Weight-rebase vision construction
According to the model card, Mistral-Small-3.2-24B-Instruct-2506 supplies the vision chassis while Magistral-Small-2507 supplies the language_model weights, with a small expected quality degradation.
Open Source Model Profile · kmouratidis
Magistral-Small-2507-Rebased-Vision
Magistral-Small-2507-Rebased-Vision is a 24.01B-parameter image-text-to-text model from kmouratidis. According to the model card, it places Magistral-Small-2507 weights into a Mistral-Small vision chassis without further training.
- Publisher
- kmouratidis
- Task
- image-text-to-text
- Model type
- mistral3
- License
- apache-2.0
- Library
- transformers
- Publication status
- Approved for indexing
Model overview
Magistral-Small-2507-Rebased-Vision is published by kmouratidis as an image-text-to-text model. The captured configuration identifies Mistral3ForConditionalGeneration, and Safetensors metadata reports 24,011,361,280 parameters. According to the model card, it combines Magistral-Small-2507 language weights with a Mistral-Small vision model.
Recorded capabilities
No post-rebase training
According to the model card, no training or fine-tuning followed the rebase, and the publisher relied on unsloth uploads for extra files and configs.
Publisher-described degradation check
According to the model card, the publisher ran an internal 7-benchmark check with lm_eval and sglang to approximate degradation, noting it is not comprehensive and not directly comparable to official Mistral numbers.
Vision-text conversational record
Captured hub records identify image-text-to-text use with vision, reasoning, and conversational tags alongside broad language-code coverage.
Use cases in the source record
- Image-plus-text conversational experiments that combine Magistral-family language behavior with vision input.
- Downstream-task testing by the user, since the publisher describes its own check as approximate and asks users to test on their own tasks.
Limitations and unknowns
- No evaluation results were extracted from this record as structured data; the card's 7-benchmark check is a publisher-described internal exercise.
- The publisher notes users should expect a small quality degradation from the weight rebase.
- According to the model card, release code, scripts, and further evaluation details were still pending at capture time.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: kmouratidis/Magistral-Small-2507-Rebased-Vision
Captured: Unknown. Processed: 2026-09-07T19:35:57.877890+00:00.
Magistral-Small-2507-Rebased-Vision This model was created by taking Mistral-Small-3.2-24B-Instruct-2506 and replacing the weights under the language_model with the weights from Magistral-Small-2507. The result is Magistral with vision capabilities, but you should expect a small quality degradation. Notes: I used unsloth's uploads of these models for convenience, since they include some extra files and configs too. I didn't name this "-Vision" because it was not trained or finetuned after weight rebase, and in case a future version by mistralai has vision. The code will be released soon. Evaluation Evaluation was performed on 7 benc…
F001F002F003F004F005F006F007F009F010F011F012F013F014