TRIM-KV is an efficient and learnable key–value eviction strategy designed to improve the efficiency of large language models (LLMs) in long-horizon inference.

This model is a version of TrimKV-Qwen3-8B-Math enhanced with TRIM-KV, as presented in the paper Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs.

The core idea behind TRIM-KV is to learn the intrinsic importance of each key–value pair at creation time, which we call token retention, and then decay this importance exponentially over time to mimic standard inference running with eviction.

The retention score is query-agnostic and captures the long-term utility of tokens. This is different from attention scores, which are query-dependent: they capture the short-term utility for predicting the next token and are recomputed at every step.

Paper: Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs
Code: GitHub - ngocbh/trimkv
Arxiv: 2512.03324

Quick Start

Installation

To use this model, you need to install the trimkv library:

git clone https://github.com/ngocbh/trimkv.git
cd trimkv
pip install -e .

Inference

python

import torch
from trimkv.models.qwen3 import TrimKVQwen3ForCausalLM
from trimkv.cache_utils import TrimKVCache
from transformers import AutoTokenizer

model_path = "ngocbh/TrimKV-Qwen3-8B-Math"

model = TrimKVQwen3ForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.bfloat16,
    load_trimkv_weights=True,
    download_from="huggingface",
    use_cache=True,
    device_map="cuda",
)

# Configure TRIM-KV settings
model.config._attn_implementation = "flash_attention_2"
model.config.compress_memory = True
model.config.memory_size = 512
model.config.buffer_size = 128

tokenizer = AutoTokenizer.from_pretrained(
    model.config.base_model,
    use_fast=True,
    padding_side="left",
)

# Use model.generate as normal.
# Note: TRIM-KV uses TrimKVCache under the hood. So please pass TrimKVCache to model.generate

Citation

bibtex

@article{bui2025cache,
  title={Cache what lasts: Token retention for memory-bounded kv cache in llms},
  author={Bui, Ngoc and Sharma, Shubham and Lamba, Simran and Mishra, Saumitra and Ying, Rex},
  journal={arXiv preprint arXiv:2512.03324},
  year={2025}
}
@article{bui2025make,
  title={Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction},
  author={Bui, Ngoc and Nguyen, Hieu Trung and Cohan, Arman and Ying, Rex},
  journal={arXiv preprint arXiv:2512.03324},
  year={2025}
}