MiniCPM4 and MiniCPM4.1 series are highly efficient large language models (LLMs) designed explicitly for end-side devices, which achieves this efficiency through systematic innovation in four key dimensions: model architecture, training data, training algorithms, and inference systems.
MiniCPM4 and MiniCPM4.1 are extremely efficient edge-side large model that has undergone efficient optimization across four dimensions: model architecture, learning algorithms, training data, and inference systems, achieving ultimate efficiency improvements.
-
🏗️ Efficient Model Architecture:
- InfLLM v2 -- Trainable Sparse Attention Mechanism: Adopts a trainable sparse attention mechanism architecture where each token only needs to compute relevance with less than 5% of tokens in 128K long text processing, significantly reducing computational overhead for long texts
-
🧠 Efficient Learning Algorithms:
- Model Wind Tunnel 2.0 -- Efficient Predictable Scaling: Introduces scaling prediction methods for performance of downstream tasks, enabling more precise model training configuration search
- BitCPM -- Ultimate Ternary Quantization: Compresses model parameter bit-width to 3 values, achieving 90% extreme model bit-width reduction
- Efficient Training Engineering Optimization: Adopts FP8 low-precision computing technology combined with Multi-token Prediction training strategy
-
📚 High-Quality Training Data:
- UltraClean -- High-quality Pre-training Data Filtering and Generation: Builds iterative data cleaning strategies based on efficient data verification, open-sourcing high-quality Chinese and English pre-training dataset UltraFinweb
- UltraChat v2 -- High-quality Supervised Fine-tuning Data Generation: Constructs large-scale high-quality supervised fine-tuning datasets covering multiple dimensions including knowledge-intensive data, reasoning-intensive data, instruction-following data, long text understanding data, and tool calling data
-
⚡ Efficient Inference System:
- CPM.cu -- Lightweight and Efficient CUDA Inference Framework: Integrates sparse attention, model quantization, and speculative sampling to achieve efficient prefilling and decoding
- ArkInfer -- Cross-platform Deployment System: Supports efficient deployment across multiple backend environments, providing flexible cross-platform adaptation capabilities
1from awq import AutoAWQForCausalLM
2import torch
3from transformers import AutoTokenizer
4
5prompt = "北京有什么好玩的地方？"
6quant_path = "MiniCPM4.1-8B-AutoAWQ"
7
8messages = [{"role": "user", "content": prompt}]
9
10model = AutoAWQForCausalLM.from_quantized(
11 quant_path,
12 fuse_layers=False,
13 trust_remote_code=True
14)
15tokenizer = AutoTokenizer.from_pretrained(
16 quant_path,
17 trust_remote_code=True
18)
19device = next(model.model.parameters()).device
20
21# if open think mode, use the following code
22formatted_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt = True, enable_thinking = True)
23# if close think mode, use the following code
24# formatted_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt = True, enable_thinking = False)
25
26input_ids = tokenizer.encode(formatted_prompt, return_tensors='pt').to(device)
27outputs = model.generate(
28 input_ids,
29 max_new_tokens=1000,
30 do_sample=True
31)
32# if open think mode, use the following code
33ans = [i.split("<|im_start|> assistant\n", 1)[1].strip() for i in tokenizer.batch_decode(outputs)]
34# if close think mode, use the following code
35# ans = [i.split("<|im_start|> assistant\n<think>\n\n</think>", 1)[1].strip() for i in tokenizer.batch_decode(outputs)]
