mirrors/mindcraft
1
0
Fork 0
mirror of https://github.com/kolbytn/mindcraft.git synced 2025-08-04 14:25:43 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

added exp name to prompter.js and splitting around think tokens

Browse source
This commit is contained in:
Isadora White 2025-03-19 23:52:08 -05:00
parent 6f8027b86f
commit 7bf97660eb
2 changed files with 6 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
evaluation_script.py
View file

@ -261,6 +261,7 @@ def launch_parallel_experiments(task_path,
total_run = results["total"] total_run = results["total"]
print(f"Total tasks run: {total_run}/{total_num_experiments}") print(f"Total tasks run: {total_run}/{total_num_experiments}")
print(results) print(results)
results["exp_name"] = exp_name
results["template_profile"] = template_profile results["template_profile"] = template_profile
results["model"] = model results["model"] = model
results["api"] = api results["api"] = api
@ -358,15 +359,6 @@ def launch_server_experiment(task_path,
set_environment_variable_tmux_session(session_name, "NUM_EXAMPLES", str(num_examples)) set_environment_variable_tmux_session(session_name, "NUM_EXAMPLES", str(num_examples))
if insecure_coding: if insecure_coding:
set_environment_variable_tmux_session(session_name, "INSECURE_CODING", "true") set_environment_variable_tmux_session(session_name, "INSECURE_CODING", "true")
# you need to add the bots to the world first before you can add them as op
# cmd = f"node main.js --task_path example_tasks.json --task_id debug_{num_agents}_agent_timeout"
# subprocess.run(["tmux", "send-keys", "-t", session_name, cmd, "C-m"])
# time.sleep(40)
# subprocess.run(["tmux", "send-keys", "-t", "server_" + session_name, f"/op @a", "C-m"])
make_ops(agent_names, session_name) make_ops(agent_names, session_name)
# add the bots as op # add the bots as op
@ -379,9 +371,6 @@ def launch_server_experiment(task_path,
set_environment_variable_tmux_session(session_name, "BLOCKED_ACTIONS", BLOCKED_ACTIONS_CRAFTING) set_environment_variable_tmux_session(session_name, "BLOCKED_ACTIONS", BLOCKED_ACTIONS_CRAFTING)
elif task_type == "construction": elif task_type == "construction":
set_environment_variable_tmux_session(session_name, "BLOCKED_ACTIONS", BLOCKED_ACTIONS_CONSTRUCTION) set_environment_variable_tmux_session(session_name, "BLOCKED_ACTIONS", BLOCKED_ACTIONS_CONSTRUCTION)
script_content = "" script_content = ""
for task_id in task_ids: for task_id in task_ids:

5
src/models/prompter.js
View file

@ -395,6 +395,11 @@ export class Prompter {
if (current_msg_time !== this.most_recent_msg_time) { if (current_msg_time !== this.most_recent_msg_time) {
console.warn(`${this.agent.name} received new message while generating, discarding old response.`); console.warn(`${this.agent.name} received new message while generating, discarding old response.`);
return ''; return '';
}
if (generation?.includes('</think>')) {
const [_, afterThink] = generation.split('</think>')
generation = afterThink
} }
return generation; return generation;