I would like to see a comparison for the inference time compute between a regular transformer and this. I’m assuming token/s is lower since you need to compute the weights of the model for each token prior to the actual attention calculations for the sequence position.
TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
11–20 of 34 posts
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#12Seems like a big deal. I feel like this could enable a new level of modularity and compatibility between publicly available weight sets, assuming they use similar channel dimensions. Maybe it also provides a nice formalism for thinking about fine tuning, where you could adopt certain heuristics for adding/removing key-value pairs from the Pattention layers. One interesting thing to note: sounds like model scaling hap…
Unless I’m reading it wrong I don’t think rows matter. Attention doesn’t take into account sequence position natively, that’s why positional encodings exist.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#13Seems like a big deal. I feel like this could enable a new level of modularity and compatibility between publicly available weight sets, assuming they use similar channel dimensions. Maybe it also provides a nice formalism for thinking about fine tuning, where you could adopt certain heuristics for adding/removing key-value pairs from the Pattention layers. One interesting thing to note: sounds like model scaling hap…
Unless I’m reading it wrong I don’t think rows matter. Attention doesn’t take into account sequence position natively, that’s why positional encodings exist.
I'm talking about the rows in the new K and V matrices introduced by the paper, not rows in the input sequence. The ordering of rows in the new K and V matrices does matter in the sense that rows that appear further down were added later in the training process to add new parameter tokens during scaling. So those newer parameters may represent knowledge that is less fundamental and more about fine tuning on the training set.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#14This is a great idea. Being able to dynamically scale up model sizes as datasets and use cases expand without needing to retrain from scratch could enable a Cambrian explosion of interesting stuff building on top of a Llama type model trained in this way.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#15Seems this would naturally translate into a mixture of experts by using a "hard" attention function so that only a fixed amount of weight tokens get included in the calculation.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#16I would like to see a comparison for the inference time compute between a regular transformer and this. I’m assuming token/s is lower since you need to compute the weights of the model for each token prior to the actual attention calculations for the sequence position.
Isn't that figure 5 in the paper? It's for training not inference, but presumably if training is faster then inference would be too. Because they don't increase the dimension of the text tokens when scaling up, which reduces the compute needed for attention. But potentially limits how well the text token attention can keep track of things, because it's got less space for passing things along.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#17Seems like a lot of existing models could be converted to this token parameter representation.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#18This seems closely related to the "Mixtral" approach of a mixture-of-experts transformer [1]... I'm not claiming the approach is not original, it just helped me understand what was going on.
Consider a case of two "experts" or two "value parameter tokens."
The mixture of experts has a "router" network that provides a weight to each expert (through a softmax) conditional on an input. The output is a (sparse) weighted sum of the outputs of the experts.
The TokenFormer has an "attention" layer combines the token and a key value to provide a weight to each "value parameter" token. A(B+C) = AB + AC definitionally, so this is like applying a weighted sum of distinct transformations.
I think the differences are: a) where the non-linearity hits (the above description doesn't consider an activation function), b) this attention softmax is not (necessarily) sparse, c) that "mixtral" networks only replace the feed-forward components of the layer, and d) that extending a "mixtral" approach would require re-training the "router" layers.
It seems like (d) is maybe the nicest feature here... my intuition would think (a) doesn't matter much, (b) is debatable (how close a sparse-MoE can approximate a dense-MoE), (c) has probably been tried (guessing the ffwd limitation was just "more-bang-for-buck-given-parameters" not an oversight)...
... I wonder, though, if there might be diminishing returns here (I believe that Mixture-of-Experts tends to struggle with imbalanced "winner-take-all" dynamics, since "early" winners get more gradient signal to improve their weights) and how different this would have been from going from 3x7B to a 8x7B to a 24x7B training approach (with a "retrain routing networks" step).
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#19Eventually people will figure out how to nest neural networks in the nodes and edges of an arbitrary graph.
Re: TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
#20The authors factorize every weight matrix with an attention mechanism: weight = attention(token_query, weight_keys, weight_values). In other words, they query weight_keys to fetch the weight_values, and mix them to compute each weight on the spot. Increasing model size becomes a matter of adding more weight_keys and weight_values, and incrementally training them. Simple, clever, and it seems to work well. Beautiful.
There is a particularly nice geometric interpretation of attention I just realised recently in a flash of enlightenment, best explained with an interactive Desmos plot (black dot is draggable): https://www.desmos.com/calculator/3rtqsyapxo The above assumes the columns of K are normalised but bear with me. K and V together form a vector database. V are the payloads, each row containing a vector of data. K describes th…
It looks fascinating, but i don't understand it. I'm haven't gone yet deeply into the theory of attention networks.
Can you explain the desmos plot in simple terms?