tuieval 0.2.0.dev65__tar.gz → 0.2.0.dev66__tar.gz

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/CHANGELOG.md

@@ -7,7 +7,7 @@
 - The run's ETA adds up each pack's answers left × that model's time per answer there (its history, then its own times once it has 5 answers), instead of extrapolating from the first few answers.
 - Faster start on Macs: the GPU core count comes from `ioreg` instead of `system_profiler`, which took seconds on some Macs.
 - A model a server won't serve (HTTP 401-404, e.g. OpenRouter's "No endpoints found") is never judged on it: those answers don't count, and the model stops at once instead of failing every question.
-- PTA index: parsimony (tokens), time and accuracy per model as bars, with tokens and time shown as multiples of the best (2.4×): the first Results tab (where `r` lands), `tuieval pta` and the report. Server errors aren't answers, and models with few answers are left out unless picked.
+- PTA index: parsimony (tokens), time and accuracy per model as bars (tokens and time as long as they took, shorter is better; accuracy longer is better), with tokens and time also shown as multiples of the best (2.4×): the first Results tab (where `r` lands), `tuieval pta` and the report. Server errors aren't answers, and models with few answers are left out unless picked.
 - Run several models at a time: `parallel_models` per machine, **Models at a time** in the TUI, or `tuieval run --parallel N` (default 1). Models wait for memory to fit, and answers note what ran alongside them.
 - More evals during a run: `n` opens Setup, and Start adds them to the run going (same tier) or queues them after it.
 - A new workspace's `models.toml` is 23 lines: the llama, local and OpenRouter servers and the defaults are built in (docs/models.md lists them in full; a server defined in `models.toml` replaces the built-in one with that name). Existing workspaces keep their settings; built-in servers they don't define become available, and adding a model still prefers the workspace's own servers.

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: tuieval
-Version: 0.2.0.dev65
+Version: 0.2.0.dev66
 Summary: Evaluate local and frontier LLMs on your own questions: accuracy, speed, tokens and PASS/FAIL verdicts, in the terminal.
 Project-URL: Homepage, https://github.com/ashe-wb/tuieval
 Project-URL: Issues, https://github.com/ashe-wb/tuieval/issues

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/docs/models.md

@@ -98,13 +98,13 @@ By default a run serves one model at a time. On a machine with room for more, se
 
 ## PTA index
 
-The **PTA index** puts parsimony (tokens), time and accuracy side by side, as a bar for each per model (longer is better), with a table underneath. It's the first tab in Results, where `r` opens, and it's also `tuieval pta` and a section in `tuieval report`. There's no combined number; verdicts and critical failures stay in Production readiness.
+The **PTA index** puts parsimony (tokens), time and accuracy side by side, as a bar for each per model, with a table underneath. It's the first tab in Results, where `r` opens, and it's also `tuieval pta` and a section in `tuieval report`. There's no combined number; verdicts and critical failures stay in Production readiness.
 
 - **P parsimony (tokens)**: the tokens used, reasoning and answer, on every question compared (each question at its median over repeats). Beside the bar: how many times the leanest model's, e.g. `2.4×` (`1.0×` is the leanest, ★ in the table's **tokens/answer**). Unlike time it doesn't depend on your machine: it's how much a model says to get to an answer, which is what you pay for on an API and what fills its context.
 - **T time**: the total time to answer those questions, as how many times the fastest model's, e.g. `28×` (★ in **time/answer**).
 - **A accuracy**: the share of answers to those questions that passed.
 
-The P and T bars use a log scale: full for the best, 15 points (of 100) shorter for every doubling, empty at about 100 times the best. So a model at 28× still shows a bar, and adding a better model shortens the others'.
+The P and T bars are as long as the tokens and time used: the model that used the most fills its bar, the others in proportion (at least one block), so **shorter is better**. The A bar is the share of answers right, so **longer is better**; it's full only when every answer passed. Each heading says which way is better.
 
 Models are compared on the questions all of them answered, so one that skipped some never looks faster. Server and connection errors aren't answers: a model with only those is listed as having none. A model with fewer than half as many answers as the most-answered one is left out (and listed), or it would shrink the shared questions for everyone. Pick the models and pack to compare in the TUI, or use `--only` and `--packs`: models you pick are always compared.
 

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/src/tuieval/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.2.0.dev65'
-__version_tuple__ = version_tuple = (0, 2, 0, 'dev65')
+__version__ = version = '0.2.0.dev66'
+__version_tuple__ = version_tuple = (0, 2, 0, 'dev66')
 
 __commit_id__ = commit_id = None

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/src/tuieval/pta.py

@@ -1,4 +1,4 @@
-"""PTA index: parsimony (tokens), time and accuracy per model, a bar each (longer = better).
+"""PTA index: parsimony (tokens), time and accuracy per model, a bar each.
 
     P  parsimony (tokens)  the tokens (reasoning and answer) used on every question compared (each
                            question at its median over repeats), shown as how many times the leanest
@@ -7,8 +7,10 @@
                            model's ("28×").
     A  accuracy            the share of answers to those questions that passed (%).
 
-Behind the P and T bars is a 0-100 score: 100 for the best, POINTS_PER_DOUBLING points less for every
-doubling (0 at about 100 times more), so models far behind the best still get a visible bar.
+The P and T bars are as long as the tokens and time used (the model that used the most fills its bar,
+the others in proportion, at least one block), so shorter is better; the A bar: longer is better.
+P and T also carry a 0-100 score (100 for the best, POINTS_PER_DOUBLING points less for every
+doubling), used to order models with the same accuracy.
 
 Models are compared on the questions all of them answered, so one that skipped some never looks
 faster. Server and connection errors aren't answers. By default a model with fewer than half as many
@@ -93,9 +95,11 @@ def index(rows, models=None, memory=None):
             "most": most}
 
 
-# (letter, name, what the number beside the bar shows, colour); the bar itself is the 0-100 score
-LETTERS = (("P", "parsimony (tokens)", "tokens_x", "cyan"), ("T", "time", "time_x", "magenta"),
-           ("A", "accuracy", "A", "green"))
+# (letter, name, which way is better, what the bar measures, the number beside it, colour). P and T bars
+# are the amount itself (tokens, time: shorter is better); the A bar is the share right (longer is better).
+LETTERS = (("P", "parsimony (tokens)", "▼ shorter is better", "tokens", "tokens_x", "cyan"),
+           ("T", "time", "▼ shorter is better", "total_s", "time_x", "magenta"),
+           ("A", "accuracy", "▲ longer is better", "A", "A", "green"))
 
 
 def times(v):
@@ -103,26 +107,41 @@ def times(v):
     return "-" if v is None else f"{v:.1f}×" if v < 9.95 else f"{v:.0f}×"
 
 
+def _blocks(width, v, most):
+    """How many of width blocks a bar fills: v as a share of most, at least one block when v > 0."""
+    if not v or not most:
+        return 0
+    return max(1, min(width, round(width * v / most)))
+
+
 def bars(result, width=20, markup=True):
-    """Lines with a row per model and a bar per score (P, T, A), longer = better: the at-a-glance view.
-    Beside P and T, how many times the leanest / fastest model's tokens / time; beside A, the percentage.
-    markup=False gives plain text (the report)."""
+    """Lines with a row per model and a bar each for P, T and A: the at-a-glance view. The P and T bars
+    are proportional to tokens and time (the longest is the model that used most; shorter is better),
+    with how many times the leanest / fastest model's beside them; the A bar is the share of answers
+    right (longer is better, full only when every answer passed). markup=False gives plain text."""
     if not result["models"]:
         return []
     name_w = max(len(x["model"]) for x in result["models"])
-    # each column: the bar, a space, "100%" or "9.8×" (5), then a gap; at least as wide as its heading
-    cell_w = max(width + 9, *(len(f"{k} {name}") + 2 for k, name, _, _ in LETTERS))
-    head = " " * (name_w + 2) + "".join(f"{f'{k} {name}':<{cell_w}}" for k, name, _, _ in LETTERS)
-    lines = [f"[b]{head.rstrip()}[/b]" if markup else head.rstrip()]
+    # each column: the bar, a space, "100%" or "9.8×" (6), then a gap; at least as wide as its heading
+    cell_w = max(width + 9, *(len(t) + 2 for k, name, better, *_ in LETTERS for t in (f"{k} {name}", better)))
+    pad = " " * (name_w + 2)
+    head = pad + "".join(f"{f'{k} {name}':<{cell_w}}" for k, name, *_ in LETTERS)
+    sub = pad + "".join(f"{better:<{cell_w}}" for _, _, better, *_ in LETTERS)
+    lines = [f"[b]{head.rstrip()}[/b]" if markup else head.rstrip(), f"[dim]{sub.rstrip()}[/dim]" if markup else sub.rstrip()]
+    most = {m: max((x.get(m) or 0 for x in result["models"]), default=0) for _, _, _, m, _, _ in LETTERS}
     for x in result["models"]:
         cells = []
-        for k, _, shown, color in LETTERS:
-            v = x[k]
-            full = 0 if v is None else round(width * max(0, min(100, v)) / 100)
-            if v is not None and v < 100:   # full only when perfect (accuracy) or the best (P, T)
-                full = min(full, width - 1)
+        for k, _, _, measure, shown, color in LETTERS:
+            if k == "A":
+                v = x["A"]
+                full = 0 if v is None else round(width * max(0, min(100, v)) / 100)
+                if v is not None and v < 100:   # full only when every answer passed
+                    full = min(full, width - 1)
+                num = "-" if v is None else compare.pct(v / 100)
+            else:
+                full = _blocks(width, x.get(measure), most[measure])
+                num = times(x.get(shown))
             bar, rest = "█" * full, "░" * (width - full)
-            num = (("-" if v is None else compare.pct(v / 100)) if shown == "A" else times(x.get(shown)))
             bar = f"[{color}]{bar}[/][grey37]{rest}[/]" if markup else bar + rest
             cells.append(f"{bar} {num:>6}" + " " * (cell_w - width - 7))
         name = f"{x['model']:<{name_w}}"
@@ -183,8 +202,9 @@ def scope_note(result, models):
     if not result["questions"]:
         return "These models have no question in common yet, so there's nothing to compare."
     return (f"Compared on the {result['questions']} questions all {len(result['models'])} model(s) answered. "
-            "Longer bars are better. P parsimony (tokens) and T time: how many times the leanest / fastest "
-            "model's (1.0×, ★ in the table). A accuracy: answers right.")
+            "P parsimony (tokens) and T time: bars as long as the tokens / time used (shorter is better), "
+            "and how many times the leanest / fastest model's (1.0×, ★ in the table). A accuracy: answers right "
+            "(longer is better).")
 
 
 def for_engine(e, labels=None, packs=None, results_dir=None):

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/src/tuieval/tui.py

@@ -332,9 +332,10 @@ with the model's reasoning and answer as they stream, then recent results with t
   v  with several models at a time: stream the next one (k skips the one streaming)""",
     "ResultsScreen": """[b]Results[/b]
 
-  PTA index              parsimony (tokens), time and accuracy per model, a bar each (longer = better;
-                         2.4× = 2.4 times the best), with per-answer numbers, tok/s, TTFT and memory in
-                         the table. Results open here.
+  PTA index              parsimony (tokens), time and accuracy per model, a bar each: tokens and time
+                         as long as they took (shorter is better; 2.4× = 2.4 times the best), accuracy
+                         longer is better. Per-answer numbers, tok/s, TTFT and memory in the table.
+                         Results open here.
   Production readiness   one verdict per model and use case; the table says why and what to run next
   Per question           every question, model by model (filters: pack, level, Models disagree)
   Failures               every wrong answer with the grader's reason; enter opens it in full

{tuieval-0.2.0.dev65 → tuieval-0.2.0.dev66}/tests/test_tuieval.py

@@ -761,10 +761,14 @@ class PTAIndex(unittest.TestCase):
         self.assertEqual((cells["slow"][1], cells["fast"][1:3]), ("10 ★", ["40", "2.0s ★"]))   # per answer
         bars = pta.bars(res, 10, markup=False)
         self.assertEqual(bars[0].split(), ["P", "parsimony", "(tokens)", "T", "time", "A", "accuracy"])
+        self.assertEqual(bars[1].split(), ["▼", "shorter", "is", "better"] * 2 + ["▲", "longer", "is", "better"])
+        slow, fast = bars[2], bars[3]
         self.assertEqual((bars[0].index("P parsimony"), bars[0].index("T time")),   # each bar starts under
-                         (bars[2].index("███████░░░"), bars[2].index("██████████")))  # its heading
-        # the bar is the score (4x the tokens: 70); beside it, how many times the best
-        self.assertEqual(bars[2].split(), ["fast", "███████░░░", "4.0×", "██████████", "1.0×", "█████░░░░░", "50.0%"])
+                         (fast.index("██████████"), fast.index("███░░░░░░░")))       # its heading
+        # P and T bars as long as tokens and time (shorter is better): fast used 4x the tokens of slow but
+        # took 4 s against 13 s, so it has the longest token bar and the shortest time bar
+        self.assertEqual(fast.split(), ["fast", "██████████", "4.0×", "███░░░░░░░", "1.0×", "█████░░░░░", "50.0%"])
+        self.assertEqual(slow.split()[1:5], ["██░░░░░░░░", "1.0×", "██████████", "3.2×"])
         self.assertEqual((pta.times(1), pta.times(9.84), pta.times(27.9)), ("1.0×", "9.8×", "28×"))
         self.assertEqual((by["fast"]["A"], by["slow"]["A"]), (50, 100))
         self.assertEqual([x["model"] for x in res["models"]], ["slow", "fast"])   # best accuracy first
@@ -774,7 +778,7 @@ class PTAIndex(unittest.TestCase):
         self.assertEqual([compare.pct(f) for f in (467 / 469, 1, 0, 0.5)], ["99.6%", "100%", "0%", "50.0%"])
         row = lambda model, a: {"model": model, "P": 100, "T": 100, "A": a, "tokens_x": 1, "time_x": 1}
         bars = pta.bars({"models": [row("two-wrong", 100 * 467 / 469), row("perfect", 100.0)]}, 10, markup=False)
-        two_wrong, perfect = bars[1].split(), bars[2].split()
+        two_wrong, perfect = bars[2].split(), bars[3].split()
         self.assertEqual((two_wrong[-2], two_wrong[-1]), ("█████████░", "99.6%"))   # not a full bar
         self.assertEqual((perfect[-2], perfect[-1]), ("██████████", "100%"))