Instructions to use moonbox-dev/Voxtral-Mini-3B-2507-ONNX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- ONNX Runtime GenAI
How to use moonbox-dev/Voxtral-Mini-3B-2507-ONNX with ONNX Runtime GenAI:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Voxtral Mini 3B ONNX
ONNX builds of Voxtral Mini 3B, made by Moonbox for on-device use with ONNX Runtime (CPU, CUDA, DirectML). Each
precision is one folder holding the graphs and a small Python pipeline.py that does the rest (features, prompt,
decoding loop); what the graphs take and return is under How to run.
Files
The last column is the benchmark on Moonbox's ONNX speech-recognition set (400 LibriSpeech test-clean utterances from all 40 speakers, plus 200 AISHELL-1 test utterances for models that speak Chinese; details under How to run): the word (or character) error rate, and how often the build reads a clip differently from the original PyTorch model.
| Precision | File | Size | sha256 | Benchmark |
|---|---|---|---|---|
| fp16 | fp16/audio_encoder.onnx |
1.32 GB | 8122fe66c8d4 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/config.json |
0.6 kB | ad7603ee5568 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/decoder.onnx |
0.2 MB | bcfd8d624637 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/decoder.onnx.data |
7.26 GB | 79230de209d5 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/embed.onnx |
805.3 MB | c183feaa6767 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/mel_filters.npy |
0.1 MB | 8abf12444e0d |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/ortllm.py |
5.7 kB | c31d132e14aa |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/pipeline.py |
4.1 kB | f43d72dd0f37 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| fp16 | fp16/tekken.json |
14.9 MB | 4aaf3836c2a5 |
WER 1.73 % (original 1.77 %, bf16); 378 of 400 transcripts identical, 0.07 % read differently (CUDA, RTF 0.12) |
| int8 | int8/audio_encoder.onnx |
671.6 MB | bcec8409cf03 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/config.json |
0.6 kB | ad7603ee5568 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/decoder.onnx |
0.2 MB | 75bd3290baa3 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/decoder.onnx.data |
3.87 GB | da3a09604ba3 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/embed.onnx |
402.9 MB | 5c05f1bc8f33 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/mel_filters.npy |
0.1 MB | 8abf12444e0d |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/ortllm.py |
5.7 kB | c31d132e14aa |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/pipeline.py |
4.1 kB | f43d72dd0f37 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int8 | int8/tekken.json |
14.9 MB | 4aaf3836c2a5 |
WER 1.74 %; 367 of 400 identical, 0.12 % read differently (CUDA, RTF 0.10); GPU build |
| int4 | int4/audio_encoder.onnx |
434.6 MB | 50e3d41af9c0 |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/config.json |
0.6 kB | ad7603ee5568 |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/decoder.onnx |
0.2 MB | 590de890388a |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/decoder.onnx.data |
2.07 GB | 639eb5254ef1 |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/embed.onnx |
402.9 MB | 5c05f1bc8f33 |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/mel_filters.npy |
0.1 MB | 8abf12444e0d |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/ortllm.py |
5.7 kB | c31d132e14aa |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/pipeline.py |
4.1 kB | f43d72dd0f37 |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
| int4 | int4/tekken.json |
14.9 MB | 4aaf3836c2a5 |
WER 1.78 %; 297 of 400 identical, 0.48 % read differently (CUDA, RTF 0.098; CPU RTF 2.7) |
How to run
Voxtral Mini 3B (2507) (Mistral's speech-understanding model; this build runs its transcription mode: English,
Spanish, French, Portuguese, Hindi, German, Dutch and Italian, 16 kHz mono input) as three ONNX graphs plus
pipeline.py and ortllm.py, a numpy runtime that does the features, the transcription prompt and greedy decoding
around them:
audio_encoder.onnx: the Whisper-style audio encoder and the projector for one 30-second chunk.input_features[1, 128, 3000] (log-mel) ->audio_embeds[375, 3072]. Longer audio is a series of 30 s chunks, as in the original processor.embed.onnx:input_ids[1, S] ->inputs_embeds[1, S, 3072].decoder.onnx: the Llama-architecture text decoder built with onnxruntime-genai's model builder.inputs_embeds,attention_mask[1, total],past_key_values.{0..29}.key / .value[1, 8, past, 128] ->logits[1, S, 131072],present.*.
The prompt is the one VoxtralProcessor.apply_transcription_request builds: <s>[INST][BEGIN_AUDIO], 375 audio
embeddings per 30 s, then [/INST]lang:en[TRANSCRIBE] (the lang:xx part left out to let the model detect the
language). The token ids are in config.json. pipeline.py needs numpy, onnxruntime (or onnxruntime-gpu),
mistral-common (for the tekken tokenizer, tekken.json) and soundfile:
python int4/pipeline.py speech.wav --language en --cuda
import sys; sys.path.insert(0, "int4")
import pipeline
asr = pipeline.Voxtral("int4", "cuda")
print(asr(pipeline.load_audio("speech.wav"), language="fr"))
On CUDA, ortllm.py keeps the KV cache on the GPU and creates the sessions with enable_skip_layer_norm_strict_mode
= 1 (the builder's fused layer norms otherwise accumulate in fp16); set it too if you run decoder.onnx yourself.
Precisions. fp16/: decoder and audio encoder in fp16 (fp32 inputs and outputs for the encoder). int8/:
decoder 8-bit MatMulNBits (block 32, quantized from the fp16 build, fp16 activations), audio encoder weight-only
int8, int8 embeddings with an fp16 scale per row. int4/: decoder 4-bit MatMulNBits (block 32, symmetric RTN, from
the fp16 build), audio encoder 4-bit MatMulNBits on its MatMuls, int8 embeddings. These are GPU builds: on a laptop
CPU int4/ runs 2.7 times slower than real time, and the 8-bit MatMulNBits of int8/ has no fast CPU kernel. No
bf16 build.
Benchmark. 400 LibriSpeech test-clean utterances (the first 10 of 2-20 s from each of its 40 speakers), English
given, greedy decoding, at most 500 new tokens. WER with Whisper's English text normaliser. The original is
transformers' VoxtralForConditionalGeneration (5.23) in bf16, the checkpoint's own precision, on CUDA (fp32 does not
fit a 16 GB GPU). "Same transcript" counts clips read exactly as the original reads them. Pass (the
speech-recognition threshold used across these repositories): WER at most 1 point above the original's. Real-time
factor (processing time / audio time) on an RTX 3080 Laptop GPU, or with 4 threads of a laptop CPU, whole pipeline.
| Clips | WER | Same transcript as the original | Read differently | Real-time factor | |
|---|---|---|---|---|---|
| original (PyTorch bf16, CUDA) | 400 | 1.77 % | - | - | 0.23 |
fp16/ (CUDA) |
400 | 1.73 % | 378 | 0.07 % | 0.12 |
int8/ (CUDA) |
400 | 1.74 % | 367 | 0.12 % | 0.10 |
int4/ (CUDA) |
400 | 1.78 % | 297 | 0.48 % | 0.098 |
int4/ (CPU) |
first 30 | 1.45 % (original 1.27 %) | 18 | 0.91 % | 2.7 |
How these were made
From Voxtral Mini 3B at revision 3060fe34b3. The graphs were exported from the original PyTorch modules with
torch.onnx (opset 17) and checked against them; text decoders of audio-language models were built with
onnxruntime-genai's model builder (inputs_embeds in, KV cache in and out). Weight-only quantization: int8 is a
per-output-channel int8 weight behind DequantizeLinear (or 8-bit MatMulNBits in a text decoder), int4 is ONNX
Runtime's MatMulNBits (block 32). Each card above says which precision every graph uses. Every precision was
benchmarked against the original before upload (pass: error rate at most 1 point above the original's), and the
sha256 column above is what the files must hash to.
License
Apache-2.0. These files are a derivative of the original model and carry its license: it is copied in this repository, and it governs your use of these files. The original model card follows unchanged.
Original model card
Voxtral Mini 1.0 (3B) - 2507
Voxtral Mini is an enhancement of Ministral 3B, incorporating state-of-the-art audio input capabilities while retaining best-in-class text performance. It excels at speech transcription, translation and audio understanding.
Learn more about Voxtral in our blog post here and our research paper.
Key Features
Voxtral builds upon Ministral-3B with powerful audio understanding capabilities.
- Dedicated transcription mode: Voxtral can operate in a pure speech transcription mode to maximize performance. By default, Voxtral automatically predicts the source audio language and transcribes the text accordingly
- Long-form context: With a 32k token context length, Voxtral handles audios up to 30 minutes for transcription, or 40 minutes for understanding
- Built-in Q&A and summarization: Supports asking questions directly through audio. Analyze audio and generate structured summaries without the need for separate ASR and language models
- Natively multilingual: Automatic language detection and state-of-the-art performance in the world’s most widely used languages (English, Spanish, French, Portuguese, Hindi, German, Dutch, Italian)
- Function-calling straight from voice: Enables direct triggering of backend functions, workflows, or API calls based on spoken user intents
- Highly capable at text: Retains the text understanding capabilities of its language model backbone, Ministral-3B
Benchmark Results
Audio
Average word error rate (WER) over the FLEURS, Mozilla Common Voice and Multilingual LibriSpeech benchmarks:
Text
Usage
The model can be used with the following frameworks;
vllm (recommended): See hereTransformers🤗: See here
Notes:
temperature=0.2andtop_p=0.95for chat completion (e.g. Audio Understanding) andtemperature=0.0for transcription- Multiple audios per message and multiple user turns with audio are supported
- System prompts are not yet supported
vLLM (recommended)
We recommend using this model with vLLM.
Installation
Make sure to install vllm >= 0.10.0, we recommend using uv:
uv pip install -U "vllm[audio]" --system
Doing so should automatically install mistral_common >= 1.8.1.
To check:
python -c "import mistral_common; print(mistral_common.__version__)"
Offline
You can test that your vLLM setup works as expected by cloning the vLLM repo:
git clone https://github.com/vllm-project/vllm && cd vllm
and then running:
python examples/offline_inference/audio_language.py --num-audios 2 --model-type voxtral
Serve
We recommend that you use Voxtral-Small-24B-2507 in a server/client setting.
- Spin up a server:
vllm serve mistralai/Voxtral-Mini-3B-2507 --tokenizer_mode mistral --config_format mistral --load_format mistral
Note: Running Voxtral-Mini-3B-2507 on GPU requires ~9.5 GB of GPU RAM in bf16 or fp16.
- To ping the client you can use a simple Python snippet. See the following examples.
Audio Instruct
Leverage the audio capabilities of Voxtral-Mini-3B-2507 to chat.
Make sure that your client has mistral-common with audio installed:
pip install --upgrade mistral_common\[audio\]
Python snippet
from mistral_common.protocol.instruct.messages import TextChunk, AudioChunk, UserMessage, AssistantMessage, RawAudio
from mistral_common.audio import Audio
from huggingface_hub import hf_hub_download
from openai import OpenAI
# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://<your-server-host>:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
models = client.models.list()
model = models.data[0].id
obama_file = hf_hub_download("patrickvonplaten/audio_samples", "obama.mp3", repo_type="dataset")
bcn_file = hf_hub_download("patrickvonplaten/audio_samples", "bcn_weather.mp3", repo_type="dataset")
def file_to_chunk(file: str) -> AudioChunk:
audio = Audio.from_file(file, strict=False)
return AudioChunk.from_audio(audio)
text_chunk = TextChunk(text="Which speaker is more inspiring? Why? How are they different from each other?")
user_msg = UserMessage(content=[file_to_chunk(obama_file), file_to_chunk(bcn_file), text_chunk]).to_openai()
print(30 * "=" + "USER 1" + 30 * "=")
print(text_chunk.text)
print("\n\n")
response = client.chat.completions.create(
model=model,
messages=[user_msg],
temperature=0.2,
top_p=0.95,
)
content = response.choices[0].message.content
print(30 * "=" + "BOT 1" + 30 * "=")
print(content)
print("\n\n")
# The speaker who is more inspiring is the one who delivered the farewell address, as they express
# gratitude, optimism, and a strong commitment to the nation and its citizens. They emphasize the importance of
# self-government and active citizenship, encouraging everyone to participate in the democratic process. In contrast,
# the other speaker provides a factual update on the weather in Barcelona, which is less inspiring as it
# lacks the emotional and motivational content of the farewell address.
# **Differences:**
# - The farewell address speaker focuses on the values and responsibilities of citizenship, encouraging active participation in democracy.
# - The weather update speaker provides factual information about the temperature in Barcelona, without any emotional or motivational content.
messages = [
user_msg,
AssistantMessage(content=content).to_openai(),
UserMessage(content="Ok, now please summarize the content of the first audio.").to_openai()
]
print(30 * "=" + "USER 2" + 30 * "=")
print(messages[-1]["content"])
print("\n\n")
response = client.chat.completions.create(
model=model,
messages=messages,
temperature=0.2,
top_p=0.95,
)
content = response.choices[0].message.content
print(30 * "=" + "BOT 2" + 30 * "=")
print(content)
Transcription
Voxtral-Mini-3B-2507 has powerful transcription capabilities!
Make sure that your client has mistral-common with audio installed:
pip install --upgrade mistral_common\[audio\]
Python snippet
from mistral_common.protocol.transcription.request import TranscriptionRequest
from mistral_common.protocol.instruct.messages import RawAudio
from mistral_common.audio import Audio
from huggingface_hub import hf_hub_download
from openai import OpenAI
# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://<your-server-host>:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
models = client.models.list()
model = models.data[0].id
obama_file = hf_hub_download("patrickvonplaten/audio_samples", "obama.mp3", repo_type="dataset")
audio = Audio.from_file(obama_file, strict=False)
audio = RawAudio.from_audio(audio)
req = TranscriptionRequest(model=model, audio=audio, language="en", temperature=0.0).to_openai(exclude=("top_p", "seed"))
response = client.audio.transcriptions.create(**req)
print(response)
Transformers 🤗
Starting with transformers >= 4.54.0 and above, you can run Voxtral natively!
Install Transformers:
pip install -U transformers
Make sure to have mistral-common >= 1.8.1 installed with audio dependencies:
pip install --upgrade "mistral-common[audio]"
Audio Instruct
➡️ multi-audio + text instruction
from transformers import VoxtralForConditionalGeneration, AutoProcessor
import torch
device = "cuda"
repo_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map=device)
conversation = [
{
"role": "user",
"content": [
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/mary_had_lamb.mp3",
},
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/winning_call.mp3",
},
{"type": "text", "text": "What sport and what nursery rhyme are referenced?"},
],
}
]
inputs = processor.apply_chat_template(conversation)
inputs = inputs.to(device, dtype=torch.bfloat16)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print("\nGenerated response:")
print("=" * 80)
print(decoded_outputs[0])
print("=" * 80)
➡️ multi-turn
from transformers import VoxtralForConditionalGeneration, AutoProcessor
import torch
device = "cuda"
repo_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map=device)
conversation = [
{
"role": "user",
"content": [
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/obama.mp3",
},
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/bcn_weather.mp3",
},
{"type": "text", "text": "Describe briefly what you can hear."},
],
},
{
"role": "assistant",
"content": "The audio begins with the speaker delivering a farewell address in Chicago, reflecting on his eight years as president and expressing gratitude to the American people. The audio then transitions to a weather report, stating that it was 35 degrees in Barcelona the previous day, but the temperature would drop to minus 20 degrees the following day.",
},
{
"role": "user",
"content": [
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/winning_call.mp3",
},
{"type": "text", "text": "Ok, now compare this new audio with the previous one."},
],
},
]
inputs = processor.apply_chat_template(conversation)
inputs = inputs.to(device, dtype=torch.bfloat16)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print("\nGenerated response:")
print("=" * 80)
print(decoded_outputs[0])
print("=" * 80)
➡️ text only
from transformers import VoxtralForConditionalGeneration, AutoProcessor
import torch
device = "cuda"
repo_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map=device)
conversation = [
{
"role": "user",
"content": [
{
"type": "text",
"text": "Why should AI models be open-sourced?",
},
],
}
]
inputs = processor.apply_chat_template(conversation)
inputs = inputs.to(device, dtype=torch.bfloat16)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print("\nGenerated response:")
print("=" * 80)
print(decoded_outputs[0])
print("=" * 80)
➡️ audio only
from transformers import VoxtralForConditionalGeneration, AutoProcessor
import torch
device = "cuda"
repo_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map=device)
conversation = [
{
"role": "user",
"content": [
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/winning_call.mp3",
},
],
}
]
inputs = processor.apply_chat_template(conversation)
inputs = inputs.to(device, dtype=torch.bfloat16)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print("\nGenerated response:")
print("=" * 80)
print(decoded_outputs[0])
print("=" * 80)
➡️ batched inference
from transformers import VoxtralForConditionalGeneration, AutoProcessor
import torch
device = "cuda"
repo_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map=device)
conversations = [
[
{
"role": "user",
"content": [
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/obama.mp3",
},
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/bcn_weather.mp3",
},
{
"type": "text",
"text": "Who's speaking in the speach and what city's weather is being discussed?",
},
],
}
],
[
{
"role": "user",
"content": [
{
"type": "audio",
"path": "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/winning_call.mp3",
},
{"type": "text", "text": "What can you tell me about this audio?"},
],
}
],
]
inputs = processor.apply_chat_template(conversations)
inputs = inputs.to(device, dtype=torch.bfloat16)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print("\nGenerated responses:")
print("=" * 80)
for decoded_output in decoded_outputs:
print(decoded_output)
print("=" * 80)
Transcription
➡️ transcribe
from transformers import VoxtralForConditionalGeneration, AutoProcessor
import torch
device = "cuda"
repo_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map=device)
inputs = processor.apply_transcription_request(language="en", audio="https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/obama.mp3", model_id=repo_id)
inputs = inputs.to(device, dtype=torch.bfloat16)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print("\nGenerated responses:")
print("=" * 80)
for decoded_output in decoded_outputs:
print(decoded_output)
print("=" * 80)
Model tree for moonbox-dev/Voxtral-Mini-3B-2507-ONNX
Base model
mistralai/Voxtral-Mini-3B-2507
