Blinorot
/

AL-Whisper-Instruct-R

@@ -1,199 +1,144 @@
 ---
 library_name: transformers
-tags: []
 ---
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## Model Details
-### Model Description
-<!-- Provide a longer summary of what this model is. -->
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-### Model Sources [optional]
-<!-- Provide the basic links for the model. -->
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-## Uses
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-### Direct Use
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-[More Information Needed]
-### Downstream Use [optional]
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-[More Information Needed]
-### Out-of-Scope Use
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-[More Information Needed]
-## Bias, Risks, and Limitations
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-[More Information Needed]
-### Recommendations
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-## How to Get Started with the Model
-Use the code below to get started with the model.
-[More Information Needed]
-## Training Details
-### Training Data
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-[More Information Needed]
-### Training Procedure
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-#### Preprocessing [optional]
-[More Information Needed]
-#### Training Hyperparameters
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-#### Speeds, Sizes, Times [optional]
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-[More Information Needed]
-## Evaluation
-<!-- This section describes the evaluation protocols and provides the results. -->
-### Testing Data, Factors & Metrics
-#### Testing Data
-<!-- This should link to a Dataset Card if possible. -->
-[More Information Needed]
-#### Factors
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-[More Information Needed]
-#### Metrics
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-[More Information Needed]
-### Results
-[More Information Needed]
-#### Summary
-## Model Examination [optional]
-<!-- Relevant interpretability work for the model goes here -->
-[More Information Needed]
-## Environmental Impact
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-## Technical Specifications [optional]
-### Model Architecture and Objective
-[More Information Needed]
-### Compute Infrastructure
-[More Information Needed]
-#### Hardware
-[More Information Needed]
-#### Software
-[More Information Needed]
-## Citation [optional]
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-**BibTeX:**
-[More Information Needed]
-**APA:**
-[More Information Needed]
-## Glossary [optional]
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-[More Information Needed]
-## More Information [optional]
-[More Information Needed]
-## Model Card Authors [optional]
-[More Information Needed]
-## Model Card Contact
-[More Information Needed]

 ---
 library_name: transformers
+datasets:
+- Blinorot/ALARM-Corpora
+base_model:
+- Qwen/Qwen3-4B-Thinking-2507
 ---
+# Model Card for AL-Whisper-Instruct-R
+This is a checkpoint for AL-Whisper-Instruct-R, audio-understanding reasoning language model, proposed in [ALARM: Audio–Language Alignment for Reasoning Models](https://arxiv.org/abs/2603.09556).
+This model is trained only on the speech instruction subset of the data.
+Can be combined with [ALARM-CA](https://huggingface.co/Blinorot/ALARM-CA) to create ALARM-E model that achieves state-of-the-art performance (the best open-source, the third overall) on [MMSU](https://arxiv.org/abs/2506.04779) and [MMAU-Speech](https://arxiv.org/abs/2410.19168) benchmarks.
+For more details regarding the model and its usage, please refer to our [GitHub](https://github.com/Blinorot/ALARM).
+## Inference
+We provide [vLLM](https://github.com/vllm-project/vllm) support using [vLLM Prompt Embedding API](https://docs.vllm.ai/en/stable/features/prompt_embeds/).
+Since ALARM uses the frozen Qwen3 model as the backbone, `vllm` just runs the original Qwen3 checkpoint, and the ALARM checkpoint is used for extracting LLM input embeddings.
+After you cloned the repo and installed the depnedencies, you can run the pretrained model as follows:
+```python
+# Import libraries
+import os
+os.environ["CUDA_VISIBLE_DEVICES"] = "0" #optional
+# run before importing torch because generate_vllm sets the multiprocessing method
+from generate_vllm import get_response
+from src.model.wrapped_llms.qwen3 import Qwen3AudioWrappedFeatureExtractor
+from omegaconf import OmegaConf
+from torchaudio.utils import _download_asset
+from torchcodec.decoders import AudioDecoder
+from transformers import AutoTokenizer
+from vllm import LLM
+# The model configuration config.
+# Handles vllm-related configuration and defines feature extractors,
+# i.e., audio -> encoder input embedding conversion.
+# All other configuration, including model architecture, will be
+# loaded from the checkpoint.
+default_model_config_name = "src/configs/model/default_inference.yaml"
+model_config = OmegaConf.load(default_model_config_name)
+# checkpoint_name = which model to run
+#   Single model version (no inference-time ensemble):
+#   checkpoint_name='Blinorot/AL-Whisper-Instruct-R'
+#   ALARM-E embedding fusion-type version (inference-time ensemble):
+#   checkpoint_name=["Blinorot/ALARM-CA","Blinorot/AL-Whisper-Instruct-R"]
+checkpoint_name = "Blinorot/AL-Whisper-Instruct-R"
+device = "cuda"
+# Load Tokenizer for Text Processing
+tokenizer = AutoTokenizer.from_pretrained(model_config.llm)
+# Load ALARM/AL-*-R checkpoints for extraction of LLM input embeddings
+if isinstance(checkpoint_name, list): # ALARM-E-style embedding fusion (inference-time ensemble)
+    feature_extractor_list = []
+    for name in checkpoint_name:
+        # Load weights into the (audio,text)->LLM embeddings converter
+        feature_extractor = Qwen3AudioWrappedFeatureExtractor(
+            model_config=model_config,
+            checkpoint_name=name,
+            tokenizer=tokenizer,
+        )
+        feature_extractor.to(device)
+        feature_extractor_list.append(feature_extractor)
+    feature_extractor = feature_extractor_list
+else: # Single Model version (no inference-time ensemble)
+    # Load weights into the (audio,text)->LLM embeddings converter
+    feature_extractor = Qwen3AudioWrappedFeatureExtractor(
+        model_config=model_config,
+        checkpoint_name=checkpoint_name,
+        tokenizer=tokenizer,
+    )
+    feature_extractor.to(device)
+# Start the offline vLLM instance of original Qwen3 RLM
+# Model will be loaded to CUDA_VISIBLE_DEVICES id
+llm = LLM(
+    model_config.llm,
+    enable_prefix_caching=True,
+    max_model_len=model_config.max_model_len,
+    max_num_seqs=model_config.max_num_seq,
+    max_num_batched_tokens=model_config.max_num_batched_tokens,
+    gpu_memory_utilization=model_config.gpu_memory_utilization,
+    enable_prompt_embeds=True,
+)
+# Set sampling arguments for the RLM
+sample = llm.get_default_sampling_params()
+sample.seed = model_config.seed
+sample.max_tokens = model_config.max_tokens
+# Define audio and prompt
+# Audio must come from torchcodec.AudioDecoder
+audio_example_path = _download_asset("tutorial-assets/ctc-decoding/1688-142285-0007.wav")
+audio = AudioDecoder(audio_example_path)
+prompt = "Describe the audio content."
+# Define a system prompt
+system_prompt = "You are an audio-understanding model."
+# Obtain response from Audio RLM
+response = get_response(
+    prompts=[prompt], # list of all the prompts
+    audio_list=[audio], # list of corresponding audio
+    llm=llm,
+    feature_extractor=feature_extractor,
+    sample=sample,
+    tokenizer=tokenizer,
+    system_prompt=system_prompt,
+    max_thinking_tokens=model_config.max_thinking_tokens, # controls thinking budget for the RLM
+    debug=False,
+)
+# Response is a list of responses, one per each (prompt, audio) input pair
+# We have only one input pair, so the final response is at index 0
+response = response[0]
+print(f"Model response:\n\n{response}")
+```
+## Citation
+If you use this work, please cite:
+```bibtex
+@article{grinberg2026alarm,
+  title={ALARM: Audio-Language Alignment for Reasoning Models},
+  author={Grinberg, Petr and Shahmohammadi, Hassan},
+  journal={arXiv preprint arXiv:2603.09556},
+  year={2026}
+}
+```
+## License
+The model checkpoint is licensed under Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0).
+It may only be used for non-commercial research purposes.