CSM 1B needs about 1.4 GB of memory at Q4_K_M (a 0.8 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 22 quants, from 0.6 GB (Q2_K) to 5.0 GB (F32).
CSM 1B – 0.457B parameter conversational English speech model
CSM 1B is a 0.457B parameter conversational English speech model. It is available in GGUF format and can be run with llama.cpp or codec.cpp. The model is licensed under apache-2.0.
Good at
- Conversational English speech
- Text‑to‑speech generation
- Voice cloning (via codec)
Limits
- No vision support
- Requires one model file plus one codec file
- Large VRAM needed for higher quantizations
Best for
- Offline TTS on CPU or GPU
- Voice cloning for custom voices
- Embedded speech applications
Who it’s for. Developers building offline text‑to‑speech systems or voice cloning solutions. The model is suitable for projects that need a small, efficient speech backbone with a separate codec for audio decoding.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 0.6 GB | 1.2 GB | Fits | Get file → | |
| Q3_K_S | 0.6 GB | 1.2 GB | Fits | Get file → | |
| Q3_K_M | 0.7 GB | 1.3 GB | Fits | Get file → | |
| Q3_K_L | 0.7 GB | 1.3 GB | Fits | Get file → | |
| Q4_0 | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q4_K_S | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q4_K_M | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q4_K_M | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q5_K_M | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q4_1 | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q8_0 | 0.8 GB | 1.4 GB | Fits | Get file → | |
| Q5_0 | 0.9 GB | 1.5 GB | Fits | Get file → | |
| Q5_K_S | 0.9 GB | 1.5 GB | Fits | Get file → | |
| Q5_K_M | 0.9 GB | 1.5 GB | Fits | Get file → | |
| F16 | 0.9 GB | 1.5 GB | Fits | Get file → | |
| Q5_1 | 1.0 GB | 1.6 GB | Fits | Get file → | |
| Q6_K | 1.0 GB | 1.6 GB | Fits | Get file → | |
| F32 | 1.2 GB | 1.8 GB | Fits | Get file → | |
| Q8_0 | 1.3 GB | 2.0 GB | Fits | Get file → | |
| BF16 | 2.5 GB | 3.3 GB | Fits | Get file → | |
| F16 | 2.5 GB | 3.3 GB | Fits | Get file → | |
| F32 | 5.0 GB | 6.0 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K_M (0.8 GB, 1.4 GB memory) or Q5_K_M (0.91 GB, 1.5 GB memory). For 4 GB VRAM, choose Q3_K_S (0.64 GB, 1.2 GB memory). Higher quantizations require more VRAM and CPU memory.
What GPU do you need for CSM 1B?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is F16 (2.5 GB file, about 3.3 GB in use).
- 6 GB VRAM: the best quant that fits is F32 (5.0 GB file, about 6.0 GB in use).
Which CSM 1B quant should you choose?
Q4_K_M is the usual starting point for CSM 1B: it balances file size and quality well. Here it is a 0.8 GB file.
The Q8_0 file is 0.8 GB, about 1.0 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is Q2_K at 0.6 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about CSM 1B
How much VRAM does CSM 1B need?
About 1.4 GB at Q4_K_M with an 8K context window. That is the 0.8 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run CSM 1B on an 8 GB GPU?
Yes, with F32: the 5.0 GB file needs about 6.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of CSM 1B GGUF?
It depends on the quant: Q2_K is 0.6 GB, Q3_K_S is 0.6 GB, Q3_K_M is 0.7 GB, Q3_K_L is 0.7 GB, Q4_0 is 0.8 GB, Q4_K_S is 0.8 GB and 16 more.
Where can I download CSM 1B GGUF?
The files are on Hugging Face in the BricksDisplay/CSM-1B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does CSM 1B use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run CSM 1B on 8 GB VRAM?
Yes, using Q4_K_M or Q5_K_M quantizations will fit within 8 GB VRAM, with estimated memory usage around 1.4–1.5 GB at 8k context.
Does CSM 1B support voice cloning?
Yes, the codec file includes voice cloning capabilities. Clone a voice with the ref_audio parameter in tts-cli or codec_cpp.