Downloads · 30 days
43
12% of all-time downloads
MiniCPM5-1B-ru-v3
vahpetr/MiniCPM5-1B-ru-v3
MiniCPM5-1B-ru-v3 is a machine learning model from vahpetr. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
openbmb/MiniCPM5-1B with a Russian LoRA (v3) merged in, to improve practical Russian (customer-support style, everyday instructions, translation) while keeping the base's coding ability. Trained on Vikhrmodels/GrandMa…
All-time downloads
354
Public
Parameters
1.1B
3.3 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors2.2 GB · 65%
At a glance
- License
- apache-2.0
- Model type
- llama
- Access
- Public
- Created
- Jul 7, 2026
- Updated
- Jul 7, 2026
- SHA
- 6b80ef5a
Base models
Model card
From the Hugging Face model README
MiniCPM5-1B-ru (v3, merged, experimental)
openbmb/MiniCPM5-1B with a Russian LoRA (v3) merged in, to improve practical Russian (customer-support style, everyday instructions, translation) while keeping the base's coding ability. Trained on Vikhrmodels/GrandMaster-PRO-MAX (Russian + English, chain-of-thought) plus an English replay slice, on an AMD Strix Halo iGPU (ROCm). Adapter-only version: vahpetr/MiniCPM5-1B-ru-lora-v3.
Honest evaluation (base vs this, thinking OFF, 3-run avg on local batteries)
| axis | base | this | note |
|---|---|---|---|
| Reasoning (/18) | ~16 | ~12 | ↓ tradeoff |
| IFEval (/10) | ~8 | ~5 | ↓ tradeoff |
| Code (/7) | 7 | 7 | preserved |
| Agentic tool-use (/10) | ~6 | ~5 | ≈ preserved |
| General / Ortho (/23) | ~12 | ~13 | ↑ (customer-service, translation) |
| Russian math battery | low | low | unchanged (1B math-capped) |
What it does well: practical Russian gets noticeably more fluent and coherent (customer-support replies become natural Russian instead of English/Chinese-mixed; RU→EN translation more accurate).
The tradeoff: sharp reasoning and strict instruction-following drop somewhat — an inherent 1B capacity limit. For strong Russian without the tradeoff, use a Russian-native model or a larger base.
Use thinking OFF (enable_thinking=false).
Usage
Load as a normal model (transformers / vLLM). A Q8_0 GGUF (MiniCPM5-1B-ru-v3-Q8_0.gguf) is included for Ollama / LM Studio / llama.cpp.
from transformers import AutoModelForCausalLM, AutoTokenizer
m = AutoModelForCausalLM.from_pretrained("vahpetr/MiniCPM5-1B-ru-v3", trust_remote_code=True)
tok = AutoTokenizer.from_pretrained("vahpetr/MiniCPM5-1B-ru-v3", trust_remote_code=True)
Attribution & license
- Base: MiniCPM5-1B (OpenBMB) — Apache-2.0
- Data: GrandMaster-PRO-MAX (Vikhr) — Apache-2.0
- This model — Apache-2.0
- Type
- llama
- License
- apache-2.0
- Languages
- ru, en
- Created
- Jul 7, 2026
- Updated
- Jul 7, 2026