File size: 8,569 Bytes
a58d1bb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 |
import json
import os
from collections import defaultdict
import gradio as gr
# Global in-memory state (kept minimal on purpose)
CURRENT = {
"file_path": None,
"records": [],
"providers": [],
"summary_rows": [],
"request_errors_by_provider": {},
"toolcall_errors_by_provider": {},
}
def _extract_error_json(error_value):
if isinstance(error_value, (dict, list)):
return error_value
if isinstance(error_value, str):
# Best-effort: if it's JSON, parse it; else return wrapped
try_parse = None
if error_value.startswith("{") or error_value.startswith("["):
try_parse = json.loads(error_value)
return try_parse if try_parse is not None else {"error": error_value}
return {"error": str(error_value)}
def _truncate(s, length=200):
if not isinstance(s, str):
return s
return s if len(s) <= length else s[: length - 1] + "\u2026"
def _compute_summary(records):
by_provider = defaultdict(lambda: {
"total": 0,
"request_errors": 0,
"tool_calls_present": 0,
"tool_call_invalid": 0,
"success": 0,
})
request_errors_by_provider = defaultdict(list)
toolcall_errors_by_provider = defaultdict(list)
for r in records:
provider = r.get("provider")
if provider is None:
continue
stats = by_provider[provider]
stats["total"] += 1
finish_reason = r.get("finish_reason")
has_error_field = bool(r.get("error"))
is_request_error = finish_reason == "error" or has_error_field
tool_calls_present = bool(r.get("tool_calls_present"))
tool_calls_valid = bool(r.get("tool_calls_valid")) if tool_calls_present else True
is_tool_call_invalid = tool_calls_present and (not tool_calls_valid)
if is_request_error:
stats["request_errors"] += 1
request_errors_by_provider[provider].append({
"sample_id": r.get("sample_id"),
"model": r.get("model"),
"finish_reason": finish_reason,
"error": r.get("error"),
"_full_record": r,
})
if tool_calls_present:
stats["tool_calls_present"] += 1
if is_tool_call_invalid:
stats["tool_call_invalid"] += 1
toolcall_errors_by_provider[provider].append({
"sample_id": r.get("sample_id"),
"model": r.get("model"),
"finish_reason": finish_reason,
"error": r.get("error"),
"_full_record": r,
})
is_success = (not is_request_error) and (not is_tool_call_invalid)
if is_success:
stats["success"] += 1
summary_rows = []
for provider, s in by_provider.items():
total = s["total"] if s["total"] > 0 else 1
req_err_rate = s["request_errors"] / total
tc_err_rate = (s["tool_call_invalid"] / s["tool_calls_present"]) if s["tool_calls_present"] > 0 else 0.0
success_rate = s["success"] / total
summary_rows.append({
"provider": provider,
"total": s["total"],
"request_errors": s["request_errors"],
"request_error_rate": round(req_err_rate, 3),
"tool_calls_present": s["tool_calls_present"],
"tool_call_invalid": s["tool_call_invalid"],
"tool_call_error_rate": round(tc_err_rate, 3),
"success": s["success"],
"success_rate": round(success_rate, 3),
})
summary_rows.sort(key=lambda x: (-x["success_rate"], x["request_error_rate"], x["tool_call_error_rate"]))
providers = [row["provider"] for row in summary_rows]
return summary_rows, providers, request_errors_by_provider, toolcall_errors_by_provider
def load_results(file_path):
assert isinstance(file_path, str) and len(file_path) > 0
assert os.path.exists(file_path)
with open(file_path, "r") as f:
data = json.load(f)
records = data.get("records") or []
assert isinstance(records, list)
summary_rows, providers, req_errs, tc_errs = _compute_summary(records)
CURRENT["file_path"] = file_path
CURRENT["records"] = records
CURRENT["providers"] = providers
CURRENT["summary_rows"] = summary_rows
CURRENT["request_errors_by_provider"] = req_errs
CURRENT["toolcall_errors_by_provider"] = tc_errs
def pct_str(x):
x = x or 0.0
return f"{x * 100:.1f}%"
df_rows = [[
r["provider"],
r["total"],
r["request_errors"],
pct_str(r["request_error_rate"]),
r["tool_calls_present"],
r["tool_call_invalid"],
pct_str(r["tool_call_error_rate"]),
r["success"],
pct_str(r["success_rate"]),
] for r in summary_rows]
return (
gr.update(choices=providers, value=(providers[0] if providers else None)),
df_rows,
"Loaded",
[],
None,
)
def _build_error_rows(provider, error_type):
if not provider:
return []
if error_type == "Request errors":
errs = CURRENT["request_errors_by_provider"].get(provider) or []
else:
errs = CURRENT["toolcall_errors_by_provider"].get(provider) or []
rows = []
for e in errs:
preview = e.get("error")
rows.append([
e.get("sample_id"),
e.get("model"),
e.get("finish_reason"),
_truncate(preview if isinstance(preview, str) else json.dumps(preview) if preview is not None else ""),
])
return rows
def update_error_table(provider, error_type):
return _build_error_rows(provider, error_type), None
def on_error_select(evt: gr.SelectData, provider, error_type):
if not provider:
return None
idx = evt.index[0] if isinstance(evt.index, (list, tuple)) else evt.index
if error_type == "Request errors":
errs = CURRENT["request_errors_by_provider"].get(provider) or []
else:
errs = CURRENT["toolcall_errors_by_provider"].get(provider) or []
if not errs:
return None
if idx is None or idx < 0 or idx >= len(errs):
return None
rec = errs[idx]
error_json = _extract_error_json(rec.get("error"))
if isinstance(error_json, dict) and len(error_json) == 1 and "error" in error_json:
# If we couldn't parse a structured error, fall back to showing the full record
return rec.get("_full_record")
return error_json
def build_ui(default_path):
with gr.Blocks(title="BFCL Viewer", css=".gradio-container {max-width: 1200px}") as demo:
gr.Markdown("## BFCL Results Viewer")
with gr.Row():
file_in = gr.Textbox(label="Results JSON path", value=default_path, scale=4)
load_btn = gr.Button("Load", scale=1)
status = gr.Markdown("")
summary_df = gr.Dataframe(
headers=[
"Provider", "Total", "Req Errs", "Req Err %",
"Tool Calls", "TC Invalid", "TC Err %",
"Success", "Success %",
],
value=[],
interactive=False,
wrap=True,
label="Provider reliability summary",
)
with gr.Row():
provider_dd = gr.Dropdown(choices=[], label="Provider", interactive=True)
err_type = gr.Radio(["Request errors", "Tool-call errors"], value="Request errors", label="Error type")
errors_df = gr.Dataframe(
headers=["sample_id", "model", "finish_reason", "error_preview"],
value=[],
interactive=False,
wrap=True,
label="Errors",
)
error_json = gr.JSON(label="Selected error (JSON)")
load_btn.click(
fn=load_results,
inputs=[file_in],
outputs=[provider_dd, summary_df, status, errors_df, error_json],
)
provider_dd.change(
fn=update_error_table,
inputs=[provider_dd, err_type],
outputs=[errors_df, error_json],
)
err_type.change(
fn=update_error_table,
inputs=[provider_dd, err_type],
outputs=[errors_df, error_json],
)
errors_df.select(
fn=on_error_select,
inputs=[provider_dd, err_type],
outputs=[error_json],
)
return demo
DEFAULT_PATH = "./bfcl_inference_results_20251029_154417.json"
if __name__ == "__main__":
app = build_ui(DEFAULT_PATH)
app.launch()
|