tuieval 0.2.0.dev74__py3-none-any.whl → 0.2.0.dev75__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
tuieval/_version.py CHANGED
@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
18
18
  commit_id: str | None
19
19
  __commit_id__: str | None
20
20
 
21
- __version__ = version = '0.2.0.dev74'
22
- __version_tuple__ = version_tuple = (0, 2, 0, 'dev74')
21
+ __version__ = version = '0.2.0.dev75'
22
+ __version_tuple__ = version_tuple = (0, 2, 0, 'dev75')
23
23
 
24
24
  __commit_id__ = commit_id = None
tuieval/tui.py CHANGED
@@ -2239,9 +2239,10 @@ class RunScreen(Screen):
2239
2239
  self.req_started, self.req_tokens, self.req_first = time.time(), 0, None
2240
2240
  job = d["job"]
2241
2241
  rep_note = f" · repeat {d['repeat'] + 1}/{job.repeat}" if job.repeat > 1 else ""
2242
- self.current = (f"[b]{job.label}[/b] · {job.pack.label} · test {job.done + 1}/{job.total}{rep_note}"
2242
+ self.current = (f"[b]{rich_escape(job.label)}[/b] · {rich_escape(job.pack.label)} · test "
2243
+ f"{job.done + 1}/{job.total}{rep_note}"
2243
2244
  f"{' [magenta]🖼 image[/magenta]' if d['has_image'] else ''}\n"
2244
- f"{diff_badge(d.get('difficulty', 'unrated'))} {d['test']}")
2245
+ f"{diff_badge(d.get('difficulty', 'unrated'))} {rich_escape(d['test'])}")
2245
2246
  elif kind == "request_done":
2246
2247
  if self.watched(kind, d):
2247
2248
  self.flush_streams()
@@ -2251,12 +2252,12 @@ class RunScreen(Screen):
2251
2252
  # follow new rows unless you're browsing an older one
2252
2253
  at_end = not recent.has_focus or recent.cursor_row >= recent.row_count - 1
2253
2254
  calls = f" [cyan]→ {r['tool_calls'][0]['name']}()[/cyan]" if r.get("tool_calls") else ""
2254
- row_key = recent.add_row(mark, diff_badge(r.get("difficulty", "unrated")), r["description"][:50], job.label,
2255
+ row_key = recent.add_row(mark, diff_badge(r.get("difficulty", "unrated")), rich_escape(r["description"][:50]), job.label,
2255
2256
  str(r.get("completion_tokens") or ""),
2256
2257
  (f"{r['ttft_s']:.1f}" + (" ⟲" if r.get("cached_tokens") else ""))
2257
2258
  if r.get("ttft_s") is not None else "",
2258
2259
  f"{r['gen_tps']:.0f}" if r.get("gen_tps") else "", f"{r['total_s']:.0f}",
2259
- r["reason"][:90] + calls
2260
+ rich_escape(r["reason"][:90]) + calls
2260
2261
  + (f" [red]copy of repeat {r['identical_to_repeat'] + 1}[/red]"
2261
2262
  if r.get("identical_to_repeat") is not None else ""))
2262
2263
  self.stats.append(r)
@@ -2364,7 +2365,7 @@ class RunScreen(Screen):
2364
2365
  self.event("[bold yellow]REUSE[/bold yellow] " + msg)
2365
2366
  self.notify(msg, severity="warning", timeout=20)
2366
2367
  elif kind == "identical_answer":
2367
- msg = (f"{d['label']} · {d['pack']} · {d['test']}: repeat {d['repeat'] + 1} is an exact copy of repeat "
2368
+ msg = (f"{d['label']} · {d['pack']} · {rich_escape(d['test'])}: repeat {d['repeat'] + 1} is an exact copy of repeat "
2368
2369
  f"{d['same'] + 1}, so the server may be returning cached answers rather than fresh ones.")
2369
2370
  self.event("[bold red]COPY[/bold red] " + msg)
2370
2371
  self.notify(msg, severity="error", timeout=20)
@@ -2711,7 +2712,7 @@ class ResultsScreen(Screen):
2711
2712
  # compare.failures lists the failed rows in order, so they line up with fail_rows
2712
2713
  self.fail_rows = [r for r in rows if not r["ok"]]
2713
2714
  for i, (model, test, reason, level) in enumerate(compare.failures(rows)):
2714
- f.add_row(model, diff_badge(level), test, rich_escape(reason[:130]), key=str(i))
2715
+ f.add_row(model, diff_badge(level), rich_escape(test), rich_escape(reason[:130]), key=str(i))
2715
2716
  self.fill_per_question()
2716
2717
  self.fill_pta()
2717
2718
  self.fill_readiness(results_dir)
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: tuieval
3
- Version: 0.2.0.dev74
3
+ Version: 0.2.0.dev75
4
4
  Summary: Evaluate local and frontier LLMs on your own questions: accuracy, speed, tokens and PASS/FAIL verdicts, in the terminal.
5
5
  Project-URL: Homepage, https://github.com/ashe-wb/tuieval
6
6
  Project-URL: Issues, https://github.com/ashe-wb/tuieval/issues
@@ -1,6 +1,6 @@
1
1
  tuieval/__init__.py,sha256=myVXVX7htbwezW6k3x_FuKgn1RSZ2WcTaDEMByGn6Wg,312
2
2
  tuieval/__main__.py,sha256=E6Gls0DNz8GQK2K-kOUIx8cYhgANW_CH54VKrfCfs14,52
3
- tuieval/_version.py,sha256=V5PkcAfWcwrKTUa1VWZBAMieeerSM4tJioAs-6yJ5Fs,535
3
+ tuieval/_version.py,sha256=Kqbd4oEF7L3v_nXB493p78wA9p35xQCT7-H7VeKPrLE,535
4
4
  tuieval/builtin.toml,sha256=Kxw22RZCJhURUyu5Wz8tj5Y1nM85A8hFVRbA4ocoPUg,4293
5
5
  tuieval/cli.py,sha256=JGtN4J6d4cr9G7OfBxDiNtxvVEew9MgUhJBlYKYPhJ4,4687
6
6
  tuieval/client.py,sha256=kpsyLEZKfxCsvIGBHBbBy83TpI97CvnGCDq-4AbJS4Y,9986
@@ -18,7 +18,7 @@ tuieval/remove.py,sha256=Vwh43nm9lVb9U9aN-h6CMYbkTwdR3UWfZObwweF72n0,5157
18
18
  tuieval/run_evals.py,sha256=LtzXYWlPDdYKMWLmrmCpD6ff2ALLQRcUlgXGUdIFjQE,47005
19
19
  tuieval/scaffold.py,sha256=YBOqniJpf_GtKKAFF3_OHxxMGbmTWdFjcClGcOnrrKE,12130
20
20
  tuieval/selftest.py,sha256=nExkrOidgymlmvkxImEBfrQG5iTUN5WuRMIKtqusCCs,9835
21
- tuieval/tui.py,sha256=Xzt6SNi7sc-3GzII2qgtxl7hr0td75ZxzIV6XeP6TTg,172616
21
+ tuieval/tui.py,sha256=MUHk9EDMpqWa-dn-fuC2LZ9F6C_wJ6A2YTKyU6XcDdg,172739
22
22
  tuieval/tune.py,sha256=azxxthgJeag8JyoQUgaRxq6oI1XvEgzb9vkszjJZH7w,39645
23
23
  tuieval/verdict.py,sha256=VB-0LTxR9y8bJHT9RicxtE9K7cUHBdCu40hOCrsbmA4,19150
24
24
  tuieval/watch_proxy.py,sha256=bnDO1i6n6yUs1hODoUWcMppJMGPavoQqs6aNYrIxpT8,11416
@@ -47,8 +47,8 @@ tuieval/templates/packs/tool_call/pack.toml,sha256=n4bx0lGJHSmW41uVqhfIXJ2BWZpmT
47
47
  tuieval/templates/packs/tool_call/system.txt,sha256=mKxu-lB-0YjLHXdgeFJa8x2C180OvcNz5cHv2N5Bg3U,254
48
48
  tuieval/templates/packs/tool_call/tests.yaml,sha256=Nv7c4syrzJ9cU8EBKSK_1ZSFx9V02aCjW4-2TBCPT7s,1859
49
49
  tuieval/templates/packs/tool_call/tools.yaml,sha256=HgxCkh0On2VwbHSsjAGM8tNCK7ZtKLGrs7lMCn3Fprs,1048
50
- tuieval-0.2.0.dev74.dist-info/METADATA,sha256=iUzgPK31iuxQNBXgjxpZnCHJ9sIr_E_nt36t_QstQ9I,14215
51
- tuieval-0.2.0.dev74.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
52
- tuieval-0.2.0.dev74.dist-info/entry_points.txt,sha256=7eIhi1GpFxBP2OLJV02Nn8LwsNuO512eKOGt1IBbj04,45
53
- tuieval-0.2.0.dev74.dist-info/licenses/LICENSE,sha256=7wfRgGtBGH1aXhZkIn47oUznqKC18kEPMvUVgol7Ffc,1077
54
- tuieval-0.2.0.dev74.dist-info/RECORD,,
50
+ tuieval-0.2.0.dev75.dist-info/METADATA,sha256=btPP6J2uKuvCypZMPXQYoXwqWzTDFdalSpBJbPbx3fo,14215
51
+ tuieval-0.2.0.dev75.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
52
+ tuieval-0.2.0.dev75.dist-info/entry_points.txt,sha256=7eIhi1GpFxBP2OLJV02Nn8LwsNuO512eKOGt1IBbj04,45
53
+ tuieval-0.2.0.dev75.dist-info/licenses/LICENSE,sha256=7wfRgGtBGH1aXhZkIn47oUznqKC18kEPMvUVgol7Ffc,1077
54
+ tuieval-0.2.0.dev75.dist-info/RECORD,,