这次真的很厉害！
代码完成准确率提升30%，错误减少一半 ✨ 最重要的是可以完全本地部署，不用担心代码泄露了 🔒
包含完整的AI编程套件：智能补全、语义搜索、自动写PR...从VS Code到JetBrains都支持 💻
#Codestral #AI编程RetryClaude
Développer avec DeepSeek-R1 et Codestral via Roo Code (ex-RooCline, fork... https://t.co/N4zzqV4EJG via @YouTube
#DeepSeekR1 #roocode #roocline #cline #boltnew #boltdiy #deepseekv3 #codestral #mistralai #ai #ia #openai #claude #NoCode
今月3本目！< 自動プログラミングでquick sort、LP、ブロック崩し実行。全てローカルLLMで可能♪
#RooCline
#Codestral 22B #gpt4o #DeepSeekR1 14B
https://t.co/wAFD2QEfUA
🚀 #Mistral presenta #Codestral 2501, la actualización de su modelo de generación de código.
⚡ Rápido, ligero y eficiente: diseñado para generar y completar código 2X más rápido.
✅ Compatible con 80+ lenguajes
🤖 Optimizado para tareas como autocompletado y tests unitarios
API キーのエラーは、Mistral AI サイトで取得したキーが違っていたからだった。紛らわしいけど、左側のサイドナビゲーションの「API Keys」ではなく、「Codestral」内でキーを取得できる。 #Codestral
🚨 Tired of getting outdated code from ChatGPT?
Meet our #Codestral-powered #LLM, which deliver high-quality code using our @reservoirpy library 🔥
💻 Try it free—only a few days left! 👇
https://t.co/YRxKTfHiHf
@Inria_Bordeaux @Inria @labriOfficial @Neuro_Bordeaux
@MistralAI
🚀 Proud to share our new beta tool developed at @Inria! 🛠️ Dive into Reservoir Computing and coding with @reservoirpy, our Large Language Model is ready to chat! 💬💡
https://t.co/YRxKTfHQwN
#LLM #RAG #reservoir_computing
@Inria_Bordeaux @Neuro_Bordeaux @labriOfficial
📣#MistralAI の Codestral、Nemo、Large2 を #VertexAI でサポートを開始。#Codestral はコード補完、ドキュメント化、テスト生成などのタスク用に最適化されています。新しいモデルには MaaS として手軽にアクセスでき、手間をかけずにセットアップと管理ができます。https://t.co/ldWLnhY75N #gcpja
Google Cloud is integrating Mistral Al's code generation model, Codestral, into its Vertex Al platform, making it the first major cloud provider to offer this model to its customers.
#AI #GoogleCloud #MistralAl #Codestral #VertexAl #CodeGeneration #sayvai
#Mistral publie #Codestral Mamba~? un modèle de langage #Mamba2 avec 7 milliards de paramètres spécialisé dans la génération de code, disponible sous licence Apache 2.0 https://t.co/AX3pix9Ll2
#Codestral by @MistralAILabs + CodeGPT = Coding Superpowers!
Codestral in @codegptAI: Use Codestral in for code generation and tab completion!
Check out our docs for more Codestral integrations:
https://t.co/Y151kN6ipz
人工智能开发商 #Mistral 开源编程模型 #Codestral Mamba，支持无限长度的输入、支持最高 256K 上下文检索。
该模型基于 Mamba 而非 Transformer，能够快速响应和不受输入长度限制，更适合在编程开发领域使用。
查看全文：https://t.co/ZH7sGqv3By
🚀Local models can outcompete Gemini or OpenAI to build AI Web Agents!
🤗We have released a blog post on @huggingface that uses our open-source Web Agent evaluating tooling to show how @MistralAI #Codestral + bge-small can have the same performance as @OpenAI but with faster inference and lower + controlled cost!
Article: https://t.co/7SD7rxH6cx
Colab to reproduce evaluation results: https://t.co/GZKDlzxjoL
🛠️In an age where #Agents are likely to become a reality but platforms like @Apple only allow developers to provide Tools but not control the full agent stack, it’s refreshing to see how we can create our own Web Agents that outperform centralized solutions!
🌊For context, LaVague (https://t.co/f5p3flMSRe) is an open-source framework for building AI Web Agents. We can answer user queries, like ‘Click on the search bar and type X,’ by performing RAG on the current HTML and generating #Selenium code to perform the desired actions.
To evaluate the impact of different design choices, e.g. embedding models for the retriever or LLM for the code generation, we have open-sourced both datasets and tooling to evaluate the performance and the latency of those solutions.
We found that:
1. Local embedding models are as performant for recall of relevant HTML chunks but also cheaper / faster!
2. Codestral can be as performant as Gemini 1.5 Flash!
More details about the dataset and methodology can be found in our article.
If you want to understand how we designed our RAG pipeline on the HTML using Llama Index, you can also access our webinar on the topic: https://t.co/iUiNcFJrMa
https://t.co/7SD7rxH6cx
🧑💻 @MistralAI has unveiled #Codestral, their first programming-focused AI model 🧑💻
Have you tried it yet? Share your experience in the comments!
#Coding 🧵
🤺@MistralAI 's #Codestral beats #GPT-3.5 and is on par with @Google's Gemini when used as a Large Action Model for #Selenium code generation for Web Agents!
A key challenge when developing an LLM-based framework is finding the right models that answer our community requirements in terms of performance, cost, latency, and control.
We have therefore evaluated several models, from open source models like #Llama3 8b using @huggingface transformers, to proprietary APIs like GPT-4o, through Mistral’s Codestral. We focused on the task of Selenium code generation, which is a critical part of our workflow for our Web Agents to navigate and perform actions on the web.
Our findings:
🥇GPT-4o still dominates the podium
🥈Codestral is on par with Gemini-1.5 Flash!
🥉Codestral outperforms GPT-3.5 and Llama 3 70b, which are roughly the same
We plotted on this graph on the x-axis the size of the models when those are known (aka for open source ones), and on y-axis the performance of the model (more on that below).
The more a model is on the top left corner, the better it is in performance/cost.
Because we don’t know the number of parameters of proprietary APIs like Gemini and OpenAI, we could not do a real one to one comparison.
💡Interesting phenomena: there seems to be a class of models where there is a certain ratio of size to performance (we could plot a line between Llama 3 8b to Llama 3 70b).
However, Codestral seems to be on a class of its own: it’s outperforming Llama 3 70b with much less parameters!
🔥Fascinating how the open-source community is able to create models on par with closed AI APIs
To evaluate them, we used WebLINX (https://t.co/rlxWT0t4sw), and the Wave (https://t.co/c2CQLh0iZA), our open-source data of user queries from our anonymous telemetry.
To do evaluation, we injected unique IDs for each element in the HTML and looked at the overlap between the IDs of the element identified by the LLM, with the IDs of the ground truth element to be interacted with, which gives the Action recall, aka does the button that will be interacted with through the generated code contain the right elements.
Code will be made available soon for people to reproduce these results, and improve them, but also use our tooling to evaluate the performance of different models on their own agents!
Also if you want to learn more about Large Action Models, don’t hesitate to check our webinar on how to design and improve Large Action Models using LLMs on the 13th June at 9 am PST!
Webinar: https://t.co/4QORaf78gt
You can also join our Discord (https://t.co/JWtWTm5tar to chat with us, to ask questions or contribute to our open source project (https://t.co/OqrfjEVUre)
Si vous bidouillez un peu de code à l'aide de #Chatgpt ou #Claude3 , vérifiez votre taf avec Mistral #Codestral .
Pour des non-dév, ça corrige un tas de petits détails qu'on laisserait passer (syntaxe, plus concis, etc.)
💻@MistralAI dévoile #Codestral, un #LLM pour générer du code efficacement. Performant en #Python, il offre un gain de temps ⏱️ et une réduction des erreurs pour les #Développeurs. 👨💻
👇https://t.co/cqgnXbCs0k
I hung out with @sqs and the @sourcegraph team at their SG HackerHouse last week 🔥
We tried out @janframework's new OpenAI API compatible local AI server #Cortex to make Cody work offline 🤯
Next up: pull in @MistralAI's #Codestral for absolute offline coding bliss 😌
#MistralAI dévoile #Codestral, un #LLM adapté aux tâches de génération de code
👉Des performances supérieures aux modèles de Meta
https://t.co/J3cdEbPwEf
Last Seen Hashtags on Sotwe
Most Popular Users
Elon Musk 
@elonmusk
240.2M followers
Barack Obama
@barackobama
119.3M followers
Donald J. Trump
@realdonaldtrump
111.6M followers
Cristiano Ronaldo
@cristiano
108.9M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
97.3M followers
NASA
@nasa
92.1M followers
Justin Bieber
@justinbieber
90.6M followers
KATY PERRY
@katyperry
86.8M followers
Taylor Swift
@taylorswift13
80.6M followers
Lady Gaga
@ladygaga
72.2M followers
Kim Kardashian
@kimkardashian
69.4M followers
YouTube
@youtube
68.6M followers
Virat Kohli
@imvkohli
68.5M followers
Bill Gates
@billgates
63.4M followers
The Ellen Show
@theellenshow
62.5M followers
CNN
@cnn
61.9M followers
Neymar Jr
@neymarjr
61.1M followers
X
@x
60.9M followers
Selena Gomez
@selenagomez
59.9M followers