Local LLMを中心に、AI技術系の投稿とネタツイ Tesla v100 32GB×2枚と共に生きる。 最近はFlash Nextちゃんと遊んでいます アイコンは羊野郎さん@maton_3110
Qwen3.8 27Bから浮気中
Joined August 2019
- Tweets15.8K
- Following1.9K
- Followers7.1K
- Likes48K
Ryu retweeted
this is actually your first decision when downloading muse in japan
こんにちは日本！
I came to eat sashimi and assist you in your everyday tasks...
and I’m all out of sashimi.
download here: muse.ai 🇯🇵
Ryu retweeted
MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token.
Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym.
Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency.
Hugging Face:
huggingface.co/MiniMaxAI/Min…
ModelScope:
modelscope.cn/models/MiniMax…
API: platform.minimax.io
M Plan: minimax.io/m-planMiniMax
MCode for desktop: agent.minimax.io/download
MiniMax M3.1 Flash: Space Performance + Bunny Speed
🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash.
💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence across key tasks such as Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed.
API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens.
Model weights will arrive on Oct 23.
API: platform.minimax.io
M Plan: minimax.io/m-plan
MiniMax Code for desktop: agent.minimax.io/download
Ryu retweeted
M3.1 Flash weights are now open 🔓
Thanks to @togethercompute for being a day-0 launch partner and bringing M3.1 Flash to production on the AI Native Cloud. Open weights, no GPUs to manage. Great building together! 🐰
minimax m3.1 flash is now live on together ai
we’re excited to be a day-0 launch partner for the model, aka space bunny
m3.1 flash is built for agentic tasks, coding, computer use + cybersecurity, with a 1m context window and vision support
try it now together.ai/models/minimax-m…
Ryu retweeted
Can we get @Googlegemma's Embedding Gemma 2 to speak?
A couple of days ago, I posted about my experiment to get Embedding Gemma 2 to transcribe. Then internet did its internet thing :D and it reached over 238k views on X and over 85K views on LinkedIN, and got it on the front page of HN
xitter.cf/TslShahir/status/21096…
So, naturally, I wanted to test the reverse. Can we get EG2 to generate speech? Turns out that we can. At least part of it. We used EG2's semantic word embeddings to plan prosody and other details, then a small model generates the audio representation as EG2 would have heard that and a vocos vocoder reconstructs the final voice form
And as it happened with STT, this turned out to be ultrafast. 1000 tokens in 3.8 seconds. 79.5x realtime on a base M3 air.
Please check out the video. The quality is not close to the best in the industry, but this is the just the first iteration.
I turned @googlegemma EmbeddingGemma 2, an embedding model, into a speech-to-text model, and it turned out to be one of the fastest STT models you can run on device.
On a base M3 MacBook Air it transcribes an hour of audio in about 13 seconds: 288× real time.
The idea came in March with Gemini Embedding 2. It puts text, images, audio and video into one shared space, so a spoken word lands right next to the written word. If that's true, you should be able to read a transcript straight out of the embedding model. I couldnt tried it then, but last week EmbeddingGemma 2 came out as an open model, so I tried it.
Ryu retweeted
Can you use EmbeddingGemma 2 for Speech-to-Text?
This experiment uses EmbeddingGemma 2’s unified audio-text vector space to match speech against text via vector similarity.
The result? Transcribing an hour of audio in about 13 seconds, locally!
Ryu retweeted
Today, we announce 𝙻𝚒𝚚𝚞𝚒𝚍 𝚁𝙸𝙿𝙻 (real-time inference pipeline), our ultra low-latency inference engine built to put LFMs directly in the request path of real-time workloads like search, ads, fraud, recommendations, and payments.
> complete generated responses end-to-end in sub 50ms
> maintains low latency under continuous peak load
> LFMs get up to 20x faster on RIPL
> RIPL supports Multimodal LFMs
> RIPL supports d1, our decision models too!
RIPL optimizes the full path from request arrival to a complete generated response. As traffic rises, its performance advantage is greatest on the slowest requests. Let’s see some production use cases:
