Downloads · 30 days
13
5% of all-time downloads
MERaLiON-3-10B-TurboQuant-MLX-4bit
majentik/MERaLiON-3-10B-TurboQuant-MLX-4bit
MERaLiON-3-10B-TurboQuant-MLX-4bit is a automatic speech recognition model from majentik. Use it when you need speech turned into text. It is set up for mlx. The card lists the license as other.
[!TIP] KV-cache quantization without any fork (recommended, 2026): upstream llama.cpp/Ollama now cover this natively — use -ctk q80 -ctv q80 (~half KV memory, negligible quality loss: perplexity +0.002–0.05) or -ctk q…
All-time downloads
284
Public
Repo size
8 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors7.9 GB · 100%
At a glance
- Task
- Automatic Speech Recognition
- Library
- mlx
- License
- other
- Model type
- meralion3
- Access
- Public
- Created
- Apr 13, 2026
- Updated
- Jul 20, 2026
- SHA
- 78023d23
Base models
Model card
From the Hugging Face model README
<!-- kv-upstream-note -->[!TIP] KV-cache quantization without any fork (recommended, 2026): upstream llama.cpp/Ollama now cover this natively — use
-ctk q8_0 -ctv q8_0(~half KV memory, negligible quality loss: perplexity +0.002–0.05) or-ctk q4_0 -ctv q4_0(~quarter memory, ≈7.6% perplexity increase). In Ollama:OLLAMA_KV_CACHE_TYPE=q8_0withOLLAMA_FLASH_ATTENTION=1. Keep K and V types symmetric to stay on the fast fused Flash-Attention path. Since April 2026, mainline llama.cpp also applies Hadamard rotation to KV activations (PR #21038), which greatly improves low-bit KV quality (opt-out:LLAMA_ATTN_ROT_DISABLE=1).The RotorQuant/TurboQuant fork flow below is experimental/legacy: the TurboQuant llama.cpp PR was closed without merging (June 2026) and the fork is unmaintained relative to mainline. It is NOT required to use this model.
MERaLiON-3-10B-TurboQuant-MLX-4bit
4-bit weight-quantized MLX version of MERaLiON/MERaLiON-3-10B-preview with the legacy TurboQuant KV-cache fork (superseded by upstream llama.cpp KV options). Optimized for Apple Silicon inference via the MLX framework.
MERaLiON-3-10B is a multimodal audio-language model built on a Gemma-2 decoder backbone, designed for speech-to-text and audio understanding tasks.
Approximate model size: ~5 GB
Model Specifications
| Property | Value |
|---|---|
| Base Model | MERaLiON/MERaLiON-3-10B-preview |
| Parameters | ~10 billion |
| Architecture | Multimodal audio-language (Gemma-2 decoder backbone) |
| Modality | Audio + text input, text output |
| License | See base model |
| Weight Quantization | 4-bit (~5 GB) |
| KV-Cache Quantization | TurboQuant |
| Framework | MLX (Apple Silicon) |
Quickstart
from mlx_lm import load, generate
model, tokenizer = load("majentik/MERaLiON-3-10B-TurboQuant-MLX-4bit")
prompt = "Transcribe the following audio:"
response = generate(model, tokenizer, prompt=prompt, max_tokens=512)
print(response)
About the RotorQuant / TurboQuant labels
RotorQuant and TurboQuant are this project's release labels, not distinct
quantization algorithms — for any given tier, both brand repos carry
byte-identical weights produced with the standard MLX / llama.cpp quantizers.
No brand-specific speedup is claimed or measured. The KV-cache fork these
labels originally referred to is legacy; for KV-cache memory savings use the
upstream options described above (-ctk/-ctv q8_0, OLLAMA_KV_CACHE_TYPE).
KV-Cache Quantization Comparison
| Method | Prefill Speed | Decode Speed | Memory Savings | Reference |
|---|---|---|---|---|
| TurboQuant | Baseline | Baseline | High | arXiv: 2504.19874 |
Memory Estimates (MERaLiON-3-10B)
| Precision | Approximate Size | MLX Variant |
|---|---|---|
| FP16 (original) | ~20 GB | -- |
| 8-bit quantized | ~10 GB | TurboQuant-MLX-8bit |
| 4-bit quantized | ~5 GB | This model |
| 2-bit quantized | ~3 GB | TurboQuant-MLX-2bit |
Hardware Requirements
This model requires approximately 5 GB of unified memory. Recommended hardware:
- Apple M1 (8 GB+)
- Any Apple Silicon Mac
See Also
- MERaLiON/MERaLiON-3-10B-preview -- Base model
- majentik/MERaLiON-3-10B-TurboQuant-MLX-8bit -- MLX 8-bit variant
- majentik/MERaLiON-3-10B-TurboQuant-MLX-2bit -- MLX 2-bit variant
- majentik/MERaLiON-3-10B-RotorQuant-MLX-4bit -- RotorQuant MLX 4-bit variant
- TurboQuant Paper (arXiv: 2504.19874)
- MLX Framework
Quant trade-off (MLX lane)
| Bits | Approx size | Use case | Recommendation |
|---|---|---|---|
| 2-bit | ~2.6 GB | Aggressive quantization | Very low-RAM Macs |
| 3-bit | ~3.6 GB | Lossy but small | Low-RAM Macs |
| 4-bit | ~4.2 GB | Balanced default | Recommended for most Macs |
| 5-bit | ~5.0 GB | Higher fidelity | Quality-sensitive |
| 6-bit | ~6.0 GB | Approaching FP16 quality | High-fidelity |
| 8-bit | ~7.6 GB | Near-lossless reference | Fidelity-critical work |
(Current variant — 4bit — is bolded.)
Variants in this family
(Showing 8 sibling variants under majentik/meralion3-10b-*. The current variant — TurboQuant-MLX-4bit — is bolded.)
| Variant | Runtime | Approx size | Use case |
|---|---|---|---|
| TurboQuant-MLX-4bit | mlx-lm | ~6.2 GB | Apple Silicon balanced |
- Task
- Automatic Speech Recognition
- Library
- mlx
- Type
- meralion3
- License
- other
- Created
- Apr 13, 2026
- Updated
- Jul 20, 2026