modelscope swift
@ms_swift2023
llm&mllm training framework from modelscope
You might like
We add some features to improve the GRPO training speed up to 200% based on the awesome work of TRL and Open-R1, check script here: github.com/modelscope/ms-… check wandb here: wandb.ai/tastelikefeet/…
SWIFT has supported the finetuning(VQA/Grounding/OCR/etc) of Qwen2-VL series models, check: github.com/modelscope/ms-…
And also support the finetuning of Florence series models: github.com/modelscope/swi…
SWIFT has support the finetuning of Gemma2: github.com/modelscope/swi…
SWIFT has supported the finetuning of Qwen2/GLM4/GLM4v: github.com/modelscope/swi…
We start a space in huggingface: huggingface.co/spaces/modelsc… meanwhile, we support peft 0.11.0 with hqq/eetq/vera/pissa training!
huggingface.co
Swift(LLM lightweight framework for fine-tuning and inference) - a Hugging Face Space by modelscope
Swift(LLM lightweight framework for fine-tuning and inference) - a Hugging Face Space by modelscope
Welcome to use quantized version of Yi1.5-6b/9b/34b! model can be found here: huggingface.co/modelscope
SWIFT is an LLM/MLLM training framework from the ModelScope community and supports models like LLaMA3/Mistral/DeepSeek-VL/YI-VL/Qwen-VL/LLaVA. We have newly supported the EETQ and HQQ QLoRA training for LLM and MLLMs: github.com/modelscope/swi… @_akhaliq
We support the training of qwen2-110b models, use this script: github.com/modelscope/swi… to train!
United States Trends
- 1. #TORQSports N/A
- 2. Malcolm Brogdon 3,020 posts
- 3. Argentina 487K posts
- 4. Banish 1,207 posts
- 5. SCOTUS 35K posts
- 6. Big Balls 26.7K posts
- 7. Waddle 4,090 posts
- 8. Russ 19.6K posts
- 9. Olave 3,443 posts
- 10. #ClockTower1Year N/A
- 11. Rickey 2,447 posts
- 12. $HIMS 4,551 posts
- 13. Aphrodite 5,442 posts
- 14. Biker Boyz N/A
- 15. #BeyondTheGates 5,638 posts
- 16. Voting Rights Act 33.2K posts
- 17. Maybe in California N/A
- 18. Capitol Police 29.9K posts
- 19. Martha 22K posts
- 20. Paul Pierce N/A
You might like
Something went wrong.
Something went wrong.