BigQwen2.5-125B I made a few self-merges with the new Qwen2.5 models: – The 125B model is based on Qwen2.5-72B-Instruct and follows Meta-Llama-3-120B-Instruct's (
https://
huggingface.co/mlabonne/Meta-
Llama-3-120B-Instruct
…) recipe, where blocks of 10 layers are repeated 6 times. – The 47B model is based on
LLMS
-
BigQwen2.5-125B: Self-Merged Large Language Model
By
–
-
SolarPro: Most Powerful LLM on Single GPU
By
–
@upstageai #SolarPro has been featured in the most popular tldr newsletter among developers and startups (600,000 viewers). The article is titled "Most powerful LLM on a single GPU (12 minute read)". Please give it a read! https://
a.tldrnewsletter.com/web-version?ep
=1&lc=3bf8d792-bd90-11ee-817b-b777f755fa8b&p=703435f8-799a-11ef-8b11-371f32244928&pt=campaign&t=1727097241&s=dcda0944a0dad5602d8b27308915cf426959da99e53da7f4f12b2e6ffb0fd41f
… -
New Gemini models expected to launch tomorrow
By
–
Looks like new Gemini models are dropping tomorrow and not Opus 3.5
-
Perplexity iOS UI update improves prompt visibility and reasoning steps
By
–
Perplexity for iOS got a new prompt screen where prompt area got moved to the top So after submitting, you will see reasoning steps appearing below (see before and after). I rly like this move The big thing I am still missing is an edit btn for submitted prompts
-
User experience with AI memory and personalization in voice mode
By
–
Personalisation isn't a new feature, but immediately after asking "How is it going?", I got a question back about one of the events I've saved into memory in the past In voice mode, it feels very different (even if it is just standard).
-
Testing new voice features in ChatGPT read aloud
By
–
The Voice selector is still very raw on desktop and these extra voices are not yet available. I've added some extra values for demo proposes. More info https://
testingcatalog.com/chatgpt-hints-
at-up-to-8-new-voices-available-for-read-aloud/
… -
Turbo LoRA: Fast Inference with Customized Models
By
–
Last chance to save your spot Learn how to get the best of both worlds: super fast #inference + high quality models customized for your use case! Join us tomorrow, to learn about Turbo LoRA, a new approach to fine-tuning that increases model #throughput by 2-3x, lowers
-
Data Wall Becomes AI’s Next Frontier Bottleneck
By
–
Now that the “data wall” is the bottleneck for AI advancement, we plan to scale that wall with a combination of data abundance, frontier data for frontier models, and the most advanced evals for measuring model improvement.
-
SambaNova Cloud Offers Free Access to Meta’s Llama 3.1 405B
By
–
Thank you, @rohanpaul_ai for being a great resource for the #dev community In case you missed it, our recently launched SambaNova Cloud offers developers the chance to use @AIatMeta
's Llama 3.1 405B model for free! Start developing http://
cloud.sambanova.ai