LFM3 1.7B: VRAM requirements & speed
LFM3 1.7B (1.73B dense, lfm1.0). Native context 256k. Smallest GGUF: 1.1 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: LiquidAI/LFM3-1.7B-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx | +256k ctx |
|---|---|---|---|---|---|
| Q4_K_M | 1.1 GB | 1.2 GB | 1.5 GB | 2.7 GB | 4.3 GB |
| Q5_K_M | 1.26 GB | 1.4 GB | 1.7 GB | 2.9 GB | 4.5 GB |
| Q6_K | 1.43 GB | 1.5 GB | 1.8 GB | 3.0 GB | 4.7 GB |
| Q8_0 | 1.85 GB | 2.0 GB | 2.3 GB | 3.5 GB | 5.1 GB |
KV cache size by context length
Computed from the model's attention layout (sliding-window layers capped at their window).
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.10 GB | 0.06 GB | 0.03 GB |
| 32k | 0.41 GB | 0.22 GB | 0.11 GB |
| 128k | 1.61 GB | 0.86 GB | 0.45 GB |
| 256k | 3.22 GB | 1.71 GB | 0.91 GB |
Which hardware runs LFM3 1.7B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX PRO 6000 96GB | Q8_0 | fits fully | 2.1 GB | 465 | scaled from measured |
| RTX 5090 32GB | Q8_0 | fits fully | 2.1 GB | 463 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 2.1 GB | 411 | estimated |
| Mac Studio M5 Ultra 512GB (orderable since Oct 28 2026) | Q8_0 | fits fully | 2.1 GB | 411 | estimated |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 2.1 GB | 349 | estimated |
| RTX 4090 24GB | Q8_0 | fits fully | 2.1 GB | 325 | scaled from measured |
| RTX 5080 16GB | Q8_0 | fits fully | 2.1 GB | 320 | scaled from measured |
| AMD RX 7900 XTX 24GB | Q8_0 | fits fully | 2.1 GB | 315 | estimated |
| RTX 5070 Ti 16GB | Q8_0 | fits fully | 2.1 GB | 277 | scaled from measured |
| RTX 3090 24GB | Q8_0 | fits fully | 2.1 GB | 264 | scaled from measured |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 2.1 GB | 254 | estimated |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 2.1 GB | 254 | estimated |
| RTX 5070 12GB | Q8_0 | fits fully | 2.1 GB | 228 | scaled from measured |
| RTX 4070 12GB | Q8_0 | fits fully | 2.1 GB | 191 | scaled from measured |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 2.1 GB | 175 | estimated |
| RTX 5060 Ti 16GB | Q8_0 | fits fully | 2.1 GB | 159 | scaled from measured |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 2.1 GB | 140 | estimated |
| NVIDIA RTX Spark N1X 128GB | Q8_0 | fits fully | 2.1 GB | 129 | estimated |
| RTX 3060 12GB (32GB RAM) | Q8_0 | fits fully | 2.1 GB | 126 | scaled from measured |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q8_0 | fits fully | 2.1 GB | 122 | scaled from measured |
| AMD Ryzen AI Max+ PRO 495 (Ryzen AI Halo) 192GB | Q8_0 | fits fully | 2.1 GB | 118 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | Q8_0 | fits fully | 2.1 GB | 115 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q8_0 | fits fully | 2.1 GB | 114 | scaled from measured |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 2.1 GB | 110 | scaled from measured |
| NVIDIA DGX Spark 64GB | Q8_0 | fits fully | 2.1 GB | 110 | scaled from measured |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 2.1 GB | 110 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 2.1 GB | 83 | estimated |
| Mac mini M6 32GB | Q8_0 | fits fully | 2.1 GB | 58 | estimated |
| MacBook Pro 14 M6 32GB (ships Nov 6 2026) | Q8_0 | fits fully | 2.1 GB | 58 | estimated |
| MacBook Air M5 32GB | Q8_0 | fits fully | 2.1 GB | 56 | estimated |
| Mac mini / iMac M4 16GB | Q8_0 | fits fully | 2.1 GB | 50 | scaled from measured |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Fits (at least with offload) on every listed configuration.
MLX conversions for Apple Silicon
No MLX conversion yet: we found no verified mlx-community 4-bit or 8-bit repo for LFM3 1.7B. Use the GGUF files with llama.cpp or ollama on a Mac. How we check.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-11-02
Can it run? LFM3 1.7B on: 8GB VRAM class (RTX 4060 8GB, 32GB RAM) · Mac mini / iMac M4 16GB (all pairs)
FAQ
How much VRAM does LFM3 1.7B need?
The smallest quant (Q4_K_M) is 1.1 GB, so you need roughly 2 GB for an 8k context including KV cache and overhead. At 256k context the f16 KV cache alone is 3.2 GB.
Can I run LFM3 1.7B on a RTX PRO 6000 96GB?
Yes: on RTX PRO 6000 96GB it fits fully as Q8_0 at about 465 tok/s (scaled from measured). 31 of 31 hardware configurations fit it fully at 16k context.
What is the maximum context for LFM3 1.7B?
256k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.