thinkwee
commited on
Commit
·
e12cf59
1
Parent(s):
655137d
update
Browse files
charts.js
CHANGED
|
@@ -561,6 +561,42 @@ document.querySelectorAll('.ranking-dim').forEach(btn => {
|
|
| 561 |
// ============================================================================
|
| 562 |
// TURN DISTRIBUTION - 3 Charts (Ridgeline style)
|
| 563 |
// ============================================================================
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 564 |
function initTurnCharts() {
|
| 565 |
const scenarios = ['mimic', '10k', 'globem'];
|
| 566 |
|
|
@@ -603,6 +639,7 @@ function initTurnCharts() {
|
|
| 603 |
displayData.forEach((model, idx) => {
|
| 604 |
const color = getModelColor(model.model);
|
| 605 |
const yOffset = idx;
|
|
|
|
| 606 |
|
| 607 |
// Scale distribution to fit in the row (max height ~0.8)
|
| 608 |
const maxDist = Math.max(...model.distribution) || 1;
|
|
@@ -616,9 +653,9 @@ function initTurnCharts() {
|
|
| 616 |
fill: 'toself',
|
| 617 |
fillcolor: color + '40', // 25% opacity
|
| 618 |
line: { color: color, width: 1.5 },
|
| 619 |
-
name:
|
| 620 |
text: model.distribution.map((d, i) =>
|
| 621 |
-
`${
|
| 622 |
),
|
| 623 |
hovertemplate: '%{text}<extra></extra>',
|
| 624 |
showlegend: false
|
|
@@ -647,7 +684,7 @@ function initTurnCharts() {
|
|
| 647 |
...darkLayout.yaxis,
|
| 648 |
tickmode: 'array',
|
| 649 |
tickvals: displayData.map((_, i) => i),
|
| 650 |
-
ticktext: displayData.map(m =>
|
| 651 |
automargin: true,
|
| 652 |
range: [-0.5, displayData.length]
|
| 653 |
},
|
|
|
|
| 561 |
// ============================================================================
|
| 562 |
// TURN DISTRIBUTION - 3 Charts (Ridgeline style)
|
| 563 |
// ============================================================================
|
| 564 |
+
// Turn distribution display name mapping
|
| 565 |
+
const TURN_DISPLAY_NAMES = {
|
| 566 |
+
'run_api_deepseek_deepseek-chat': 'DeepSeek-V3.2',
|
| 567 |
+
'qwen3-next-80b-a3b-instruct': 'Qwen3-Next-80BA3B',
|
| 568 |
+
'qwen3-next-80b-a3b-instruct-note': 'Qwen3-Next-80BA3B-Note',
|
| 569 |
+
'qwen3-next-80b-a3b-instruct-noreasoning': 'Qwen3-Next-80BA3B-NoR',
|
| 570 |
+
'qwen3-next-80b-a3b-instruct-longreasoning': 'Qwen3-Next-80BA3B-LR',
|
| 571 |
+
'qwen3-next-80b-a3b-instruct-shortreasoning': 'Qwen3-Next-80BA3B-SR',
|
| 572 |
+
'qwen2.5-14B-Instruct-1M': 'Qwen2.5-14B-1M',
|
| 573 |
+
'qwen2.5-7B-Instruct-1M': 'Qwen2.5-7B-1M',
|
| 574 |
+
'qwen2.5-14B-Instruct': 'Qwen2.5-14B',
|
| 575 |
+
'qwen2.5-7B-Instruct': 'Qwen2.5-7B',
|
| 576 |
+
'qwen2.5-72B-Instruct': 'Qwen2.5-72B',
|
| 577 |
+
'qwen2.5-32b-instruct': 'Qwen2.5-32B',
|
| 578 |
+
'qwen3-4B-Instruct-2507': 'Qwen3-4B',
|
| 579 |
+
'gemini2.5-flash-lite': 'Gemini2.5-Flash-Lite',
|
| 580 |
+
'gemini2.5-flash': 'Gemini2.5-Flash',
|
| 581 |
+
'gemini2.5-pro': 'Gemini2.5-Pro',
|
| 582 |
+
'claude4.5-sonnet': 'Claude4.5-Sonnet',
|
| 583 |
+
'llama3.3-70B': 'Llama3.3-70B',
|
| 584 |
+
'llama-3.3-70B': 'Llama3.3-70B',
|
| 585 |
+
'minimax-m2': 'MiniMax-M2',
|
| 586 |
+
'gpt5mini': 'GPT-5-mini',
|
| 587 |
+
'gpt5-mini': 'GPT-5-mini',
|
| 588 |
+
'gpt5.1': 'GPT-5.1',
|
| 589 |
+
'gpt5.2': 'GPT-5.2',
|
| 590 |
+
'kimi-k2': 'Kimi-K2',
|
| 591 |
+
'glm4.6': 'GLM-4.6',
|
| 592 |
+
'qwen3': 'Qwen3-30B-A3B',
|
| 593 |
+
'gemini3-flash': 'Gemini3-Flash',
|
| 594 |
+
};
|
| 595 |
+
|
| 596 |
+
function getTurnDisplayName(model) {
|
| 597 |
+
return TURN_DISPLAY_NAMES[model] || model;
|
| 598 |
+
}
|
| 599 |
+
|
| 600 |
function initTurnCharts() {
|
| 601 |
const scenarios = ['mimic', '10k', 'globem'];
|
| 602 |
|
|
|
|
| 639 |
displayData.forEach((model, idx) => {
|
| 640 |
const color = getModelColor(model.model);
|
| 641 |
const yOffset = idx;
|
| 642 |
+
const displayName = getTurnDisplayName(model.model);
|
| 643 |
|
| 644 |
// Scale distribution to fit in the row (max height ~0.8)
|
| 645 |
const maxDist = Math.max(...model.distribution) || 1;
|
|
|
|
| 653 |
fill: 'toself',
|
| 654 |
fillcolor: color + '40', // 25% opacity
|
| 655 |
line: { color: color, width: 1.5 },
|
| 656 |
+
name: displayName,
|
| 657 |
text: model.distribution.map((d, i) =>
|
| 658 |
+
`${displayName}<br>${binLabels[i]} turns: ${d.toFixed(1)}%<br>Median: ${model.median}`
|
| 659 |
),
|
| 660 |
hovertemplate: '%{text}<extra></extra>',
|
| 661 |
showlegend: false
|
|
|
|
| 684 |
...darkLayout.yaxis,
|
| 685 |
tickmode: 'array',
|
| 686 |
tickvals: displayData.map((_, i) => i),
|
| 687 |
+
ticktext: displayData.map(m => getTurnDisplayName(m.model)),
|
| 688 |
automargin: true,
|
| 689 |
range: [-0.5, displayData.length]
|
| 690 |
},
|
data.js
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|