Thank You for Open-Sourcing Your Model & Feedback
Qwen Team,
Thank you for open-sourcing "Qwen/Qwen2.5-VL-32B-Instruct" ! We truly appreciate the effort and dedication that went into making this resource available to the community. Your work is an invaluable contribution to the field, and we’re excited to explore its capabilities further.
While testing the model, we noticed a few areas that could benefit from improvement:
Inclusion of Chinese Characters in English Conversations – In some cases, the model unexpectedly generates Chinese characters within predominantly English responses.Lack of Word Spacing – Words occasionally appear without proper spacing, making the text difficult to read.
Frequent Spelling Mistakes – There are noticeable spelling errors that affect readability and accuracy.
Use of Emojis at the End of Conversations – While emojis can enhance casual interactions, their presence in formal or technical discussions may not always be appropriate.
Also, doesn't respect the system prompt
We greatly appreciate your hard work and look forward to future updates. Thank you again for sharing this with the community!
Best regards,
Ramesh
Team, Thanks for the updated tensors. The above issues are seems to be resolved with the latest update. Thanks again for an excellent work and making it available.