OpenChat
@OpenChatDev
Followers
1,970
Following
42
Media
10
Statuses
57
Advancing Open Source LLMs with Mixed Quality Data through offline RL-inspired C-RLFT. ⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗟𝗲𝗮𝗱: Guan Wang, @AlpayAriyak
Joined July 2023
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
ドラフト
• 398173 Tweets
#わたしの宝物
• 73115 Tweets
NuNew Medley Song
• 70918 Tweets
青春18きっぷ
• 51033 Tweets
Baakhabar Sant Rampal Ji
• 49987 Tweets
LINEスタンプ
• 28200 Tweets
ILL BE THERE TODAY
• 26895 Tweets
Errejón
• 26334 Tweets
船井電機
• 24689 Tweets
LIA AT DG BEAUTY POP UP
• 19645 Tweets
WALL SONG OHMLENG
• 18486 Tweets
まっすー
• 15059 Tweets
Razem
• 13738 Tweets
FURIA EN BONDI
• 12954 Tweets
もやしもん
• 12097 Tweets
Adoちゃん
• 11572 Tweets
GENTO SA TAIWAN
• 11216 Tweets
Pinned Tweet
🚀Introducing OpenChat 3.6
🌟Surpassed official Llama3-Instruct—with 1-2M synthetic data compared to ~10M human labels
🤫GPTs are close to limits—excel at generation but fall short at complex tasks
🎯We are training next gen—capable of deterministic reasoning and planning
🔗
9
68
298
Introducing the 𝗪𝗼𝗿𝗹𝗱’𝘀 𝗕𝗲𝘀𝘁 𝗢𝗽𝗲𝗻 𝗦𝗼𝘂𝗿𝗰𝗲 𝟳𝗕 𝗟𝗟𝗠 - OpenChat-3.5-1210, further surpassing ChatGPT and Grok models.
This upgrade to the widely adopted OpenChat-3.5 is focused on increasing the performance in one of the most important areas for LLMs -
25
145
803
🚀Announcing OpenChat-3.5 Update 0106: 𝗪𝗼𝗿𝗹𝗱’𝘀 𝗕𝗲𝘀𝘁 𝗢𝗽𝗲𝗻 𝗦𝗼𝘂𝗿𝗰𝗲 𝟳𝗕 𝗟𝗟𝗠!
Experience ChatGPT & Grok-level AI locally 💿!
Surpassing Grok-0 (33B) across all 4 benchmarks and Grok-1 (???B) on average and 3/4 benchmarks 🔥.
🎯 This update mainly enhanced
32
132
641
🚀 The World's First Gemma fine-tune based on openchat-3.5-0106 data and method (C-RLFT). Almost the same performance as the Mistral-based version.
6T tokens = secret recipe?
HuggingFace:
11
31
184
🎉OpenChat 3.2 SUPER is here!
🚀 Built with innovative fine-tuning techniques, it outperforms all Llama-2-based 13B models, even with the same 80K mixed-quality ShareGPT data set.
🥇 Ranking
#1
on AgentBench, MT-bench, and AlpacaEval among 13B models.
5
9
72
We achieved almost ideal MoE fine-tuning performance (equivalent to a dense model with the same active parameters). <10% overhead
3
1
66
Additional Benchmarks, including the scores for ChatGPT, Grok, and Open Source LLMs.
4
6
61
🚀Kudos to
@huggingface
! OpenChat-3.5 Update 0106 has landed on HuggingChat & Spaces! Explore now!
Experience open-source AI at ChatGPT & Grok level!
🤗 HuggingChat:
🌌 Spaces:
🖥️ OpenChat UI:
🚀Announcing OpenChat-3.5 Update 0106: 𝗪𝗼𝗿𝗹𝗱’𝘀 𝗕𝗲𝘀𝘁 𝗢𝗽𝗲𝗻 𝗦𝗼𝘂𝗿𝗰𝗲 𝟳𝗕 𝗟𝗟𝗠!
Experience ChatGPT & Grok-level AI locally 💿!
Surpassing Grok-0 (33B) across all 4 benchmarks and Grok-1 (???B) on average and 3/4 benchmarks 🔥.
🎯 This update mainly enhanced
32
132
641
2
11
60
🚀 OpenChat: our new paper on enhancing open-source language models!
C-RLFT utilizes mixed-quality data—no preference labels!
OpenChat-13B excels, using only ShareGPT data (like Vicuna). Discover more!👇
#NLP
#AI
#OpenSource
5
13
46
It is also available on our hosted demo:
If you’d like to deploy it yourself, you can follow the instructions on our GitHub to serve OpenChat models with a vLLM backend, API keys and more:
6
0
30
(1/3) Additional benchmark results, including ChatGPT and other open-source models
3
0
23
@NSarrazin_
Great job😀Tried HuggingChat with the web search functionality, and it's working very well!
0
0
11
I wonder if the remaining OpenAI people will open-source GPT4.
I just want **Open**AI to be back.
2
0
11
Magic recipe:
Set eps = 1e-5 in AdamW and you will get a very smooth loss curve
Tested with betas (0.9, 0.95) weight decay 0.1 and a lot of different learning rates
.
0
1
7
@_philschmid
Thank you! We added this feature to advocate reproducible evaluations with open-source LLMs. It should behave similarly to Prometheus. We're testing using the methodology in their paper😀
1
0
7
@andersonbcdefg
Conditioning: Use a different prompt for GPT4 and GPT3.5 data
Token-wise Loss: The same as HF loss calculation, the total loss is the average of all token losses
0
0
2
@yar_vol
@alignment_lab
@OpenAI
@Tim_Dettmers
The model in the paper is based on Llama 2. However, openchat 3.5 is based on Mistral 7B, so it is 1/3 the size.
1
0
3
@_philschmid
@Teknium1
@Mascobot
Just replaced the Gemma's tokenizer with the instruction-tuned version's. Feel free to use it
0
0
3
@alignment_lab
@_philschmid
@lvwerra
@Teknium1
@erhartford
OpenChat masks all user input following the instructgpt paper. We plan to test it soon.
0
0
3
@yar_vol
@alignment_lab
@OpenAI
@Tim_Dettmers
Yes, exactly. We experimented with mistral 7b and llama 2 13b. Mistral is definitely better.
1
0
3
@abacaj
Today's OSS models are much better than before. Our OrcaPlaty 13B has nearly the same benchmark results as the Gopher 280B
0
0
1
@aisinne
Me, gpt4 and openchat have the same answer 🤣
"The banana is now in the living room, on the plate you took with you."
0
0
1