unslothai/unsloth

Issues

• LLaMA-3.1-8B finetune: Unsloth does NOT load pre-finetuned QLoRA adapter correctly but its default ?

  #1045 opened 3 days ago by thusinh1969
  2

• Qwen-2.5 Coder-7B-Instruct: ValueError: Unsloth: Untrained tokens found, but embed_tokens & lm_head not trainable, causing NaNs. Restart then add `embed_tokens` & `lm_head` to `FastLanguageModel.get_peft_model(target_modules = [..., "embed_tokens", "lm_head",])

  #1053 opened a day ago by dante3112
  0

• A package issue?

  #1052 opened a day ago by savour-it-last
  0

• about finetuning

  #1051 opened a day ago by djahzo
  0

• TypeError: '<' not supported between instances of 'NoneType' and 'int'

  #1050 opened a day ago by BBaekdabang
  0

• Example notebook to train on responses only runs into error

  #1047 opened a day ago by NazimHAli
  0

• qwen2.5 support?

  #1043 opened 4 days ago by milsun
  2

• Qwen model loading Error says model not found on hugging face

  #1049 opened 2 days ago by vanshnawander
  0

• Fine tune and infer llama3 with cpu

  #1037 opened 5 days ago by SidneyLann
  2

• RuntimeError: Unsloth currently does not support multi GPU setups - but we are working on it!

  #1046 opened 2 days ago by YasamanJafari
  0

• Llama 3.1 (8B) fine-tuning demo suddenly stopped working during local training.

  #1010 opened 3 days ago by brainchen2020
  5

• How to use Custom Trainer from hugginface instead of SFTtrainer

  #1038 opened 5 days ago by ankitprezent
  1

• How to convert unsloth model to standard HF model e.g., to lead to vLLM?

  #1039 opened 4 days ago by brando90
  1

• ValueError: Unknown RoPE scaling type longrope

  #1042 opened 4 days ago by XCYXHL
  1

• What is the right way to load Qwen2's chat interface?

  #1040 opened 4 days ago by brando90
  4

• Issue With Mistral Small

  #1044 opened 4 days ago by DaddyCodesAlot
  1

• No Validation Loss logged (possibly related to train_on_responses_only?)

  #1019 opened 11 days ago by selalipop
  3

• train_on_responses_only doesn't map `eval_dataset`, breaking evaluation

  #1041 opened 4 days ago by fpgaminer
  2

• Can we continue finetuning LlaMA-3.1-8B's QLora pre-finetuned adapter with Unsloth to extend context length ?

  #1032 opened 8 days ago by thusinh1969
  6

• Any Plans To Support Solar Pro?

  #1031 opened 8 days ago by DaddyCodesAlot
  6

• Internlm2_5 support

  #1007 opened 14 days ago by Ammar-Alnagar
  3

• Discrepancy in LLaMA 3.1 performance when using custom trainer and SFTTrainer

  #1025 opened 6 days ago by salokr
  4

• Feature Request: Qwen2-VL support

  #1020 opened 11 days ago by 0xSt1ng3R
  3

• release cycle

  #1002 opened 8 days ago by aarnphm
  5

• ModuleNotFoundError: No module named 'unsloth'

  #1030 opened 8 days ago by vidithirve
  5

• More problems with train_on_responses_only

  #1017 opened 12 days ago by LostRuins
  3

• Is it safe to use FastLanguageModel.for_inference() during finetuning ?

  #1023 opened 9 days ago by Andrei997
  2

• When Flash Attention 2 is used and "use_dora = True", errored out: "RuntimeError: FlashAttention only support fp16 and bf16 data type"

  #1013 opened 13 days ago by rohhro
  2

• Flash Attention 2 doesn't work with Gemma 2 models

  #1014 opened 13 days ago by rohhro
  2

• Unsloth & XFormers keep crashing on each other !

  #1026 opened 10 days ago by thusinh1969
  4

• llama 2 based model does not stop generating answers during inference

  #1008 opened 14 days ago by Karoljv
  4

• Train on responses only does not work with TinyLlama-chat

  #1015 opened 13 days ago by akhlakm
  2

• Should throw exception for the current text generation call will exceed the model's predefined maximum length (8192)

  #1016 opened 12 days ago by SidneyLann
  1

• Inferencing on CPU (using fine tuned version of llama 3.1)

  #1012 opened 13 days ago by ApurvPujari
  14

• Full Finetune with Unsloth

  #1021 opened 11 days ago by user074
  1

• RuntimeError: Failed to import transformers.data.data_collator because of the following error (look up to see its traceback): maximum recursion depth exceeded

  #1022 opened 11 days ago by karacupa21
  2

• Add support for Qwen2Audio

  #1018 opened 12 days ago by jonflynng
  2

• target_modules "embed_tokens", "lm_head" really needed? try out FT with Nous Research LLAMA31 8B

  #1024 opened 11 days ago by carstendraschner
  1

• TemplateError: Only user and assistant roles are supported!

  #994 opened 12 days ago by DeeraDigiSpoc
  5

• model.push_to_hub_merged saves padding_side = "left"?

  #1006 opened 13 days ago by Karoljv
  2

• Issues with saving to hub -> Gemma based models

  #1005 opened 14 days ago by Ammar-Alnagar
  2

• Problem with Phi 3.5

  #1003 opened 16 days ago by DRXD1000
  2

• RuntimeError: expected self and mask to be on the same device, but got mask on cuda:7 and self on cuda:0

  #996 opened 18 days ago by Silentssss
  2

• Installation on Kaggle no longer works

  #998 opened 18 days ago by abedkhooli
  2

• push_to_hub_merged: How can I push to an organization?

  #1009 opened 14 days ago by ysy970923
  1

• Can add its Google Notebooks to THUDM/glm-4-9b-chat, thanks

  #1011 opened 14 days ago by FTDRTD
  1

• Can this fine tune the llama3.1 70b model

  #995 opened 15 days ago by xwan07017
  2

• Loading from Checkpoint

  #997 opened 18 days ago by MuhammadBilalKhan267
  0

• `construct_chat_template` raises RunTimeError for default template with trailing newline

  #992 opened 19 days ago by rodrigomeireles
  1

• "multiplicative LoRAs" for LLMs?

  #991 opened 19 days ago by jukofyork
  1