Yoni Gottesman

Model Reproducibility

3

Hi, I think its important to provide scripts to reproduce paper results for the models the package implements. To be some kind of proof everything is working correctly. What do...

the line: self.Q = tf.reduce_sum(tf.multiply(self.output, self.actions_)) should be: self.Q = tf.reduce_sum(tf.multiply(self.output, self.actions_), axis=1)

In fasttext model: def forward(self, x): embedded_sent = self.embeddings(x).permute(1,0,2) h = self.fc1(embedded_sent.mean(1)) z = self.fc2(h) return self.softmax(z) Why is there no call to relu on h?

Return assistant generated tokens mask in apply_chat_template

2

# What does this PR do? This PR addresses issue #28950 and enhances the functionality of the `tokenizer.apply_chat_template` method when finetuning on chat datasets. The method tokenizer.apply_chat_template is recommended for...

This class is a singleton so keeps a reference to all blocks/ByteBuffers at all time so off heap memory is never released even if oak is closed and not referenced...

If i create a new builder like this: `OakMapBuilder builder = new OakMapBuilder() .setKeySerializer(stringSer) .setValueSerializer(stringSer) .setComparator(comparator)` I get an exception: `java.lang.NullPointerException at com.oath.oak.Insert$BenchmarkState$1.calculateSize(Insert.java:75) at com.oath.oak.Insert$BenchmarkState$1.calculateSize(Insert.java:52) at com.oath.oak.InternalOakMap.(InternalOakMap.java:66) at com.oath.oak.OakMap.(OakMap.java:74) at...

PreTrainedTokenizerFast `char_to_token` `token_to_char` not working as expected

4

### System Info - `transformers` version: 4.44.0 - Platform: macOS-13.6.9-arm64-arm-64bit - Python version: 3.11.4 - Huggingface_hub version: 0.23.4 - Safetensors version: 0.4.3 - Accelerate version: 0.32.1 - Accelerate config: not...

1. shadow cells per family 2. reduce client memory footprint