ML Engineer & PhD in NLP
Joined August 2009
- Tweets 152
- Following 183
- Followers 155
- Likes 184
32 Photos and videos
Voice Writer is live on Product Hunt 👍🔼 Check it out and show your support by giving an upvote if you find it useful! producthunt.com/posts/voice-…
1
135
I built a voice writer tool to help you write things quickly. ⚡️ It uses AI for speech recognition and grammar correction. I have been using it for my book reviews, emails, Slack messages, and more. Here is a demo video. 😊
Try it out here: efficientnlp.com/voice-write…
1
146
🎥 New Video! In this video, we train a speech recognition model (using OpenAI's Whisper) to recognize our family's Chinese dialect, Teochew, or Chaozhou dialect (潮州话). It has about 10 million speakers and is a part of the Min Nan language family.
youtube.com/watch?v=JH_78KmP…
1
2
196
Challenges we faced:
- Teochew is related to Mandarin, a high-resource language, but how do we apply transfer learning?
- With zero resources for training, we had to build our dataset from scratch. 🛠️
- Teochew doesn't even have a writing system! How do we model that? 🤔
96
📹 New Video! Ever had trouble deciding which AI to use for your projects? Let's solve that with AI. In this video, I will build an AI to find the best AI for you 🤯
youtu.be/2r-SqtxhgmY
1
1
140
📹 New Video! #LLMs can be slow, so this @GoogleDeepMind paper proposed to speed it up by running two LLMs at the same time. 😕 Wait what?
1
123
It's a new technique called speculative sampling. A smaller LLM generates the easier tokens and a larger LLM checks them. And using a rejection sampling trick, there is no difference in accuracy!
Check out my video on how this works ➡️youtube.com/watch?v=S-8yr_Ri…
113
📺 New video: Map of Natural Language Processing
youtube.com/watch?v=zF69Eqhi…
1
1
839
Ever since its introduction in 2017, the transformer architecture has remained largely unchanged...until now.
youtube.com/watch?v=o29P0Kpo…
1
2
126
In this video, I explain RoPE - Rotary Positional Embeddings. Proposed in 2022, RoPE is making its way into LLMs like Google's PaLM and Meta's LLaMa. I unpack the magic behind rotary embeddings and reveal how they combine the strengths of both absolute and relative embeddings.
2
115
📹 New Video: Understanding the KV Cache
Why is it so difficult to run LLMs with longer context windows? 🤔 The culprit behind this memory hog is none other than the KV cache. In this video, I explain this essential component and its impact on GPU memory during inference.
1
1
147
Watch the video here ➡️ youtu.be/80bIUggRJf4
📈 Don't forget to subscribe and hit that like button if you find this content informative!
1
92
Video link below
youtube.com/watch?v=UcwDgsMg…
#neuralnetworks #video #inference #llmops #llms #naturallanguageprocessing
87