amr-parsing-grpo-single-single-turn-20260224-1412-global-step-2841

Model Information

• Base Model: checkpoints/amr-parsing-grpo-single/single-turn-20260224-1412/global_step_2841/actor
• Timestamp: 20260224-1412

Benchmark Results

• Benchmark File: amr-parsing-grpo-single-single-turn-20260224-1412-global-step-2841_20260224-1412.jsonl

• Score:

  • F1: 82.08
  • Precision: 82.74
  • Recall: 81.44

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("viamr-project/amr-parsing-grpo-single-single-turn-20260224-1412-global-step-2841", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("viamr-project/amr-parsing-grpo-single-single-turn-20260224-1412-global-step-2841", trust_remote_code=True)

Sample Prompt:

<|im_start|>user
Convert the following English sentence into its Abstract Meaning Representation (AMR):

<sentence>2007-08-21</sentence><|im_end|>
<|im_start|>assistant
<think>

Training Details

• Framework: veRL (Reinforcement Learning)
• Task: AMR Parsing