From 2334fc42aa5a69eeabec7e6c21be5b882966aa4c Mon Sep 17 00:00:00 2001
From: Akshay Tripathi
 <142379735+AkshayTripathiShorthillsAI@users.noreply.github.com>
Date: Thu, 7 May 2026 23:51:19 -0700
Subject: [PATCH 1/3] [Fix] Qwen3-VL: set do_sample explicitly to avoid
 sampling param conflicts

The generate_kwargs dict passed top_p, top_k, and temperature to
model.generate() without setting do_sample. HuggingFace defaults
do_sample to False, which silently ignores sampling parameters and
falls back to greedy decoding, causing evaluation results to not
reflect the configured sampling behavior.

This commit derives do_sample from temperature (do_sample=True when
temperature > 0) and only includes top_p, top_k, and temperature in
generate_kwargs when sampling is enabled, avoiding conflicting-flag
warnings in the greedy path. Users can override via kwargs.

Closes #1275
---
 vlmeval/vlm/qwen3_vl/model.py | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/vlmeval/vlm/qwen3_vl/model.py b/vlmeval/vlm/qwen3_vl/model.py
index 7f224371f..0c26fd8d8 100644
--- a/vlmeval/vlm/qwen3_vl/model.py
+++ b/vlmeval/vlm/qwen3_vl/model.py
@@ -74,13 +74,19 @@ def __init__(
         self.temperature = temperature
         if self.total_pixels and self.total_pixels > 24576 * 32 * 32:
             print('The total number of video tokens might too large, resulting in an overly long input sequence.')
+        do_sample = kwargs.pop('do_sample', temperature is not None and temperature > 0)
+
         self.generate_kwargs = dict(
             max_new_tokens=self.max_new_tokens,
-            top_p=top_p,
-            top_k=top_k,
-            temperature=temperature,
             repetition_penalty=repetition_penalty,
+            do_sample=do_sample,
         )
+        if do_sample:
+            self.generate_kwargs.update(
+                top_p=top_p,
+                top_k=top_k,
+                temperature=temperature,
+            )
         self.system_prompt = system_prompt
         self.verbose = verbose
         self.post_process = post_process

From d8e95021228a9f140c12096483837fed8cf71a2f Mon Sep 17 00:00:00 2001
From: Akshay Tripathi
 <142379735+SHAI-Akshay-Tripathi@users.noreply.github.com>
Date: Sat, 16 May 2026 19:35:00 +0530
Subject: [PATCH 2/3] Remove unused Polygon3 from requirements.txt

Removed polygon3 dependency from requirements.
---
 requirements.txt | 1 -
 1 file changed, 1 deletion(-)

diff --git a/requirements.txt b/requirements.txt
index ef9c78bbc..27c8a2014 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -38,7 +38,6 @@ openpyxl
 pandas
 pdf2image>=1.17.0
 pillow
-polygon3>=3.0.9.1
 portalocker
 protobuf
 pylatexenc==2.10

From 672428333bc5f474ff0b65c7a998f0ed0f0a888f Mon Sep 17 00:00:00 2001
From: Akshay Tripathi
 <142379735+SHAI-Akshay-Tripathi@users.noreply.github.com>
Date: Sat, 16 May 2026 19:36:09 +0530
Subject: [PATCH 3/3] Remove unused Polygon3 from Ocrbench_v2 requirements

Removed 'Polygon3' from the requirements.
---
 vlmeval/dataset/utils/Ocrbench_v2/requirements.txt | 1 -
 1 file changed, 1 deletion(-)

diff --git a/vlmeval/dataset/utils/Ocrbench_v2/requirements.txt b/vlmeval/dataset/utils/Ocrbench_v2/requirements.txt
index ecdff4394..801d02b3c 100644
--- a/vlmeval/dataset/utils/Ocrbench_v2/requirements.txt
+++ b/vlmeval/dataset/utils/Ocrbench_v2/requirements.txt
@@ -7,6 +7,5 @@ Levenshtein
 lxml
 nltk
 numpy
-Polygon3
 tqdm
 zss