FRANK-ZERO-38B

<span style="font-weight: bold; color: red;">FRANK-ZERO</span>: a training-FRee And Native r1-liKe multimodal LLMs

[📂 GitHub]

Introduction

We have successfully migrated r1-like reasoning abilities into the vision-language joint space. Based on a unimodal language reasoning model (QwQ-32B) and a multimodal non-reasoning model (InternVL-38B), we constructed a training-FRee And Native r1-liKe multimodal reasoning model, <span style="font-weight: bold; color: red;">FRANK-ZERO</span> 38B. It achieves an accuracy of 74.1% on the MathVista test-mini benchmark, surpassing the closed-source OpenAI o1 model (73.9% vs. 74.1%), and demonstrates excellent long chain-of-thought reasoning and reflection capabilities.

Example

Describe this image in detail.

<span style="font-weight: bold; color: red;">FRANK-ZERO</span> output:

<think>

Okay, let me look at this chart. The title says it's about average precipitation in Christchurch, New Zealand. The x-axis has the months from January to December, and the y-axis is precipitation in millimeters. There are different colored bars for each month, each representing a different value.

First, I need to figure out what each color corresponds to. The key might be missing, but usually in these charts, each color represents a different category. Wait, actually, looking at the chart, each month has a single bar with a different color. Maybe each color is just a different month's data? Hmm, maybe not. Let me check again.

Wait, the title mentions "Average precipitation in Christchurch, New Zealand," so it's likely that each bar represents the average precipitation for each month. The colors might just be for visual distinction between the months. Let me see the values.

Starting from January: the bar is purple, and the value is around 53 mm. February is green, around 47 mm. March is red, about 53 mm. April is blue, around 51 mm. May is orange, the highest at 69 mm. June is light blue, around 67 mm. July is light green, also around 67 mm. August is teal, around 54 mm. September is pink, around 46 mm. October is dark green, around 46 mm. November is red, around 48 mm. December is blue, around 57 mm.

Wait, but the colors might actually represent different categories. Let me think again. The title says "Average precipitation," but maybe each color represents a different type of precipitation? Or perhaps different years? But the x-axis is months, so probably each bar is a month's precipitation, with colors just for differentiation.

Alternatively, maybe the chart is comparing different types of precipitation across months? But the title doesn't mention that. It just says average precipitation. Hmm, maybe the colors are just random. Let me check the values again.

Looking at the numbers:

January: ~53 mm (purple)
February: ~47 mm (green)
March: ~53 mm (red)
April: ~51 mm (blue)
May: ~69 mm (orange)
June: ~67 mm (light blue)
July: ~67 mm (light green)
August: ~54 mm (teal)
September: ~46 mm (pink)
October: ~46 mm (dark green)
November: ~48 mm (red)
December: ~57 mm (blue)

So the highest precipitation is in May at 69 mm, followed by June and July at 67 mm. The lowest is in September and October at 46 mm. The chart shows that precipitation peaks in late spring and early summer, then decreases towards autumn and winter.

The question is probably asking to describe the trend. The chart shows that precipitation is highest in May, June, and July, then decreases in August, and continues to drop in September and October. November and December see a slight increase again but not as high as the summer months.

So the main point is that the highest precipitation occurs in the late spring and early summer months (May to July), with the lowest in autumn (September and October). The rest of the months have moderate precipitation levels.

I need to make sure I didn't misinterpret the colors. Since each month has a different color, it's just for visual separation. The key might have been omitted, but the data is clear from the bars' heights.

Another thing to note is that the chart uses different colors for each month, which might be unnecessary but just for aesthetics. The important part is the height of each bar indicating precipitation in mm.

So the answer should summarize that precipitation peaks in May, June, and July, then decreases, with the lowest in September and October. The rest of the months have varying levels but not as extreme.

</think>

The bar chart displays the average precipitation in Christchurch, New Zealand, across 12 months. Each month is represented by a differently colored bar, with the height indicating precipitation in millimeters (mm). 

- **Key Observations:**
  - **Highest Precipitation:** May (69 mm), June (67 mm), and July (67 mm) show the highest precipitation, indicating a peak during late spring and early summer.
  - **Lowest Precipitation:** September (46 mm) and October (46 mm) have the lowest precipitation, marking a significant drop in autumn.
  - **Moderate Levels:** Other months (January to April, August, November, and December) have precipitation ranging between 46–57 mm, with some fluctuations but no extreme highs or lows.

- **Trend:** The chart reveals a clear seasonal pattern where precipitation peaks in the late spring/early summer months (May–July), decreases sharply in autumn (September–October), and stabilizes at moderate levels during the rest of the year. This suggests Christchurch experiences its wettest period from May to July and driest in September and October.

Overall, the data highlights seasonal variations in precipitation, with the highest rainfall occurring in the early part of the year and the lowest in late autumn.

License

This project is released under the QwQ-32B (License: Apache-2.0) and InternVL2.5 (License: MIT) licenses. Portions of this project contain code and models from other sources, which are subject to their respective licenses.

Acknowledgement

This code base is mainly built upon InternVL2.5. Thanks for their awesome work!

We would also like to recognize and commend the following open source projects (e.g., Qwen2.5VL, QwQ-32B), thank you for your great contribution to the open source community.