Instructions to use Xirui1208/readtwice-v7-RL-step-100 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Xirui1208/readtwice-v7-RL-step-100 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Xirui1208/readtwice-v7-RL-step-100")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Xirui1208/readtwice-v7-RL-step-100")
model = AutoModelForCausalLM.from_pretrained("Xirui1208/readtwice-v7-RL-step-100", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Xirui1208/readtwice-v7-RL-step-100 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Xirui1208/readtwice-v7-RL-step-100"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Xirui1208/readtwice-v7-RL-step-100",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Xirui1208/readtwice-v7-RL-step-100

SGLang

How to use Xirui1208/readtwice-v7-RL-step-100 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Xirui1208/readtwice-v7-RL-step-100" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Xirui1208/readtwice-v7-RL-step-100",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Xirui1208/readtwice-v7-RL-step-100" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Xirui1208/readtwice-v7-RL-step-100",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Xirui1208/readtwice-v7-RL-step-100 with Docker Model Runner:
```
docker model run hf.co/Xirui1208/readtwice-v7-RL-step-100
```

ReadTwice v7 RL — logical step 100

This is the ReadTwice v7 reinforcement-learning checkpoint at logical step 100, exported from the VERL FSDP checkpoint global_step_140.

The step mapping follows the convention used by the earlier v7 RL releases:

physical steps 1–50 were pilot training with batch size 32 and 8 rollouts, and together map to logical step 10;
physical steps 51–140 used full-scale training with batch size 128 and 16 rollouts;
therefore physical step 140 maps to logical step 100.

The model was initialized from Xirui1208/memagent-readtwice-all-v7-7B. Its architecture is Qwen2ForCausalLM, and the exported weights use BF16.

Usage note

This checkpoint was trained as a recurrent ReadTwice/MemAgent model. Reproducing the trained behavior requires the matching recurrent skim/update/final protocol; loading it as an ordinary one-shot text-generation model does not reproduce that protocol.

Evaluation

Formal logical-step-100 evaluation results are not included yet.

Downloads last month: -

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for Xirui1208/readtwice-v7-RL-step-100

Base model

Xirui1208/memagent-readtwice-all-v7-7B

Finetuned

(1)

this model