VE

Veena

VeenaGGUF by MungertUpdated 2025-09-24
Download on Hugging Face
Parameters3.78B
Context128K
Layers28
Architecturellama
Smallest quant1.3 GB
Quick answer

Veena needs about 3.9 GB of memory at Q4_K_M (a 2.5 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 31 quants, from 1.3 GB (TQ1_0) to 7.6 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 3.8 GB.

3B parameter Llama‑based TTS model for Hindi and English

Veena is a 3.78B parameter autoregressive transformer built on the Llama architecture. It synthesizes speech in Hindi and English at 24kHz using the SNAC codec. The model is released under Apache‑2.0.

Good at

  • High‑quality Hindi and English TTS
  • Low‑latency inference
  • Code‑mixed language support

Limits

  • No vision support
  • Large context size requires significant VRAM
  • Requires llama.cpp for inference

Best for

  • Embedded TTS in local devices
  • Rapid prototyping of multilingual voice assistants
  • Audio generation for accessibility tools

Who it’s for. Developers and researchers needing a compact, open‑source TTS model for Indian languages. The model is suitable for local deployment on machines with limited GPU memory.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
TQ1_01.3 GB 2.7 GB
Fits Get file →
TQ2_01.4 GB 2.8 GB
Fits Get file →
IQ1_S1.4 GB 2.8 GB
Fits Get file →
IQ1_M1.5 GB 2.9 GB
Fits Get file →
IQ2_XXS1.5 GB 2.9 GB
Fits Get file →
IQ2_XS1.6 GB 3.0 GB
Fits Get file →
IQ2_S1.6 GB 3.1 GB
Fits Get file →
IQ2_M1.7 GB 3.1 GB
Fits Get file →
Q2_K_S1.7 GB 3.1 GB
Fits Get file →
IQ3_XXS1.8 GB 3.2 GB
Fits Get file →
Q2_K_M1.8 GB 3.3 GB
Fits Get file →
IQ3_XS1.8 GB 3.3 GB
Fits Get file →
IQ3_M2.0 GB 3.4 GB
Fits Get file →
IQ3_S2.0 GB 3.4 GB
Fits Get file →
Q3_K_S2.0 GB 3.5 GB
Fits Get file →
Q4_02.1 GB 3.6 GB
Fits Get file →
Q3_K_M2.1 GB 3.6 GB
Fits Get file →
IQ4_XS2.2 GB 3.6 GB
Fits Get file →
IQ4_NL2.3 GB 3.7 GB
Fits Get file →
Q4_12.4 GB 3.8 GB
Fits Get file →
Q4_K_S2.4 GB 3.9 GB
Fits Get file →
Q4_K_M2.5 GB 3.9 GB
Fits Get file →
Q5_02.6 GB 4.0 GB
Fits Get file →
Q5_K_S2.8 GB 4.2 GB
Fits Get file →
Q5_K_M2.8 GB 4.2 GB
Fits Get file →
Q5_12.8 GB 4.3 GB
Fits Get file →
Q6_K_M3.1 GB 4.6 GB
Fits Get file →
Q8_04.0 GB 5.5 GB
Fits Get file →
BF165.9 GB 7.4 GB
Tight Get file →
F165.9 GB 7.4 GB
Tight Get file →
BF167.6 GB 9.0 GB
Too big Get file →

For 8 GB VRAM, use Q4_0 (file size 2.14 GB, estimated 3.6 GB memory). For 12 GB VRAM, Q5_0 (2.61 GB, 4 GB memory) or Q4_K_M (2.47 GB, 3.9 GB memory). For 16 GB VRAM, Q6_K_M (3.11 GB, 4.6 GB memory) or Q8_0 (4.03 GB, 5.5 GB memory).

What GPU do you need for Veena?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) 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 Q5_0 (2.6 GB file, about 4.0 GB in use).
  • 6 GB VRAM: the best quant that fits is Q8_0 (4.0 GB file, about 5.5 GB in use).
  • 8 GB VRAM: the best quant that fits is BF16 (5.9 GB file, about 7.4 GB in use).
  • 12 GB VRAM: the best quant that fits is BF16 (7.6 GB file, about 9.0 GB in use).

Which Veena quant should you choose?

Q4_K_M is the usual starting point for Veena: it balances file size and quality well. Here it is a 2.5 GB file.

The Q8_0 file is 4.0 GB, about 1.6 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 TQ1_0 at 1.3 GB, with the quality trade-offs that come with fewer bits. Lower bit counts mean smaller files and more quality loss.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters3.78B
Layers28
Trained context131,072 tokens
KV cache (fp16)112 KB per token

Source

Quantized byMungert
FamilyVeena
Licenseapache-2.0
Downloads (HF)1,795
Last updated2025-09-24

Frequently asked questions about Veena

How much VRAM does Veena need?

About 3.9 GB at Q4_K_M with an 8K context window. That is the 2.5 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 Veena on an 8 GB GPU?

Yes, with BF16: the 5.9 GB file needs about 7.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Veena GGUF?

It depends on the quant: TQ1_0 is 1.3 GB, TQ2_0 is 1.4 GB, IQ1_S is 1.4 GB, IQ1_M is 1.5 GB, IQ2_XXS is 1.5 GB, IQ2_XS is 1.6 GB and 25 more.

What context length does Veena support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.9 GB at 8K and 3.8 GB at 32K.

Where can I download Veena GGUF?

The files are on Hugging Face in the Mungert/Veena-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 Veena 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 Veena on 8 GB VRAM?

Yes, the Q4_0 quantization requires about 3.6 GB VRAM at 8k context, which fits within 8 GB.

Does Veena support other Indian languages?

The model is trained for Hindi and English; other Indian languages are not supported.

Similar models

Browse all →

More from Mungert

All by Mungert →

More Veena models

All Veena →

Other small models

All small models →