gavindiaz/NexQuant
1
0
Fork 0
mirror of https://github.com/NicolasBohn/NexQuant.git synced 2026-07-27 23:47:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix debug_tpl and debug_llm saving method(#874)

Browse Source
This commit is contained in:
XianBW
2025-05-15 18:37:51 +08:00
committed by GitHub
parent 7943bbcc86
commit 5e1292974b
3 changed files with 61 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rdagent/log/logger.py
-16
View File
@@ -114,22 +114,6 @@ class RDAgentLog(SingletonBaseClass):
# TODO: I think we can merge the log_object function with other normal log methods to make the interface simpler.
caller_info = get_caller_info()
tag = f"{self._tag}.{tag}.{self.get_pids()}".strip(".")
# FIXME: it looks like a hacking... We should redesign it...
if "debug_" in tag:
debug_log_path = self.log_trace_path / "debug_llm.pkl"
debug_data = {"tag": tag, "obj": obj}
if debug_log_path.exists():
with debug_log_path.open("rb") as f:
existing_data = pickle.load(f)
existing_data.append(debug_data)
with debug_log_path.open("wb") as f:
pickle.dump(existing_data, f)
else:
with debug_log_path.open("wb") as f:
pickle.dump([debug_data], f)
return
logp = self.storage.log(obj, name=tag, save_type="pkl")
file_handler_id = logger.add(
rdagent/log/ui/ds_summary.py
Regular → Executable
+3 -1
View File
@@ -257,7 +257,9 @@ def get_summary_df(log_folders: list[str]) -> tuple[dict, pd.DataFrame]:
base_df["SOTA Exp"] = base_df["SOTA Exp"].replace("", pd.NA)
base_df["SOTA Exp Score (valid)"] = base_df["SOTA Exp Score (valid)"].replace("Not Calculated", 0)
base_df["SOTA Exp Score (valid)"] = (
base_df["SOTA Exp Score (valid)"].replace("Not Calculated", 0).replace("Not Computed", 0)
)
base_df = base_df.astype(
{
"Total Loops": int,
rdagent/log/ui/ds_trace.py
+58 -36
View File
@@ -67,25 +67,47 @@ def load_times(log_path: Path):
@st.cache_data(persist=True)
def load_data(log_path: Path):
data = defaultdict(lambda: defaultdict(dict))
llm_data = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
for msg in FileStorage(log_path).iter_msg():
if msg.tag and "llm" not in msg.tag and "session" not in msg.tag and "batch embedding" not in msg.tag:
if not msg.tag:
continue
li, fn = extract_loopid_func_name(msg.tag)
ei = extract_evoid(msg.tag)
if li:
li = int(li)
if ei is not None:
ei = int(ei)
if "debug_" in msg.tag:
if "debug_tpl" in msg.tag and "filter_" in msg.content["uri"]:
continue
if ei is not None:
llm_data[li][fn][ei].append(
{
"tag": msg.tag,
"obj": msg.content,
}
)
else:
llm_data[li][fn]["no_tag"].append(
{
"tag": msg.tag,
"obj": msg.content,
}
)
elif "llm" not in msg.tag and "session" not in msg.tag and "batch embedding" not in msg.tag:
if msg.tag == "competition":
data["competition"] = msg.content
continue
li, fn = extract_loopid_func_name(msg.tag)
if li:
li = int(li)
ei = extract_evoid(msg.tag)
msg.tag = re.sub(r"\.evo_loop_\d+", "", msg.tag)
msg.tag = re.sub(r"Loop_\d+\.[^.]+\.?", "", msg.tag)
msg.tag = msg.tag.strip()
if ei:
if int(ei) not in data[li][fn]:
data[li][fn][int(ei)] = {}
data[li][fn][int(ei)][msg.tag] = msg.content
if ei is not None:
if ei not in data[li][fn]:
data[li][fn][ei] = {}
data[li][fn][ei][msg.tag] = msg.content
else:
if msg.tag:
data[li][fn][msg.tag] = msg.content
@@ -93,30 +115,30 @@ def load_data(log_path: Path):
if not isinstance(msg.content, str):
data[li][fn]["no_tag"] = msg.content
# debug_llm data
llm_data = defaultdict(lambda: defaultdict(lambda: defaultdict(list)))
# To be compatible with old version log trace, keep this
llm_log_p = log_path / "debug_llm.pkl"
try:
rd = pickle.loads(llm_log_p.read_bytes())
except:
rd = []
for d in rd:
t = d["tag"]
if "debug_exp_gen" in t:
continue
if "debug_tpl" in t and "filter_" in d["obj"]["uri"]:
continue
lid, fn = extract_loopid_func_name(t)
ei = extract_evoid(t)
if lid:
lid = int(lid)
if ei:
ei = int(ei)
if llm_log_p.exists():
try:
rd = pickle.loads(llm_log_p.read_bytes())
except:
rd = []
for d in rd:
t = d["tag"]
if "debug_exp_gen" in t:
continue
if "debug_tpl" in t and "filter_" in d["obj"]["uri"]:
continue
lid, fn = extract_loopid_func_name(t)
ei = extract_evoid(t)
if lid:
lid = int(lid)
if ei is not None:
ei = int(ei)
if ei is not None:
llm_data[lid][fn][ei].append(d)
else:
llm_data[lid][fn]["no_tag"].append(d)
if ei is not None:
llm_data[lid][fn][ei].append(d)
else:
llm_data[lid][fn]["no_tag"].append(d)
return convert_defaultdict_to_dict(data), convert_defaultdict_to_dict(llm_data)
@@ -652,7 +674,7 @@ with st.sidebar:
st.radio(
f"Select :blue[**one log folder**]",
state.log_folders,
format_func=lambda x: x[x.rfind("amlt") + 5 :].split("/")[0],
format_func=lambda x: x[x.rfind("amlt") + 5 :].split("/")[0] if "amlt" in x else x,
)
)
if not state.log_folder.exists():
@@ -702,11 +724,11 @@ if state.data["competition"]:
st.title(state.data["competition"])
st.markdown(f"[share_link](/ds_trace?log_folder={state.log_folder}&selection={state.log_path})")
summarize_data()
if len(state.data) > 2:
min_id, max_id = get_state_data_range(state.data)
min_id, max_id = get_state_data_range(state.data)
if max_id > min_id:
loop_id = st.slider("Loop", min_id, max_id, min_id)
else:
loop_id = 0
loop_id = min_id
if state.show_stdout:
stdout_win(loop_id)
main_win(loop_id, state.llm_data[loop_id] if loop_id in state.llm_data else None)