Add files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text

.python-version

@@ -0,0 +1 @@
+3.12

README.md

@@ -1,10 +1,11 @@
 ---
 title: Chroma 4B
-emoji: 👀
+emoji: 🎙️
 colorFrom: indigo
-colorTo: red
+colorTo: purple
 sdk: gradio
 sdk_version: 6.11.0
+python_version: "3.12"
 app_file: app.py
 pinned: false
 ---

app.py

@@ -0,0 +1,171 @@
+import gradio as gr
+import numpy as np
+import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoProcessor
+
+MODEL_ID = "FlashLabs/Chroma-4B"
+SAMPLE_RATE = 24_000
+SYSTEM_PROMPT = (
+    "You are Chroma, an advanced virtual human created by FlashLabs. "
+    "You possess the ability to understand auditory inputs and generate both text and speech."
+)
+
+processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    trust_remote_code=True,
+    dtype=torch.bfloat16,
+    device_map="auto",
+    attn_implementation="eager",
+)
+
+
+@spaces.GPU(duration=120)
+def generate(
+    text_input: str,
+    audio_input: str | None,
+    reference_audio: str | None,
+    reference_text: str,
+    max_new_tokens: int,
+    temperature: float,
+    top_p: float,
+) -> tuple[int, np.ndarray]:
+    if not reference_audio:
+        raise gr.Error("Reference audio is required.")
+    if not reference_text or not reference_text.strip():
+        raise gr.Error("Reference text is required.")
+
+    content = []
+    if text_input and text_input.strip():
+        content.append({"type": "text", "text": text_input.strip()})
+    if audio_input is not None:
+        content.append({"type": "audio", "audio": audio_input})
+    if not content:
+        raise gr.Error("Please provide text or audio input.")
+
+    conversation = [
+        [
+            {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT}]},
+            {"role": "user", "content": content},
+        ]
+    ]
+
+    inputs = processor(
+        conversation,
+        prompt_audio=[reference_audio],
+        prompt_text=[reference_text.strip()],
+        add_generation_prompt=True,
+        tokenize=False,
+    )
+    inputs = {
+        k: v.to(device=model.device, dtype=model.dtype)
+        if isinstance(v, torch.Tensor) and v.is_floating_point()
+        else v.to(model.device)
+        if isinstance(v, torch.Tensor)
+        else v
+        for k, v in inputs.items()
+    }
+
+    output = model.generate(
+        **inputs,
+        max_new_tokens=max_new_tokens,
+        do_sample=True,
+        temperature=temperature,
+        top_p=top_p,
+        use_cache=True,
+    )
+
+    audio_values = model.codec_model.decode(output.permute(0, 2, 1)).audio_values
+    audio_np = audio_values[0].cpu().detach().float().numpy()
+    if audio_np.ndim > 1:
+        audio_np = audio_np.squeeze()
+
+    return (SAMPLE_RATE, audio_np)
+
+
+with gr.Blocks() as demo:
+    gr.Markdown(
+        "# Chroma 4B\n"
+        "Multimodal speech generation model by "
+        "[FlashLabs](https://huggingface.co/FlashLabs/Chroma-4B)"
+    )
+
+    with gr.Row():
+        with gr.Column():
+            text_input = gr.Textbox(
+                label="Text input", lines=3, placeholder="Type your message..."
+            )
+            audio_input = gr.Audio(
+                label="Audio input", type="filepath", sources=["microphone", "upload"]
+            )
+            reference_audio = gr.Audio(
+                label="Reference audio (required)",
+                type="filepath",
+                sources=["upload"],
+            )
+            reference_text = gr.Textbox(
+                label="Reference text (required)",
+                placeholder="Transcript of the reference audio...",
+            )
+            with gr.Accordion("Generation settings", open=False):
+                max_new_tokens = gr.Slider(
+                    label="Max new tokens",
+                    minimum=10,
+                    maximum=500,
+                    step=10,
+                    value=256,
+                )
+                temperature = gr.Slider(
+                    label="Temperature",
+                    minimum=0.1,
+                    maximum=2.0,
+                    step=0.1,
+                    value=0.7,
+                )
+                top_p = gr.Slider(
+                    label="Top-p",
+                    minimum=0.1,
+                    maximum=1.0,
+                    step=0.05,
+                    value=0.9,
+                )
+            btn = gr.Button("Generate")
+
+        with gr.Column():
+            audio_output = gr.Audio(label="Generated speech", autoplay=True)
+
+    btn.click(
+        fn=generate,
+        inputs=[
+            text_input,
+            audio_input,
+            reference_audio,
+            reference_text,
+            max_new_tokens,
+            temperature,
+            top_p,
+        ],
+        outputs=audio_output,
+    )
+
+    gr.Examples(
+        examples=[
+            [
+                "Tell me a fun fact about space.",
+                None,
+                "assets/peter-piper.wav",
+                "Peter Piper picked a peck of pickled peppers. A peck of pickled peppers Peter Piper picked. If Peter Piper picked a peck of pickled peppers, where's the peck of pickled peppers Peter Piper picked?",
+            ],
+            [
+                "What's the weather like today?",
+                None,
+                "assets/seashells.wav",
+                "She sells seashells by the seashore. The shells she sells are surely seashells, so if she sells shells on the seashore, I'm sure she sells seashore shells.",
+            ],
+        ],
+        inputs=[text_input, audio_input, reference_audio, reference_text],
+    )
+
+if __name__ == "__main__":
+    demo.launch()

assets/peter-piper.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e700f0918c28b292a09b58f862a033e26e6f050f7e8ea42cb3323349220a0273
+size 615644

assets/seashells.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b557afe5a1ee56012907b4583439b3e1ad6a3b93a6ebeed361bdf179f68a7571
+size 452444

pyproject.toml

@@ -0,0 +1,26 @@
+[project]
+name = "chroma-4b"
+version = "0.1.0"
+description = "Add your description here"
+readme = "README.md"
+requires-python = ">=3.12"
+dependencies = [
+    "accelerate",
+    "av>=17.0.0",
+    "gradio>=6.11.0",
+    "librosa>=0.11.0",
+    "spaces",
+    "torch==2.9.1",
+    "torchaudio==2.9.1",
+    "torchcodec==0.9.1",
+    "torchvision>=0.24.1",
+    "transformers==5.0.0rc1",
+]
+
+[dependency-groups]
+dev = [
+    "ruff>=0.15.9",
+]
+hf-spaces = [
+    "datasets>=4.8.4",
+]

requirements.txt

@@ -0,0 +1,384 @@
+# This file was autogenerated by uv via the following command:
+#    uv export --no-hashes --no-dev --group hf-spaces --no-emit-package typer-slim --no-emit-package spaces -o requirements.txt
+accelerate==1.13.0
+    # via chroma-4b
+aiofiles==24.1.0
+    # via gradio
+aiohappyeyeballs==2.6.1
+    # via aiohttp
+aiohttp==3.13.5
+    # via fsspec
+aiosignal==1.4.0
+    # via aiohttp
+annotated-doc==0.0.4
+    # via
+    #   fastapi
+    #   typer
+annotated-types==0.7.0
+    # via pydantic
+anyio==4.13.0
+    # via
+    #   gradio
+    #   httpx
+    #   starlette
+attrs==26.1.0
+    # via aiohttp
+audioop-lts==0.2.2 ; python_full_version >= '3.13'
+    # via
+    #   gradio
+    #   standard-aifc
+    #   standard-sunau
+audioread==3.1.0
+    # via librosa
+av==17.0.0
+    # via chroma-4b
+brotli==1.2.0
+    # via gradio
+certifi==2026.2.25
+    # via
+    #   httpcore
+    #   httpx
+    #   requests
+cffi==2.0.0
+    # via soundfile
+charset-normalizer==3.4.7
+    # via requests
+click==8.3.2
+    # via
+    #   typer
+    #   uvicorn
+colorama==0.4.6 ; sys_platform == 'win32'
+    # via
+    #   click
+    #   tqdm
+datasets==4.8.4
+decorator==5.2.1
+    # via librosa
+dill==0.4.1
+    # via
+    #   datasets
+    #   multiprocess
+fastapi==0.135.3
+    # via gradio
+ffmpy==1.0.0
+    # via gradio
+filelock==3.25.2
+    # via
+    #   datasets
+    #   huggingface-hub
+    #   torch
+    #   transformers
+frozenlist==1.8.0
+    # via
+    #   aiohttp
+    #   aiosignal
+fsspec==2026.2.0
+    # via
+    #   datasets
+    #   gradio-client
+    #   huggingface-hub
+    #   torch
+gradio==6.11.0
+    # via
+    #   chroma-4b
+    #   spaces
+gradio-client==2.4.0
+    # via
+    #   gradio
+    #   hf-gradio
+groovy==0.1.2
+    # via gradio
+h11==0.16.0
+    # via
+    #   httpcore
+    #   uvicorn
+hf-gradio==0.3.0
+    # via gradio
+hf-xet==1.4.3 ; platform_machine == 'AMD64' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64'
+    # via huggingface-hub
+httpcore==1.0.9
+    # via httpx
+httpx==0.28.1
+    # via
+    #   datasets
+    #   gradio
+    #   gradio-client
+    #   huggingface-hub
+    #   safehttpx
+    #   spaces
+huggingface-hub==1.9.2
+    # via
+    #   accelerate
+    #   datasets
+    #   gradio
+    #   gradio-client
+    #   tokenizers
+    #   transformers
+idna==3.11
+    # via
+    #   anyio
+    #   httpx
+    #   requests
+    #   yarl
+jinja2==3.1.6
+    # via
+    #   gradio
+    #   torch
+joblib==1.5.3
+    # via
+    #   librosa
+    #   scikit-learn
+lazy-loader==0.5
+    # via librosa
+librosa==0.11.0
+    # via chroma-4b
+llvmlite==0.47.0
+    # via numba
+markdown-it-py==4.0.0
+    # via rich
+markupsafe==3.0.3
+    # via
+    #   gradio
+    #   jinja2
+mdurl==0.1.2
+    # via markdown-it-py
+mpmath==1.3.0
+    # via sympy
+msgpack==1.1.2
+    # via librosa
+multidict==6.7.1
+    # via
+    #   aiohttp
+    #   yarl
+multiprocess==0.70.19
+    # via datasets
+networkx==3.6.1
+    # via torch
+numba==0.65.0
+    # via librosa
+numpy==2.4.4
+    # via
+    #   accelerate
+    #   datasets
+    #   gradio
+    #   librosa
+    #   numba
+    #   pandas
+    #   scikit-learn
+    #   scipy
+    #   soundfile
+    #   soxr
+    #   torchvision
+    #   transformers
+nvidia-cublas-cu12==12.8.4.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via
+    #   nvidia-cudnn-cu12
+    #   nvidia-cusolver-cu12
+    #   torch
+nvidia-cuda-cupti-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cuda-nvrtc-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cuda-runtime-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cudnn-cu12==9.10.2.21 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cufft-cu12==11.3.3.83 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cufile-cu12==1.13.1.3 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-curand-cu12==10.3.9.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cusolver-cu12==11.7.3.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-cusparse-cu12==12.5.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via
+    #   nvidia-cusolver-cu12
+    #   torch
+nvidia-cusparselt-cu12==0.7.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nccl-cu12==2.27.5 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nvjitlink-cu12==12.8.93 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via
+    #   nvidia-cufft-cu12
+    #   nvidia-cusolver-cu12
+    #   nvidia-cusparse-cu12
+    #   torch
+nvidia-nvshmem-cu12==3.3.20 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+nvidia-nvtx-cu12==12.8.90 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+orjson==3.11.8
+    # via gradio
+packaging==26.0
+    # via
+    #   accelerate
+    #   datasets
+    #   gradio
+    #   gradio-client
+    #   huggingface-hub
+    #   lazy-loader
+    #   pooch
+    #   spaces
+    #   transformers
+pandas==3.0.2
+    # via
+    #   datasets
+    #   gradio
+pillow==12.2.0
+    # via
+    #   gradio
+    #   torchvision
+platformdirs==4.9.6
+    # via pooch
+pooch==1.9.0
+    # via librosa
+propcache==0.4.1
+    # via
+    #   aiohttp
+    #   yarl
+psutil==5.9.8
+    # via
+    #   accelerate
+    #   spaces
+pyarrow==23.0.1
+    # via datasets
+pycparser==3.0 ; implementation_name != 'PyPy'
+    # via cffi
+pydantic==2.12.5
+    # via
+    #   fastapi
+    #   gradio
+    #   spaces
+pydantic-core==2.41.5
+    # via pydantic
+pydub==0.25.1
+    # via gradio
+pygments==2.20.0
+    # via rich
+python-dateutil==2.9.0.post0
+    # via pandas
+python-multipart==0.0.24
+    # via gradio
+pytz==2026.1.post1
+    # via gradio
+pyyaml==6.0.3
+    # via
+    #   accelerate
+    #   datasets
+    #   gradio
+    #   huggingface-hub
+    #   transformers
+regex==2026.4.4
+    # via transformers
+requests==2.33.1
+    # via
+    #   datasets
+    #   pooch
+    #   spaces
+    #   transformers
+rich==14.3.3
+    # via typer
+safehttpx==0.1.7
+    # via gradio
+safetensors==0.7.0
+    # via
+    #   accelerate
+    #   transformers
+scikit-learn==1.8.0
+    # via librosa
+scipy==1.17.1
+    # via
+    #   librosa
+    #   scikit-learn
+semantic-version==2.10.0
+    # via gradio
+setuptools==82.0.1
+    # via torch
+shellingham==1.5.4
+    # via typer
+six==1.17.0
+    # via python-dateutil
+soundfile==0.13.1
+    # via librosa
+soxr==1.0.0
+    # via librosa
+standard-aifc==3.13.0 ; python_full_version >= '3.13'
+    # via
+    #   audioread
+    #   librosa
+standard-chunk==3.13.0 ; python_full_version >= '3.13'
+    # via standard-aifc
+standard-sunau==3.13.0 ; python_full_version >= '3.13'
+    # via
+    #   audioread
+    #   librosa
+starlette==1.0.0
+    # via
+    #   fastapi
+    #   gradio
+sympy==1.14.0
+    # via torch
+threadpoolctl==3.6.0
+    # via scikit-learn
+tokenizers==0.22.2
+    # via transformers
+tomlkit==0.13.3
+    # via gradio
+torch==2.9.1
+    # via
+    #   accelerate
+    #   chroma-4b
+    #   torchaudio
+    #   torchvision
+torchaudio==2.9.1
+    # via chroma-4b
+torchcodec==0.9.1
+    # via chroma-4b
+torchvision==0.24.1
+    # via chroma-4b
+tqdm==4.67.3
+    # via
+    #   datasets
+    #   huggingface-hub
+    #   transformers
+transformers==5.0.0rc1
+    # via chroma-4b
+triton==3.5.1 ; platform_machine == 'x86_64' and sys_platform == 'linux'
+    # via torch
+typer==0.24.1
+    # via
+    #   gradio
+    #   hf-gradio
+    #   huggingface-hub
+    #   typer-slim
+typing-extensions==4.15.0
+    # via
+    #   aiosignal
+    #   anyio
+    #   fastapi
+    #   gradio
+    #   gradio-client
+    #   huggingface-hub
+    #   librosa
+    #   pydantic
+    #   pydantic-core
+    #   spaces
+    #   starlette
+    #   torch
+    #   typing-inspection
+typing-inspection==0.4.2
+    # via
+    #   fastapi
+    #   pydantic
+tzdata==2026.1 ; sys_platform == 'emscripten' or sys_platform == 'win32'
+    # via pandas
+urllib3==2.6.3
+    # via requests
+uvicorn==0.44.0
+    # via gradio
+xxhash==3.6.0
+    # via datasets
+yarl==1.23.0
+    # via aiohttp