Instructions to use diffuse-cpp/Dream-v0-Instruct-7B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use diffuse-cpp/Dream-v0-Instruct-7B-GGUF with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0 # Run inference directly in the terminal: llama cli -hf diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0 # Run inference directly in the terminal: llama cli -hf diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0 # Run inference directly in the terminal: ./build/bin/llama cli -hf diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
Use Docker
docker model run hf.co/diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
- Llama new
- LM Studio
- Jan
- vLLM
How to use diffuse-cpp/Dream-v0-Instruct-7B-GGUF with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "diffuse-cpp/Dream-v0-Instruct-7B-GGUF" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "diffuse-cpp/Dream-v0-Instruct-7B-GGUF", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
- Ollama
How to use diffuse-cpp/Dream-v0-Instruct-7B-GGUF with Ollama:
ollama run hf.co/diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
- Unsloth Desktop
- Docker Model Runner
How to use diffuse-cpp/Dream-v0-Instruct-7B-GGUF with Docker Model Runner:
docker model run hf.co/diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
- Lemonade
How to use diffuse-cpp/Dream-v0-Instruct-7B-GGUF with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull diffuse-cpp/Dream-v0-Instruct-7B-GGUF:Q8_0
Run and chat with the model
lemonade run user.Dream-v0-Instruct-7B-GGUF-Q8_0
List all available models
lemonade list
- Atomic Chat
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,92 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- diffusion
|
| 5 |
+
- masked-diffusion
|
| 6 |
+
- dream
|
| 7 |
+
- qwen2
|
| 8 |
+
- gguf
|
| 9 |
+
- diffuse-cpp
|
| 10 |
+
base_model: Dream-org/Dream-v0-Instruct-7B
|
| 11 |
+
pipeline_tag: text-generation
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# Dream-v0-Instruct-7B-GGUF
|
| 15 |
+
|
| 16 |
+
GGUF quantizations of [Dream-org/Dream-v0-Instruct-7B](https://huggingface.co/Dream-org/Dream-v0-Instruct-7B) for use with [diffuse-cpp](https://github.com/iafiscal1212/diffuse-cpp), a CPU inference engine for Diffusion Language Models.
|
| 17 |
+
|
| 18 |
+
Dream is a masked diffusion language model based on the Qwen2.5-7B backbone with bidirectional attention and Grouped Query Attention (GQA, 28 query heads / 4 KV heads).
|
| 19 |
+
|
| 20 |
+
## Available Quantizations
|
| 21 |
+
|
| 22 |
+
| File | Type | Size | Description |
|
| 23 |
+
|------|------|------|-------------|
|
| 24 |
+
| `dream-7b-f16.gguf` | F16 | ~15 GB | Full precision, best quality |
|
| 25 |
+
| `dream-7b-q8_0.gguf` | Q8_0 | ~8.2 GB | 8-bit quantization, near-lossless |
|
| 26 |
+
| `dream-7b-q4km.gguf` | Q4_K_M | ~5.0 GB | 4-bit mixed quantization, best quality/size ratio |
|
| 27 |
+
|
| 28 |
+
**Recommended:** Q4_K_M for most users. Q8_0 if you have enough RAM and want minimal quality loss.
|
| 29 |
+
|
| 30 |
+
## Performance
|
| 31 |
+
|
| 32 |
+
Benchmarked on diffuse-cpp with entropy_exit + inter-step KV cache, 12 threads, seed=42:
|
| 33 |
+
|
| 34 |
+
| Prompt | tok/s | Steps | vs llama.cpp |
|
| 35 |
+
|--------|-------|-------|-------------|
|
| 36 |
+
| Capital of France? | 21.6 | 2 | 2.5x |
|
| 37 |
+
| Translate to French | 14.3 | 6 | 1.7x |
|
| 38 |
+
| 15 x 23? | 21.6 | 2 | 2.5x |
|
| 39 |
+
| Translate to Spanish | 13.2 | 10 | 1.6x |
|
| 40 |
+
| Python is_prime() | 8.2 | 7 | 1.0x |
|
| 41 |
+
| Why sky blue? | 4.9 | 16 | 0.6x |
|
| 42 |
+
| List planets | 4.9 | 16 | 0.6x |
|
| 43 |
+
| Poem about ocean | 4.5 | 16 | 0.5x |
|
| 44 |
+
| **Average** | **11.6** | | **1.4x** |
|
| 45 |
+
|
| 46 |
+
- Easy prompts (factual, math): **14-22 tok/s** (1.6-2.5x faster than llama.cpp)
|
| 47 |
+
- Hard prompts (creative, long-form): **4.5-4.9 tok/s**
|
| 48 |
+
- llama.cpp baseline: 8.51 tok/s (Qwen2.5-7B-Instruct, Q4_K_M, same hardware)
|
| 49 |
+
|
| 50 |
+
## Usage
|
| 51 |
+
|
| 52 |
+
```bash
|
| 53 |
+
# Download
|
| 54 |
+
huggingface-cli download diffuse-cpp/Dream-v0-Instruct-7B-GGUF dream-7b-q4km.gguf
|
| 55 |
+
|
| 56 |
+
# Run (requires diffuse-cpp v0.2.0+)
|
| 57 |
+
./diffuse-cli -m dream-7b-q4km.gguf -p "What is the capital of France?" -n 64 -s 16
|
| 58 |
+
```
|
| 59 |
+
|
| 60 |
+
## Model Details
|
| 61 |
+
|
| 62 |
+
- **Architecture:** Qwen2.5-7B backbone with bidirectional attention
|
| 63 |
+
- **Parameters:** 7.62B
|
| 64 |
+
- **Layers:** 28
|
| 65 |
+
- **Hidden size:** 3584
|
| 66 |
+
- **Attention:** GQA (28 query heads, 4 KV heads, head dim 128)
|
| 67 |
+
- **FFN:** SwiGLU, intermediate size 18944
|
| 68 |
+
- **Vocabulary:** 152,064 tokens
|
| 69 |
+
- **RoPE theta:** 1,000,000
|
| 70 |
+
- **Mask token ID:** 151666
|
| 71 |
+
- **Training:** Masked diffusion on text, with autoregressive logit shift
|
| 72 |
+
|
| 73 |
+
## Conversion Details
|
| 74 |
+
|
| 75 |
+
Converted from SafeTensors using `convert-dream.py` from diffuse-cpp:
|
| 76 |
+
- 339 tensors total (255 weights + 84 QKV biases)
|
| 77 |
+
- QKV biases kept at F32 in all quantizations
|
| 78 |
+
- Edge layers (first/last) quantized to Q6_K in Q4_K_M scheme
|
| 79 |
+
|
| 80 |
+
## Citation
|
| 81 |
+
|
| 82 |
+
```bibtex
|
| 83 |
+
@misc{dream2025,
|
| 84 |
+
title={Dream 7B - Scalable Discrete Denoising Diffusion Models for Text Generation},
|
| 85 |
+
author={Ye, Jiacheng and others},
|
| 86 |
+
year={2025}
|
| 87 |
+
}
|
| 88 |
+
```
|
| 89 |
+
|
| 90 |
+
## License
|
| 91 |
+
|
| 92 |
+
Apache 2.0, following the original Dream model license.
|