CroissantLLM 1.3B
By CroissantLLM · France
Updated 2026-07-13
Overview
A 1.3B bilingual French/English model from Sorbonne's MLIA lab, light enough to run on a CPU and shipped with a fully auditable training corpus.
When to pick this model
- CPU-only or extreme edge deployments
- Academic research needing a transparent, reproducible model
- Lightweight bilingual French/English classification or completion
- Teaching and demos where size and openness matter
- Embedded devices with under 2GB of memory
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 1 GB |
| Q5_K_M | 1.2 GB |
| Q8_0 | 2 GB |
| FP16 (no quantization) | 3 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, CroissantLLM 1.3B fits an 8 GB consumer card at Q4_K_M (1 GB). Stepping up to Q8_0 nearly doubles the footprint to 2 GB, and unquantized FP16 weights take 3 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, CroissantLLM 1.3B needs roughly 4 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 40 tokens/sec on entry-level GPUs, on the order of 120 tokens/sec on a mid-range card, and up to 250 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches CroissantLLM 1.3B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for CroissantLLM 1.3B |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | FP16 (3 GB used) |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | FP16 (3 GB used) |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | FP16 (3 GB used) |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | FP16 (3 GB used) |
| 32 GB | RTX 5090 | FP16 (3 GB used) |
Which GPU should you buy to run CroissantLLM 1.3B?
To run CroissantLLM 1.3B locally at Q4, you need ~1 GB of VRAM. The best value for this is a RTX 5060 (8 GB VRAM).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Published benchmark scores
| Benchmark | Score |
|---|---|
| FrenchBench | 38 |
Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.
Strengths
- Runs in roughly 1GB VRAM at Q4
- Native French/English balance, not an afterthought
- Fully auditable training corpus
- Permissive MIT-style licensing
Limitations
- 2048-token context is too tight for most real tasks
- Quality is well below any modern 2025 model
- No vision, tools, or chain-of-thought reasoning
- Limited ecosystem and tooling support
Typical workloads
In our catalog grid, CroissantLLM 1.3B is filed under Low-End Devices, Experimentation — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: French-language output where quality matters.
Note the 2k-token context window — fine for short interactions, limiting for long documents or big retrieval contexts. The MIT license is permissive, so shipping it inside a commercial product raises no special legal questions.
Architecture & training
Architecture: Dense Transformer · 1.3B · trained on Jean Zay (IDRIS)
Training: MLIA project (Sorbonne) — balanced FR/EN web corpus, transparent public corpus.
An academic milestone for transparent bilingual training — not competitive for production use in 2025.
Quick start
ollama pull hf.co/manu/croissant-llm-chat-v0.1-GGUFOr use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does CroissantLLM 1.3B need?
At the recommended Q4_K_M quantization, CroissantLLM 1.3B needs about 1 GB of VRAM. Q8_0 takes 2 GB, and unquantized FP16 weights take 3 GB.
Can CroissantLLM 1.3B run without a GPU?
Yes — with roughly 4 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does CroissantLLM 1.3B support?
CroissantLLM 1.3B supports a 2k-token context window (2,048 tokens).
Can I use CroissantLLM 1.3B commercially?
Yes. CroissantLLM 1.3B is released under MIT, a permissive open-source license that allows commercial use, modification and redistribution.
How fast is CroissantLLM 1.3B on consumer hardware?
Our compatibility engine estimates on the order of 120 tokens/sec on a mid-range GPU and up to 250 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of CroissantLLM 1.3B should I download first?
Start with Q4_K_M (1 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It already fits an 8 GB card at FP16.