xiamoent

Agent-G2-alfworld-1.5b

README

License: apache-2.0

Model Details

Table with columns: Item, Description
Item	Description
Base model	`Qwen/Qwen2.5-1.5B-Instruct`
Architecture	`Qwen2ForCausalLM`
Checkpoint format	BF16 Safetensors
Configured context length	32,768 tokens
Target environment	ALFWorld / ALFRED text environment
Post-training	Agent-G2 with GRPO
Language	English
Required output format	`<think>...</think><action>...</action>`

Although the tokenizer metadata contains a larger generic maximum length, the model configuration declares 32,768 positions and the released training recipe uses at most 4,096 prompt tokens plus 512 response tokens.

Evaluation

The Agent-G2 project reports the following ALFWorld success rates for this 1.5B checkpoint:

Table with columns: Task group, Success rate
Task group	Success rate
Pick	96.8%
Look	100.0%
Clean	100.0%
Heat	92.9%
Cool	84.2%
Pick Two	94.7%
All tasks	95.3%

Expert-prefix guidance is enabled during training but disabled during validation in the released configuration (gmsv.apply_on_validation=false). The reported results therefore do not require an expert trajectory at inference time.

These results are reported by the Agent-G2 repository and have not been independently reproduced in this model card. Evaluation variance is not currently available. Results may vary with the ALFWorld version, task split, prompt template, action history, random seed, and decoding configuration.

Intended Use

This checkpoint is intended for:

reproducing Agent-G2 results in the ALFWorld text environment;
research on long-horizon language agents and agentic reinforcement learning;
studying adaptive expert-prefix guidance;
evaluating action selection over an environment-provided admissible action set.

For faithful evaluation, use the ALFWorld environment, prompt template, action parser, and rollout loop provided by the Agent-G2 repository. A standalone generation only demonstrates that the checkpoint loads successfully; it does not reproduce the interactive benchmark.

Environment Interface

At every environment step, provide the task, current observation, recent history, and admissible actions. The released parser expects English output containing reasoning and one action selected from the current admissible set:

text
<think>Reason about the observation and admissible actions.</think>
<action>put apple 1 in/on fridge 1</action>

Missing tags or outputs containing Chinese characters are marked invalid by the released ALFWorld parser. The action inside <action>...</action> must match an action that the environment currently allows.

Quick Start

bash
pip install -U transformers accelerate torch

The following example performs one ALFWorld-style generation step. Replace the placeholders with state supplied by the environment:

python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "xiamoent/Agent-G2-alfworld-1.5b"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)
model.eval()

task_description = "<ALFWorld task>"
current_observation = "<current observation>"
admissible_actions = ["<admissible action 1>", "<admissible action 2>"]
actions_text = ", ".join(admissible_actions)

prompt = f"""
You are an expert agent operating in the ALFRED Embodied Environment.
Your task is to: {task_description}
Your current observation is: {current_observation}
Your admissible actions of the current situation are: [{actions_text}].

Now take one action. Enclose your reasoning within <think> </think> tags, then
present one admissible action within <action> </action> tags.
""".strip()

inputs = tokenizer.apply_chat_template(
    [{"role": "user", "content": prompt}],
    tokenize=True,
    add_generation_prompt=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

with torch.inference_mode():
    output_ids = model.generate(
        **inputs,
        max_new_tokens=512,
        do_sample=True,
        temperature=0.4,
        top_p=0.8,
        top_k=20,
        repetition_penalty=1.1,
    )

new_tokens = output_ids[0, inputs["input_ids"].shape[-1]:]
response = tokenizer.decode(new_tokens, skip_special_tokens=True)
print(response)

The released checkpoint's generation_config.json defaults to temperature 0.7. The example uses temperature 0.4 to match the released validation configuration.

Training

Agent-G2 uses expert ALFWorld trajectories as prefix guidance during training, followed by policy rollouts and GRPO updates. The guidance depth is sampled per task from a Gaussian distribution estimated online from existing rollout statistics. Prefix guidance is a training mechanism; it is not required for validation or deployment.

The expert-prefix store contains 3,553 ALFWorld trajectories with action lengths from 3 to 15. The released training recipe specifies:

Table with columns: Configuration, Value
Configuration	Value
Learning rate	`1e-6`
Training batch size	`16`
Rollouts per task	`8`
Maximum prompt length	`4096`
Maximum response length	`512`
Maximum ALFWorld steps	`20`

See the paper-locked run_alfworld.sh for the complete recipe. The public repository does not identify the exact checkpoint step or selection rule used for this Hub upload, so the table documents the released recipe rather than claiming that this artifact is the final epoch checkpoint.

Limitations

The model is specialized for the text-based ALFWorld environment and may not generalize to other simulators or physical environments.
It can produce malformed or inadmissible actions; environment-side validation is required.
Performance is sensitive to prompt formatting, observation history, decoding settings, random seed, and environment configuration.
The reported evaluation does not include variance across repeated runs.
The model may inherit factual errors, biases, and other limitations from the base model and training data.
This checkpoint should not directly control physical systems or be used for consequential real-world actions without independent safety mechanisms.

Citation

If you find this checkpoint useful, please cite Agent-G2:

bibtex
@misc{wang2026agentg2gaussianguidanceagentic,
      title={Agent-G$^2$: Gaussian Guidance for Agentic Reinforcement Learning},
      author={Zixuan Wang and Yanrui Miao and Zhengxi Lu and Teng Pan and Yiwen Qiu and Hongxing Li and Peng Qiu and Ruiqing Zhang and Yongliang Shen},
      year={2026},
      eprint={2608.23318},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2608.23318},
}

The paper has been accepted to the EMNLP 2026 Main Conference. A public paper link will be added when available.

Acknowledgements

Agent-G2 builds on verl-agent, veRL, and ALFWorld.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

xiamoent

Model Tree

Base

Qwen/Qwen2.5-1.5B-Instruct

Fine-tuned

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Model Details

Table with columns: Item, Description
Item	Description
Base model	`Qwen/Qwen2.5-1.5B-Instruct`
Architecture	`Qwen2ForCausalLM`
Checkpoint format	BF16 Safetensors
Configured context length	32,768 tokens
Target environment	ALFWorld / ALFRED text environment
Post-training	Agent-G2 with GRPO
Language	English
Required output format	`<think>...</think><action>...</action>`

Evaluation

The Agent-G2 project reports the following ALFWorld success rates for this 1.5B checkpoint:

Table with columns: Task group, Success rate
Task group	Success rate
Pick	96.8%
Look	100.0%
Clean	100.0%
Heat	92.9%
Cool	84.2%
Pick Two	94.7%
All tasks	95.3%

Intended Use

This checkpoint is intended for:

reproducing Agent-G2 results in the ALFWorld text environment;
research on long-horizon language agents and agentic reinforcement learning;
studying adaptive expert-prefix guidance;
evaluating action selection over an environment-provided admissible action set.

Environment Interface

text
<think>Reason about the observation and admissible actions.</think>
<action>put apple 1 in/on fridge 1</action>

Quick Start

bash
pip install -U transformers accelerate torch

The following example performs one ALFWorld-style generation step. Replace the placeholders with state supplied by the environment:

python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "xiamoent/Agent-G2-alfworld-1.5b"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)
model.eval()

task_description = "<ALFWorld task>"
current_observation = "<current observation>"
admissible_actions = ["<admissible action 1>", "<admissible action 2>"]
actions_text = ", ".join(admissible_actions)

prompt = f"""
You are an expert agent operating in the ALFRED Embodied Environment.
Your task is to: {task_description}
Your current observation is: {current_observation}
Your admissible actions of the current situation are: [{actions_text}].

Now take one action. Enclose your reasoning within <think> </think> tags, then
present one admissible action within <action> </action> tags.
""".strip()

inputs = tokenizer.apply_chat_template(
    [{"role": "user", "content": prompt}],
    tokenize=True,
    add_generation_prompt=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

with torch.inference_mode():
    output_ids = model.generate(
        **inputs,
        max_new_tokens=512,
        do_sample=True,
        temperature=0.4,
        top_p=0.8,
        top_k=20,
        repetition_penalty=1.1,
    )

new_tokens = output_ids[0, inputs["input_ids"].shape[-1]:]
response = tokenizer.decode(new_tokens, skip_special_tokens=True)
print(response)

The released checkpoint's generation_config.json defaults to temperature 0.7. The example uses temperature 0.4 to match the released validation configuration.

Training

The expert-prefix store contains 3,553 ALFWorld trajectories with action lengths from 3 to 15. The released training recipe specifies:

Table with columns: Configuration, Value
Configuration	Value
Learning rate	`1e-6`
Training batch size	`16`
Rollouts per task	`8`
Maximum prompt length	`4096`
Maximum response length	`512`
Maximum ALFWorld steps	`20`

Limitations

The model is specialized for the text-based ALFWorld environment and may not generalize to other simulators or physical environments.
It can produce malformed or inadmissible actions; environment-side validation is required.
Performance is sensitive to prompt formatting, observation history, decoding settings, random seed, and environment configuration.
The reported evaluation does not include variance across repeated runs.
The model may inherit factual errors, biases, and other limitations from the base model and training data.
This checkpoint should not directly control physical systems or be used for consequential real-world actions without independent safety mechanisms.

Citation

If you find this checkpoint useful, please cite Agent-G2:

bibtex
@misc{wang2026agentg2gaussianguidanceagentic,
      title={Agent-G$^2$: Gaussian Guidance for Agentic Reinforcement Learning},
      author={Zixuan Wang and Yanrui Miao and Zhengxi Lu and Teng Pan and Yiwen Qiu and Hongxing Li and Peng Qiu and Ruiqing Zhang and Yongliang Shen},
      year={2026},
      eprint={2608.23318},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2608.23318},
}

The paper has been accepted to the EMNLP 2026 Main Conference. A public paper link will be added when available.

Acknowledgements

Agent-G2 builds on verl-agent, veRL, and ALFWorld.