v2.2.3

New Features:

1. support bnb and ollama export
2. suport Q-Galore

New Models:

1. numina-math-7b

Bug Fix:

1. fix vllm>=0.5.1, TP
2. fix internvl2 template
3. fix glm4v merge-lora

What's Changed

• fix internvl doc by @hjh0119 in #1394
• Fix link by @Jintao-Huang in #1397
• fix vllm==0.5.1 by @Jintao-Huang in #1404
• [TorchAcc] update accelerate API and add llama3-70B by @baoleai in #1400
• Support Ollama and BNB for export by @tastelikefeet in #1407
• Fix glm4v merge lora by @Jintao-Huang in #1410
• [TorchAcc] fix model download when using TorchAcc distributed training by @baoleai in #1408
• Support padding left by @tastelikefeet in #1414
• Fix ollama export by @tastelikefeet in #1416
• fix web-ui params by @tastelikefeet in #1417
• fix hub_token by @Jintao-Huang in #1420
• Update ms hub token by @Jintao-Huang in #1424
• Add numina math model by @tastelikefeet in #1421
• fix internvl template by @Jintao-Huang in #1433
• Internvl series models update by @hjh0119 in #1426
• fix internvl2 template by @Jintao-Huang in #1436
• Fix bug and make lazydataset more stable by @tastelikefeet in #1438
• Fix llava-hf by @tastelikefeet in #1439
• [WIP]Support Q-Galore by @tastelikefeet in #1440
• 1. support deepspeed on ui 2. add tools to client_utils by @tastelikefeet in #1446
• fix read csv (float) by @Jintao-Huang in #1447
• fix dataset by @tastelikefeet in #1448
• update internvl doc by @hjh0119 in #1449

Full Changelog: v2.2.2...v2.2.3