jetstream-v0.2.3

What's Changed

• Enable jax profiler server in run with ray by @FanhaiLu1 in #112
• Add for readme interleave multiple host with ray by @FanhaiLu1 in #114
• Fix conversion bug by @yeandy in #116
• Integrate disaggregated serving with JetStream by @FanhaiLu1 in #117
• Support HF LLaMA ckpt conversion by @lsy323 in #118
• Add guide on adding HF ckpt conversion support by @lsy323 in #119
• Add support for Llama3-70b by @bhavya01 in #101
• Fix convert_checkpoint.py for hf and gemma by @qihqi in #121
• Mixtral enablement. by @wang2yn84 in #120
• add script to isntall for GPU by @qihqi in #122
• Add activation quantization support to per-channel quantized linear layers by @lsy323 in #105
• Remove JSON config mangling for Gemma ckpt by @lsy323 in #124
• Add different token sampling algorithms to decoder. by @bvrockwell in #123
• Add lock in prefill and generate to prevent starvation by @FanhaiLu1 in #126
• Update submodules, prepare for leasing v0.2.4 by @qihqi in #127
• Update README.md by @qihqi in #128
• Update summary.md by @qihqi in #125
• Update README.md by @bhavya01 in #129
• make sure GPU works by @qihqi in #130

New Contributors

• @yeandy made their first contribution in #116
• @bvrockwell made their first contribution in #123

Full Changelog: jetstream-v0.2.2...jetstream-v0.2.3

jetstream-v0.2.2

https://github.com/google/JetStream/releases/tag/v0.2.2

jetstream-pytorch 0.2.2

• Miscellaneous bug fixes.
• Support of Tiktoken tokenizer
• Support for Gemma - 2b model (running data parallel)

jetstream-v0.2.1

Key Changes

• Support Llama3
• Support Gemma
• Ray Multiple Host Single Pod Slice MVP
• Enable unit test and format check

jetstream-v0.2.0

Release JetStream Pytorch with JetStream v0.2.0 for inference