We are releasing MiniCPM5-2B, the second model in the MiniCPM5 series, following MiniCPM5-1B. It is a dense 2B Transformer that scales up the same training recipe, built for on-device, local deployment, and resource-constrained scenarios, reaching 2B-class open-source SOTA.
LFM3-1.7B-Base
LiquidAI·Base model·released 26 Oct 2026
ollama run lfm3:1.7bLFM3-1.7B-Base is the pre-trained checkpoint behind LFM3-1.7B, a text model for local agents and specialized applications based on our new tri-hybrid architecture.
How much VRAM does LFM3-1.7B-Base need?
At Q4_K_M LFM3-1.7B-Base needs about 1.9 GB with an 8K context (1.9 GB at Q4, 4.3 GB at full precision). Pick your GPU to see what fits and how fast it runs.
Needs: ≈1.9 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.
Advanced: context 8K · KV cache 16-bit
Smallest GPU for each quant and context (KV cache 16-bit)
| Quantisation | 8K | 16K | 32K | 64K | 128K | 256K |
|---|---|---|---|---|---|---|
| Q3_K_M | ||||||
| Q4_K_M | ||||||
| Q5_K_M | ||||||
| Q6_K | ||||||
| Q8_0 | ||||||
| FP16 |
≤16 GB card24–32 GB48–96 GBmore
| Quantisation | Total |
|---|---|
| Q2_KHeavy lossSmallest, noticeably worse quality | ≈1.6 GB |
| Q3_K_MNoticeable lossSmall, some quality loss | ≈1.8 GB |
| Q4_K_MrecommendedSmall lossRecommended default for local use | ≈1.9 GB |
| Q5_K_MNear-losslessNear-lossless for most tasks | ≈2.1 GB |
| Q6_KNear-losslessPractically lossless | ≈2.3 GB |
| Q8_0LosslessLossless in practice, twice the size of Q4 | ≈2.7 GB |
| FP16 / BF16LosslessFull precision, as released | ≈4.3 GB |
Estimates for 8K tokens of context: weights + KV cache + runtime overhead. Speed assumes memory-bound decoding at your GPU’s bandwidth. Pick a GPU above and the fit column checks every row against it. Weights marked “file” use the real GGUF download size; “est.” rows are computed from the parameter count.
How to pick a quant
Take the biggest quant that fits with room left for context. Q4_K_M is the usual sweet spot. Below Q4 quality drops fast, above Q6 you mostly pay memory for nothing. Big models (70B+, large MoE) tolerate low quants better than small ones.
- LosslessF16, Q8_0
- Near-losslessQ6_K, Q5_K_M
- Small lossQ5_K_S, Q4_K_M, IQ4_NL, IQ4_XS, Q4_K_S
- Noticeable lossQ4_0, Q3_K_L, Q3_K_M, IQ3_M
- Heavy lossIQ3_XXS, Q3_K_S, IQ2_M, Q2_K
Typical quality loss, not measured per model. Tiers come from KL divergence against FP16 on Llama 3 8B: llama.cpp perplexity README: LLaMA 3 8B scoreboard (KLD and ΔPPL vs FP16, Wikitext-2); llama.cpp quantize tool: ΔPPL per quant type @ Llama-3-8B
How much context fits?
At 8K context, Q4_K_M needs ≈1.9 GB(1.0 GB weights + 0.4 GB KV cache + 0.6 GB overhead). Pick a GPU above to see how much context fits.
Weights + overheadKV cache
Can the RTX 4060 run LFM3-1.7B-Base? →
Speed vs quality per quant
Smaller quants run faster and fit in less memory, but lose quality.
Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.
Real-world speed of LFM3-1.7B-Base
Tokens per second measured by people running it at home. A GPU shows a median once 3 reports agree on the same setup. Reports far from our estimate wait for a review. Measure yours with llama-bench →
No medians yet. Be one of the first 3 to report a setup.
All reports
Run LFM3-1.7B-Base locally
--quantQ4_K_Mdefault; pick your GPU to tailor itmradermacher/LFM3-1.7B-Base-GGUF
ollama run lfm3:1.7bThe library build picks its own default quantisation.
ollama run hf.co/mradermacher/LFM3-1.7B-Base-GGUF:Q4_K_M llama-server -hf mradermacher/LFM3-1.7B-Base-GGUF:Q4_K_MDownloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.
from transformers import pipeline
pipe = pipeline("text-generation", model="LiquidAI/LFM3-1.7B-Base", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])Quantised versions and downloads
| Repository | Format | Quant | Size |
|---|---|---|---|
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q2_K | 0.8 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q3_K_S | 0.9 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q3_K_M | 0.9 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q3_K_L | 1.0 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | IQ4_XS | 1.0 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q4_K_S | 1.1 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q4_K_M | 1.1 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q5_K_S | 1.2 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q5_K_M | 1.3 GB |
| mradermacher/LFM3-1.7B-Base-GGUF | GGUF | Q6_K | 1.4 GB |
About LFM3-1.7B-Base
LFM3-1.7B-Base is the pre-trained checkpoint behind LFM3-1.7B, a text model for local agents and specialized applications based on our new tri-hybrid architecture. • Pre-trained on 22T tokens: the last 3.5T tokens are curated mid-training data with agentic traces and tool calls, and extend the context window to 256K tokens. • Designed for the edge: at 4-bit precision, it takes about 1 GB of memory, so it fits on any phone or laptop. • Made to be specialized: a starting point for continued pre-training and your own post-training. Find more information about the LFM3 family in our release blog post. The LFM3 family includes base, text, vision, audio, omni, and decision models. LFM3-1.7B-Base is the checkpoint used to create LFM3-1.7B. It is only recommended for tasks that require heavy fine-tuning, like language-specific or domain-specific assistants, training on proprietary data, or experimenting with novel post-training approaches. For chat, tool calling, and agents, use LFM3-1.7B. LFM3-1.7B-Base uses the same architecture and modeling code as LFM3-1.7B, with BF16 weights. Transformers quick start: Mid-training included chat data with and without thinking traces, formatted with the same chat template as LFM3-1.7B. With apply_chat_template, reasoning is enabled by default, so responses start with a thinking trace. Set enable_thinking=False to prefill an empty thinking trace and have the model answer directly.
Full model card on Hugging Face →Alternatives to LFM3-1.7B-Base
All alternatives →Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.
Qwen2.5 is the latest series of Qwen large language models. For Qwen2.5, we release a number of base language models and instruction-tuned language models ranging from 0.5 to 72 billion parameters. Qwen2.5 brings the following improvements upon Qwen2:
We introduce Spark-X2.6-1.7B, the second model in the Spark-X2.6 series, following the release of Spark-X2.6-4B. Building on Spark-X2.5-1.7B, this release further advances our compact, general-purpose language models for on-device deployment.
This checkpoint is a five-layer draft model for Qwen3.8-27B, with block size 16 and an 8,192-token sliding window.
LFM3-1.7B-Base FAQ
- How much VRAM does LFM3-1.7B-Base need?
- About 1.9 GB at Q4_K_M with an 8K context. Q8_0 needs about 2.7 GB and full precision about 4.3 GB. Longer contexts need more memory for the KV cache.
- Can LFM3-1.7B-Base run on the RTX 4060?
- Yes. It fits at FP16 / BF16 (about 4.3 GB), at an estimated 43 tokens/s. It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.
- Can I use LFM3-1.7B-Base commercially?
- Yes, with conditions. It is released under lfm1.0. Custom licence: read the model card before using it commercially.
- What is the context length of LFM3-1.7B-Base?
- 262,144 tokens (256K). The full window needs far more memory than an 8K context.
- How do I run LFM3-1.7B-Base with Ollama or llama.cpp?
- With Ollama: ollama run lfm3:1.7b. With llama.cpp: llama-server -hf mradermacher/LFM3-1.7B-Base-GGUF:Q4_K_M, which starts an OpenAI-compatible server.
Rate this model
Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.
0 comments