ver Mar18th

checked Claude agent
This commit is contained in:
David Chang
2024-03-18 17:42:13 +08:00
parent 30cc887088
commit 3db0591868
4 changed files with 11 additions and 2 deletions

View File

@@ -5,5 +5,8 @@ evaluation_examples
logs
mm_agents
run.py
lib_run_single.py
settings.json
quick_evaluate.py

View File

@@ -1 +1 @@
test
claude

View File

@@ -256,7 +256,6 @@ class PromptAgent:
if self.observation_type == "screenshot_a11y_tree":
_screenshot = previous_obs["screenshot"]
_linearized_accessibility_tree = previous_obs["accessibility_tree"]
logger.debug("LINEAR AT: %s", _linearized_accessibility_tree)
messages.append({
"role": "user",
@@ -343,6 +342,7 @@ class PromptAgent:
if self.observation_type in ["screenshot", "screenshot_a11y_tree"]:
base64_image = encode_image(obs["screenshot"])
linearized_accessibility_tree = linearize_accessibility_tree(accessibility_tree=obs["accessibility_tree"])
logger.debug("LINEAR AT: %s", linearized_accessibility_tree)
if self.observation_type == "screenshot_a11y_tree":
self.observations.append({
@@ -376,6 +376,7 @@ class PromptAgent:
})
elif self.observation_type == "a11y_tree":
linearized_accessibility_tree = linearize_accessibility_tree(accessibility_tree=obs["accessibility_tree"])
logger.debug("LINEAR AT: %s", linearized_accessibility_tree)
self.observations.append({
"screenshot": None,
@@ -423,6 +424,8 @@ class PromptAgent:
# with open("messages.json", "w") as f:
# f.write(json.dumps(messages, indent=4))
#logger.info("PROMPT: %s", messages)
response = self.call_llm({
"model": self.model,
"messages": messages,
@@ -522,6 +525,8 @@ class PromptAgent:
claude_messages[1]['content'].insert(0, claude_system_message_item)
claude_messages.pop(0)
logger.debug("CLAUDE MESSAGE: %s", repr(claude_messages))
# headers = {
# "x-api-key": os.environ["ANTHROPIC_API_KEY"],
# "anthropic-version": "2023-06-01",

1
run.py
View File

@@ -140,6 +140,7 @@ def test(
env = DesktopEnv(
path_to_vm=args.path_to_vm,
snapshot_name="Snapshot 35",
action_space=agent.action_space,
screen_size=(args.screen_width, args.screen_height),
headless=args.headless,