7 Pull requests merged by 6 people

• Bug Fixes

  #1470 merged Dec 24, 2024

• Name Error Bug Fix - import from packaging.version import Version

  #1468 merged Dec 23, 2024

• Bug fixes

  #1458 merged Dec 20, 2024

• Pass position embeddings explicitly from decoder layer.

  #1442 merged Dec 20, 2024

• Change _fix_chat_template in case a template has both endif and endfor

  #1388 merged Dec 20, 2024

• Update save.py warning message

  #1425 merged Dec 20, 2024

• Fix loader.py to work on Windows

  #1453 merged Dec 20, 2024

7 Issues closed by 6 people

• Support for Aya-23-8B

  #761 closed Dec 25, 2024

• Issues with saving to hub -> Gemma based models

  #1005 closed Dec 24, 2024

• Issue with fine-tuning Vision models (Pixtral and mistral )

  #1441 closed Dec 24, 2024

• ImportError: cannot import name 'EntryNotFoundError' from 'huggingface_hub.errors' (/usr/local/lib/python3.10/dist-packages/huggingface_hub/errors.py)

  #1459 closed Dec 23, 2024

• huggingface_hub.errors.HFValidationError: Repo id must use alphanumeric chars or '-', '_', '.', '--' and '..' are forbidden, '-' and '.' cannot start or end the name, max length is 96: 'unsloth/llama-3-8b-bnb-4bit\*.json'.

  #1455 closed Dec 23, 2024

• How to use train_on_responses_only on Qwen2VL model?

  #1466 closed Dec 23, 2024

• Train a model from scratch

  #1451 closed Dec 20, 2024

15 Issues opened by 14 people

• how to use pipeline load the fastlanguage model

  #1471 opened Dec 24, 2024

• issues saving finetuned Vision models to hub "NameError: name 'Version' is not defined

  #1469 opened Dec 23, 2024

• Does train_on_responses_only support Qwen2VL model? How to use train_on_responses_only on Qwen2VL model?

  #1467 opened Dec 23, 2024

• Extracting Image-Text Fusion Features from Fine-Tuned LLaMA 3.2-Vision Architecture

  #1464 opened Dec 22, 2024

• Triton dep in conda installation

  #1463 opened Dec 22, 2024

• Sampling from adapters different than sampling from saved merged model

  #1462 opened Dec 22, 2024

• Adapting Unsloth's implementation for classification tasks

  #1461 opened Dec 21, 2024

• Finetuning QwQ - Validation Loss NaN

  #1460 opened Dec 21, 2024

• Feature: Insight into when a concept has been understood / grokked (code available)

  #1457 opened Dec 20, 2024

• Batch inference produces inconsistent results for self-trained model

  #1456 opened Dec 20, 2024

• Question about Custom Attention Mask Support

  #1454 opened Dec 20, 2024

• Chat template mappings to the `get_chat_template` keys

  #1452 opened Dec 19, 2024

• tied embeddings saving with qwen2-vl

  #1450 opened Dec 19, 2024

• Dynamic 4bit quant for merged adapter models

  #1449 opened Dec 19, 2024

• RuntimeError: Unsloth: Your repo has a LoRA adapter and a base model.

  #1448 opened Dec 19, 2024

25 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• To not print unsloth introduction

  #1438 commented on Dec 19, 2024 • 0 new comments

• Train on responses only does not seem to work for Mistral format

  #1290 commented on Dec 19, 2024 • 0 new comments

• PermissionError: [WinError 5]

  #914 commented on Dec 19, 2024 • 0 new comments

• PPO

  #884 commented on Dec 19, 2024 • 0 new comments

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

  #1041 commented on Dec 19, 2024 • 0 new comments

• Support for model trained by OLMo?

  #774 commented on Dec 19, 2024 • 0 new comments

• Performance of fine-tuned model imported into Ollama from adapters differs from Unsloth inference

  #1073 commented on Dec 20, 2024 • 0 new comments

• Apple Mac Book MLX support

  #68 commented on Dec 20, 2024 • 0 new comments

• Recent paper(s) about memory reduction improvements on optimizers

  #1439 commented on Dec 20, 2024 • 0 new comments

• Installation Guide for unsloth on Windows

  #402 commented on Dec 20, 2024 • 0 new comments

• llama.cpp GGUF breaks [FIXED]

  #1376 commented on Dec 21, 2024 • 0 new comments

• Continue pretraining an instruction-fine-tuned LLM model like Qwen2.5-7B-Instruct.

  #1405 commented on Dec 21, 2024 • 0 new comments

• how to fix flash attention broken installation

  #1437 commented on Dec 21, 2024 • 0 new comments

• Kaggle unsloth module not found

  #1432 commented on Dec 22, 2024 • 0 new comments

• FastLanguageModel.from_pretrained fails validate_repo_id in huggingface_hub

  #1222 commented on Dec 22, 2024 • 0 new comments

• Unslot fine tunes no longer work out of the box with GPT4ALL as of update 3.5 and 3.5.1 .... 3.4.2 still working.

  #1424 commented on Dec 23, 2024 • 0 new comments

• OSError: Can't load tokenizer for hf uploaded model. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure hf upleaded is the correct path to a directory containing all relevant files for a PreTrainedTokenizerFast tokenizer.

  #1447 commented on Dec 23, 2024 • 0 new comments

• Error with gguf conversion.

  #1416 commented on Dec 23, 2024 • 0 new comments

• KTOTrainer don't have 'rewards/accuracies' metric

  #1440 commented on Dec 23, 2024 • 0 new comments

• Train Text Only for VLMs

  #1436 commented on Dec 23, 2024 • 0 new comments

• TypeError: expected string or bytes-like object

  #1363 commented on Dec 23, 2024 • 0 new comments

• Validation during training for VLMs?

  #1341 commented on Dec 24, 2024 • 0 new comments

• RuntimeError: CUDA driver error: operation not supported

  #1446 commented on Dec 24, 2024 • 0 new comments

• Full Finetune with Unsloth

  #1021 commented on Dec 25, 2024 • 0 new comments

• RuntimeError: Unsloth: The file 'llama.cpp/llama-quantize' or 'llama.cpp/quantize' does not exist

  #748 commented on Dec 25, 2024 • 0 new comments