@objectgraph Tried it with a batch size of 2 and lora layers of 4. The model is underfitting by a lot (val loss at 2 from 0 to 600 iter and train loss from 2 to 1.7). Should I try to increase batch size and lora layers (my dataset is 27k messages long)
@NBMSacha@KashanSid3@aiden0x4@Mobulaio Why not fine tuning a LLM model to return a search API request based on a user query in natural language? Peak friendly UX for the 95% of users
No sugar coating this, it's been a challenging few months. The Truffle Suite will be sunsetting over the next 90 days.
To all the developers who've supported us, thank you! It's been an absolute joy to strive with you & the Ethereum ecosystem for the past 6 years. 🧵