Vidyasagar Bhargava
@vidyasagarbhargava.bsky.social
Lead Data Scientist
#data #machinelearning #datascience #llm
#data #machinelearning #datascience #llm
I see..Since there was option to connect via Ollama so I was trying that. Anyways all the best to positron team. Great IDE.
June 26, 2025 at 8:31 AM
I see..Since there was option to connect via Ollama so I was trying that. Anyways all the best to positron team. Great IDE.
Summing up these weighted vectors gives us the context vector.
June 25, 2025 at 12:20 PM
Summing up these weighted vectors gives us the context vector.
Now before we pass all the hidden states to decoder we give them some scores and these scores are convert to softmax scores and multiplied by hidden states.
After multiplication some hidden state will amplify and some will drown based on score.
After multiplication some hidden state will amplify and some will drown based on score.
June 25, 2025 at 12:18 PM
Now before we pass all the hidden states to decoder we give them some scores and these scores are convert to softmax scores and multiplied by hidden states.
After multiplication some hidden state will amplify and some will drown based on score.
After multiplication some hidden state will amplify and some will drown based on score.
This is big!
June 24, 2025 at 5:02 PM
This is big!
How Attention model differs from prior sequence to sequence model?
1. Instead of passing only the last hidden state to decoder, now encoder passes all the hidden states.
2. Decoder does an extra step in order to focus on the part of the input that are relevent.
1. Instead of passing only the last hidden state to decoder, now encoder passes all the hidden states.
2. Decoder does an extra step in order to focus on the part of the input that are relevent.
June 24, 2025 at 4:49 PM
How Attention model differs from prior sequence to sequence model?
1. Instead of passing only the last hidden state to decoder, now encoder passes all the hidden states.
2. Decoder does an extra step in order to focus on the part of the input that are relevent.
1. Instead of passing only the last hidden state to decoder, now encoder passes all the hidden states.
2. Decoder does an extra step in order to focus on the part of the input that are relevent.
Cool..how you created this amazing hexsticker?
June 5, 2025 at 3:41 AM
Cool..how you created this amazing hexsticker?