Roll like @AndrewYNg 😉 and enjoy experiencing @AIatMeta 's 405B model at 120 tokens/sec on @SambaNovaAI fast API service. If you are looking to power your generative AI applications and workloads, sign up here -
sambanova.ai/fast-api?api_re…
I've been playing with @SambaNovaAI's API serving fast Llama 3.1 405B tokens. Really cool to see leading model running at speed. Congrats to Samba Nova for hitting a 114 tokens/sec speed record (and also thanks @KunleOlukotun for getting me an API key!) sambanova.ai/blog/speed-reco…
Thanks for the shout out @AndrewYNg ! If anyone needs an API key, reach out to me or DM @SambaNovaAI . Happy to provide you one.
We have 405B running at 120 tokens/sec at full precision and on a really small footprint.
I've been playing with @SambaNovaAI's API serving fast Llama 3.1 405B tokens. Really cool to see leading model running at speed. Congrats to Samba Nova for hitting a 114 tokens/sec speed record (and also thanks @KunleOlukotun for getting me an API key!) sambanova.ai/blog/speed-reco…
Generative AI based summarization gone wrong? :)
LLM hallucinations can be real. Thats why we ask our customers at @SambaNovaAI to use composite systems to reduce such issues. Models finetuned for NLI tasks can help catch such contradictions or using LLM as a judge in your pipeline might also go a long way to reduce such issues.
You can try tulu2-7b/70b for NLI task and autoj-13b for GPT-as-a-judge from sambaverse.sambanova.ai/ to reduce your hallucinations or a good system prompt for llama3 at fast.snova.ai/#generativeai#llm#cwc
SambaNova is honored to take part in the Trillion Parameter Consortium (TPC) event in Barcelona this week: tpc.dev/tpc-european-kick-of….
The main stage will feature talks by industry leaders including Rick Stevens, Associate Laboratory Director, @argonne, @ProfMatsuoka, Director, @RIKEN_RCCS, and Mateo Valero, Director, @BSC_CNS.
SambaNova’s @UrmishThakker will participate in a workshop, titled “AI Hardware Acceleration Strategies at Scale”, where they will discuss how the explosion in the size of training and inference workloads caused by the race to LLMs has created unprecedented computational demands.
#AI#GenAI#AIChips@jenniferglore@MarshallChoy
The era of spatial computing is here. Where digital content blends seamlessly with your physical space. So you can do the things you love in ways never before possible. This is Apple Vision Pro.
The secret to growing in your career: managing up.
I thought it was BS but it’s critical to career growth. Your manager wants to you manage them so it’s easier for them to manage you.
5 things I’d tell my younger self: