too bad this is fake :'( not even sure you can fit 16 h100s in a single machine lol
@jxmnop
-
Gemini vs Mistral: COT Performance Comparison Analysis
By
–
gemini went for absolute reasoning and it killed them when people started talking to the model see here where open-source mistral matches gemini pro performance on "regular" inputs but loses (slightly) with COT
-
LLM Training Trade-offs: Chain-of-Thought vs Chatbot Optimization
By
–
biggest lesson I learned from gemini is that LLM trainers now have to choose between overfitting to chain-of-thought-type inputs (best for absolute reasoning ability) and over-fitting to human chatbot interactions (best for talking to humans) no free lunch here, have to choose
-
Karpathy appears influenced by social media discussions
By
–
guess karpathy has been reading my tweets
-
Eigenfaces for AI Interpretability: Understanding Computer Vision
By
–
dunno, just seems pretty cool. i guess, interpretability? i mean, what do people use eigenfaces for?
-
Sequence Text Embeddings vs Word Embeddings Explained
By
–
i'm talking about sequence text embeddings not word embeddings
-
Sequence-Based Models Capture High-Level Features
By
–
don't think so! these are based on sequences so they should capture much higher-level features
-
Recreating Articles Using Eigentext Vector Representations
By
–
the obvious next step from there is to recreate specific articles based on their eigentext vectors – essentially this part
-
Eigenfaces for Text: PCA on Domain-Specific Embeddings
By
–
project request: eigenfaces, but for text do PCA on text embeddings from a particular domain and find the "eigentext" embeddings that represent that domain then project back to text with vec2text. what would the eigentexts look like? surprised no one has done this yet
-
Google researcher struggles to beat Mistral 7B model performance
By
–
> be google researcher
> exec says we need to beat mistral 7b
> “reclaim the narrative"
> train 7b model
> run eval. check plots
> cant do it. maybe with 8b
> exec says no dice
> “we'll look like chumps"
> ok
> maybe we can tie embeddings and linear_f, save some params
> not