Feat: Add voxtral, magistral small 1.1, and misc gemma3n fixes

README.md

@@ -25,6 +25,7 @@
 
 ## 🎉 Latest Updates
 
+- 2025/07: Voxtral with mistral-common tokenizer support has been integrated in Axolotl. Read the [docs](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/voxtral)!
 - 2025/07: TiledMLP support for single-GPU to multi-GPU training with DDP, DeepSpeed and FSDP support has been added to support Arctic Long Sequence Training. (ALST). See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/alst) for using ALST with Axolotl!
 - 2025/06: Magistral with mistral-common tokenizer support has been added to Axolotl. See [examples](https://github.com/axolotl-ai-cloud/axolotl/tree/main/examples/magistral) to start training your own Magistral models with Axolotl!
 - 2025/05: Quantization Aware Training (QAT) support has been added to Axolotl. Explore the [docs](https://docs.axolotl.ai/docs/qat.html) to learn more!

examples/colab-notebooks/colab-axolotl-example.ipynb

@@ -40,7 +40,7 @@
         "%%capture\n",
         "# This step can take ~5-10 minutes to install dependencies\n",
         "!pip install --no-build-isolation axolotl[flash-attn]>=0.9.1\n",
-        "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@631d646\""
+        "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@010c3ac3f1e725098961832830303eeb4142dd88\""
       ]
     },
     {

examples/gemma3n/README.md

@@ -1,19 +1,65 @@
-# Gemma-3n
+# Finetune Gemma-3n with Axolotl
 
-## Requirements
+Gemma-3n is a family of multimodal models from Google found on [HuggingFace](https://huggingface.co/collections/google/gemma-3n-685065323f5984ef315c93f4). This guide shows how to fine-tune it with Axolotl.
 
-In addition to Axolotl's requirements, Gemma-3n requires
+## Getting started
 
-```
-pip3 install timm
+1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as Gemma3n is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
+
+    Here is an example of how to install from main for pip:
+
+```bash
+# Ensure you have Pytorch installed (Pytorch 2.6.0 min recommended)
+git clone https://github.com/axolotl-ai-cloud/axolotl.git
+cd axolotl
+
+pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
+pip3 install --no-build-isolation -e '.[flash-attn]'
 ```
 
-If you will load audio datasets, please also install
+2. In addition to Axolotl's requirements, Gemma-3n requires:
 
+```bash
+pip3 install timm==1.0.17
+
+# for loading audio data
+pip3 install librosa==0.11.0
 ```
-pip3 install librosa
+
+3. Run the finetuning example:
+
+```bash
+# text only
+axolotl train examples/gemma3n/gemma-3n-e2b-qlora.yml
+
+# text + vision
+axolotl train examples/gemma3n/gemma-3n-e2b-vision-qlora.yml
+
+# text + vision + audio
+axolotl train examples/gemma3n/gemma-3n-e2b-vision-audio-qlora.yml
 ```
 
-## Usage
+Let us know how it goes. Happy finetuning! 🚀
+
+WARNING: The loss and grad norm will be much higher than normal. We suspect this to be inherent to the model as of the moment. If anyone would like to submit a fix for this, we are happy to take a look.
+
+### TIPS
+
+- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
+- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
+- The text dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
+- The multimodal dataset format follows the OpenAI multi-content Messages format as seen [here](https://docs.axolotl.ai/docs/multimodal.html#dataset-format).
+
+## Optimization Guides
+
+- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
+- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
+- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
+
+## Related Resources
 
-See example configs and the [multimodal doc](https://docs.axolotl.ai/docs/multimodal.html).
+- [Gemma 3n Blog](https://ai.google.dev/gemma/docs/gemma-3n)
+- [Axolotl Docs](https://docs.axolotl.ai)
+- [Axolotl Website](https://axolotl.ai)
+- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
+- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)

examples/gemma3n/gemma-3n-e2b-vision-audio-qlora.yml

@@ -34,8 +34,6 @@ eot_tokens:
 datasets:
   - path: Nanobit/text-vision-audio-2k-test
     type: chat_template
-    data_files:
-      - dataset.jsonl
 dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./outputs/out

examples/magistral/README.md

@@ -1,6 +1,6 @@
 # Finetune Magistral Small with Axolotl
 
-Magistral Small is a 24B parameter opensource model from MistralAI found on [HuggingFace](https://huggingface.co/mistralai/Magistral-Small-2506). This guide shows how to fine-tune it with Axolotl with multi-turn conversations with proper masking.
+Magistral Small is a 24B parameter opensource model from MistralAI found on HuggingFace at [2506](https://huggingface.co/mistralai/Magistral-Small-2506) and [2507](https://huggingface.co/mistralai/Magistral-Small-2507) (see [Thinking](#thinking)). This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
 
 MistralAI has also released a proprietary medium-sized version called Magistral Medium.
 
@@ -13,7 +13,7 @@ Thanks to the team at MistralAI for giving us early access to prepare for this r
     Here is an example of how to install from main for pip:
 
 ```bash
-# Ensure you have Pytorch installed (Pytorch 2.6.0 recommended)
+# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
 git clone https://github.com/axolotl-ai-cloud/axolotl.git
 cd axolotl
 
@@ -31,12 +31,37 @@ This config uses about 24GB VRAM.
 
 Let us know how it goes. Happy finetuning! 🚀
 
+### Thinking
+
+MistralAI has released their [2507](https://huggingface.co/mistralai/Magistral-Small-2507) model with thinking capabilities. The model requires the multi-content dataset format with support for an extra `role: thinking` within system and assistant messages.
+
+Example format:
+
+```json
+{
+    "messages": [
+        {"role": "system", "content": [{ "type": "text", "text": "{SYSTEM_PROMPT}"}]},
+        {"role": "user", "content": [{ "type": "text", "text": "..."}]},
+        {"role": "assistant", "content": [{ "type": "thinking", "thinking": "..."}, { "type": "text", "text": "..." }]},
+    ],
+}
+```
+
+Example config: `./magistral-small-think-qlora.yaml`.
+
+The `thinking` section also supports an optional arg `closed: bool` (`True` default) which controls adding the closing `[/THINK]` tag.
+
+Limitations:
+- You cannot mix `content: str` with `content: list[dict]` as the `dataset.load_dataset` may complain about different types for `content` key.
+- This mode does not work with custom `train_detail` and `training` at the moment.
+
 ### TIPS
 
+- We recommend adding the same/similar SystemPrompt that the model is tuned for. You can find this within the repo's files titled `SYSTEM_PROMPT.txt`.
 - For inference, the official MistralAI team recommends `top_p: 0.95` and `temperature: 0.7` with `max_tokens: 40960`.
 - You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
 - Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
-- The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
+- The text dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
 
 ## Optimization Guides
 

examples/magistral/magistral-small-fsdp-qlora.yaml

@@ -6,6 +6,9 @@ tokenizer_use_mistral_common: true
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+
 load_in_8bit: false
 load_in_4bit: true
 

examples/magistral/magistral-small-qlora.yaml

@@ -6,6 +6,9 @@ tokenizer_use_mistral_common: true
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+
 load_in_8bit: false
 load_in_4bit: true
 

examples/magistral/magistral-small-think-qlora.yaml

@@ -0,0 +1,68 @@
+base_model: mistralai/Magistral-Small-2507
+
+# Enable to use mistral-common tokenizer
+tokenizer_use_mistral_common: true
+
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+
+load_in_8bit: false
+load_in_4bit: true
+
+datasets:
+  - path: Nanobit/text-think-2k-test
+    type: chat_template
+
+dataset_prepared_path: last_run_prepared
+val_set_size: 0
+output_dir: ./outputs/lora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 2048
+sample_packing: true
+
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: false
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 1
+saves_per_epoch: 1
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/voxtral/README.md

@@ -0,0 +1,76 @@
+# Finetune Voxtral with Axolotl
+
+Voxtral is a [3B](https://huggingface.co/mistralai/Voxtral-Mini-3B-2507)/[24B](https://huggingface.co/mistralai/Voxtral-Small-24B-2507) parameter opensource model from MistralAI found on HuggingFace. This guide shows how to fine-tune it with Axolotl.
+
+Thanks to the team at MistralAI for giving us early access to prepare for this release.
+
+## Getting started
+
+1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html). You need to install from main as Voxtral is only on nightly or use our latest [Docker images](https://docs.axolotl.ai/docs/docker.html).
+
+    Here is an example of how to install from main for pip:
+
+```bash
+# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
+git clone https://github.com/axolotl-ai-cloud/axolotl.git
+cd axolotl
+
+pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
+pip3 install --no-build-isolation -e '.[flash-attn]'
+```
+
+2. Please install the below.
+
+```bash
+# audio
+pip3 install librosa==0.11.0
+pip3 install 'mistral_common[audio]==1.8.3'
+```
+
+3. Run the finetuning example:
+
+```bash
+# text only
+axolotl train examples/voxtral/voxtral-mini-qlora.yml
+
+# text + audio
+axolotl train examples/voxtral/voxtral-mini-audio-qlora.yml
+```
+
+These configs use about 4.8 GB VRAM.
+
+Let us know how it goes. Happy finetuning! 🚀
+
+### TIPS
+
+- For inference, the official MistralAI team recommends `temperature: 0.2` and `top_p: 0.95` for audio understanding and `temperature: 0.0` for transcription.
+- You can run a full finetuning by removing the `adapter: qlora` and `load_in_4bit: true` from the config.
+- Read more on how to load your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
+- The text dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
+- The multimodal dataset format follows the OpenAI multi-content Messages format as seen [here](https://docs.axolotl.ai/docs/multimodal.html#dataset-format).
+
+
+## Optimization Guides
+
+- [Multi-GPU Training](https://docs.axolotl.ai/docs/multi-gpu.html)
+- [Multi-Node Training](https://docs.axolotl.ai/docs/multi-node.html)
+- [LoRA Optimizations](https://docs.axolotl.ai/docs/lora_optims.html)
+
+## Limitations
+
+We only support the `mistral-common` tokenizer for Supervised Fine-tuning at the moment and for `type: chat_template` only.
+
+In addition, we do not support overriding tokens yet.
+
+## Related Resources
+
+- [MistralAI Magistral Blog](https://mistral.ai/news/magistral/)
+- [Axolotl Docs](https://docs.axolotl.ai)
+- [Axolotl Website](https://axolotl.ai)
+- [Axolotl GitHub](https://github.com/axolotl-ai-cloud/axolotl)
+- [Axolotl Discord](https://discord.gg/7m9sfhzaf3)
+
+## Future Work
+
+- Add parity to Preference Tuning, RL, etc.
+- Add parity to other tokenizer configs like overriding tokens.