Peter Albert

Model Request: Blenderbot 2.0

26

# 🌟 New model addition ## Model description Facebook released Blenderbot 2.0, a chatbot that builds on RAG and Blenderbot 1.0. It can save interactions for later reference and use...

Include run id in train.log

1

Added a --run-id flag that appears in logging at each training step. As default value I added a time stamp of the beginning of training. This allows for easy identification...

Unify sweep.py and slurm.py between metaseq and metaseq-internal

3

Continuation of https://github.com/facebookresearch/metaseq/pull/476 . As metaseq-internals unification PR was not merged, a few other features got added to metaseq-internal's sweep and slurm. I brought these here into metaseq. Note: Gpu...

Allow to skip batches during training

2

Allows skipping ranges of batches during training.

## 🐛 Bug When starting a training run the model starts hanging at the first forward pass. This happened when I used the small book dataset used in the gpu_test/test_training_integrity.py....

Rewrite of the load_checkpoint function

2

This is a rewrite of how we determine which checkpoint to load when starting/restarting a training run. (Originally there was also a refactor of how our different checkpoint paths are...

### What happened? When using the new tool call functionality of anthropic and you add an image to your message (according to the openai's message format) the message conversion to...

Allow multiple tools in anthropic response

3

Currently in litellm's anthropic response only the first tool is extracted and the rest is ignored. This PR allows to also parse the rest of the tools.

Error while clustering

2

After starting the clustering I get this error: ``` [local/evol1][1 shards] map "extract_text" to "('prompt__cluster',)": 100%|████████████████████████████████████████████████████████████████████████████████| 319/319 [00:00