LiveMem-RL

LiveMem-RL

LiveMem-RL is the reinforcement-learning checkpoint of LiveMem-4B-SFT. It uses a Qwen3 attention path in parallel with a Gated DeltaNet 2 (GDN2) recurrent memory path at every decoder layer:

layer output = Qwen3 attention output + GDN2 memory output

This checkpoint was trained with GRPO from chen-l/LiveMem-SFT. During RL, the Qwen3 main path remained frozen and the memory side path was updated. The configured maximum context length is 262,144 tokens; actual usable context depends on GPU memory and inference backend.

Transformers usage

LiveMem uses custom model code and GDN2 Triton kernels. A CUDA environment is required for inference.

pip install -r requirements.txt

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "chen-l/LiveMem-4B-RL"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
    dtype=torch.bfloat16,
    device_map="auto",
)

messages = [{"role": "user", "content": "Answer using the supplied long context."}]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
    return_dict=True,
).to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

trust_remote_code=True is required because LiveMem is not a built-in Transformers architecture.