处理超长prompt

Author: root

README.md

@@ -51,7 +51,22 @@ except:
         ),
         )
 
-
+处理过长的prompt：src/agentscope/model/_openai_model.py OpenAIChatModel 的__call__ 函数
+conversations = [{"role":msg["role"], "content":msg["content"][0]['text'] if type(msg["content"]) == list else msg["content"]} for msg in messages]
+input_ids = self.tokenizer.apply_chat_template(
+        conversations,
+        add_generation_prompt=True,
+        tokenize=True,
+)
+
+while len(input_ids) > 10000: （比maxlen稍微小一点）
+        messages[1]["content"][0]['text'] = messages[1]["content"][0]['text'][:150] + messages[1]["content"][0]['text'][200:]
+        conversations = [{"role":msg["role"], "content":msg["content"][0]['text'] if type(msg["content"]) == list else msg["content"]} for msg in messages]
+        input_ids = self.tokenizer.apply_chat_template(
+        conversations,
+        add_generation_prompt=True,
+        tokenize=True,
+        )
 
 verlv0.5.0 改动
 

examples/werewolf/train.sh

@@ -45,11 +45,12 @@ python -m agentlightning.verl \
     algorithm.use_kl_in_reward=False \
     trainer.default_local_dir='/root/dataDisk/checkpoints' \
     trainer.rollout_data_dir='/root/dataDisk/rollout' \
+    trainer.resume_mode=auto \
     trainer.critic_warmup=0 \
     trainer.logger=['console','wandb'] \
     trainer.project_name=${PROJECT_NAME} \
     trainer.experiment_name=${EXPERIMENT_NAME} \
     trainer.nnodes=1 \
-    trainer.save_freq=10 \
+    trainer.save_freq=1 \
     trainer.test_freq=0 \
     trainer.total_epochs=1 $@