xiaxinhang/pr-agent
1
0
Fork 0
mirror of https://github.com/qodo-ai/pr-agent.git synced 2025-07-04 21:00:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

docs: add Gemini-2.5-pro-preview model comparison to benchmark documentation

Browse Source
This commit is contained in:
mrT23
2025-05-12 08:11:57 +03:00
parent c5bcc21a61
commit bd911b67d8
1 changed files with 184 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files

259
docs/docs/finetuning_benchmark/index.md
View File

@ -1,102 +1,211 @@
# Qodo Merge Pull Request Benchmark # Qodo Merge Pull Request Benchmark
On coding tasks, the gap between open-source models and top closed-source models such as Claude and GPT is significant. ## Methodology
<br>
In practice, open-source models are unsuitable for most real-world code tasks, and require further fine-tuning to produce acceptable results.
_Qodo Merge pull request benchmark_ aims to benchmark models on their ability to be fine-tuned for a coding task. ...
Specifically, we chose to fine-tune open-source models on the task of analyzing a pull request, and providing useful feedback and code suggestions.
Here are the results: ## Gemini-2.5-pro-preview-05-06
<br>
<br>
**Model performance:** ### Model 'Gemini-2.5-pro-preview-05-06' vs 'Sonnet 3.7'
| Model name | Model size [B] | Better than gpt-4 rate, after fine-tuning [%] | ![Comparison](https://codium.ai/images/qodo_merge_benchmark/sonnet_37_vs_gemini-2.5-pro-preview-05-06_judge_o3.png){width=768}
|-----------------------------|----------------|----------------------------------------------|
| **DeepSeek 34B-instruct** | **34** | **40.7** |
| DeepSeek 34B-base | 34 | 38.2 |
| Phind-34b | 34 | 38 |
| Granite-34B | 34 | 37.6 |
| Codestral-22B-v0.1 | 22 | 32.7 |
| QWEN-1.5-32B | 32 | 29 |
| | | |
| **CodeQwen1.5-7B** | **7** | **35.4** |
| Llama-3.1-8B-Instruct | 8 | 35.2 |
| Granite-8b-code-instruct | 8 | 34.2 |
| CodeLlama-7b-hf | 7 | 31.8 |
| Gemma-7B | 7 | 27.2 |
| DeepSeek coder-7b-instruct | 7 | 26.8 |
| Llama-3-8B-Instruct | 8 | 26.8 |
| Mistral-7B-v0.1 | 7 | 16.1 |
<br> #### Analysis Summary
**Fine-tuning impact:** Model 'Gemini-2.5-pro-preview-05-06' is the stronger reviewer—more frequently identifies genuine, high-impact bugs and provides well-formed, actionable fixes. Model 'Sonnet 3.7' is safer against false positives and tends to be concise but often misses important defects or offers low-value or incorrect suggestions.
| Model name | Model size [B] | Fine-tuned | Better than gpt-4 rate [%] | See raw results [here](https://github.com/Codium-ai/pr-agent-settings/blob/main/benchmark/sonnet_37_vs_gemini-2.5-pro-preview-05-06.md)
|---------------------------|----------------|------------|----------------------------|
| DeepSeek 34B-instruct | 34 | yes | 40.7 |
| DeepSeek 34B-instruct | 34 | no | 3.6 |
## Results analysis
- **Fine-tuning is a must** - without fine-tuning, open-source models provide poor results on most real-world code tasks, which include complicated prompt and lengthy context. We clearly see that without fine-tuning, deepseek model was 96.4% of the time inferior to GPT-4, while after fine-tuning, it is better 40.7% of the time. #### Model 'Gemini-2.5-pro-preview-05-06' vs 'Sonnet 3.7' - Detailed Analysis
- **Always start from a code-dedicated model** — When fine-tuning, always start from a code-dedicated model, and not from a general-usage model. The gaps in downstream results are very big.
- **Don't believe the hype** —newer models, or models from big-tech companies (Llama3, Gemma, Mistral), are not always better for fine-tuning.
- **The best large model** - For large 34B code-dedicated models, the gaps when doing proper fine-tuning are small. The current top model is **DeepSeek 34B-instruct**
- **The best small model** - For small 7B code-dedicated models, the gaps when fine-tuning are much larger. **CodeQWEN 1.5-7B** is by far the best model for fine-tuning.
- **Base vs. instruct** - For the top model (deepseek), we saw small advantage when starting from the instruct version. However, we recommend testing both versions on each specific task, as the base model is generally considered more suitable for fine-tuning.
## Dataset strengths:
### Training dataset - higher_accuracy_and_coverage: finds real critical bugs and supplies actionable patches in most examples (better in 78 % of cases).
- guideline_awareness: usually respects new-lines-only scope, ≤3 suggestions, proper YAML, and stays silent when no issues exist.
- detailed_reasoning_and_patches: explanations tie directly to the diff and fixes are concrete, often catching multiple related defects that 'Sonnet 3.7' overlooks.
Our training dataset comprises 25,000 pull requests, aggregated from permissive license repos. For each pull request, we generated responses for the three main tools of Qodo Merge: weaknesses:
[Describe](https://qodo-merge-docs.qodo.ai/tools/describe/), [Review](https://qodo-merge-docs.qodo.ai/tools/improve/) and [Improve](https://qodo-merge-docs.qodo.ai/tools/improve/).
On the raw data collected, we employed various automatic and manual cleaning techniques to ensure the outputs were of the highest quality, and suitable for instruct-tuning. - occasional_rule_violations: sometimes proposes new imports, package-version changes, or edits outside the added lines.
- overzealous_suggestions: may add speculative or stylistic fixes that exceed the “critical” scope, or mis-label severity.
- sporadic_technical_slips: a few patches contain minor coding errors, oversized snippets, or duplicate/contradicting advice.
Here are the prompts, and example outputs, used as input-output pairs to fine-tune the models:
| Tool | Prompt | Example output | ### Model 'Gemini-2.5-pro-preview-05-06' vs 'GPT-4.1'
|----------|------------------------------------------------------------------------------------------------------------|----------------|
| Describe | [link](https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_description_prompts.toml) | [link](https://github.com/Codium-ai/pr-agent/pull/910#issue-2303989601) |
| Review | [link](https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_reviewer_prompts.toml) | [link](https://github.com/Codium-ai/pr-agent/pull/910#issuecomment-2118761219) |
| Improve | [link](https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_code_suggestions_prompts.toml) | [link](https://github.com/Codium-ai/pr-agent/pull/910#issuecomment-2118761309) |
### Evaluation dataset
- For each tool, we aggregated 200 additional examples to be used for evaluation. These examples were not used in the training dataset, and were manually selected to represent diverse real-world use-cases. [//]: # (On coding tasks, the gap between open-source models and top closed-source models such as Claude and GPT is significant.)
- For each test example, we generated two responses: one from the fine-tuned model, and one from the best code model in the world, `gpt-4-turbo-2024-04-09`.
- We used a third LLM to judge which response better answers the prompt, and will likely be perceived by a human as better response. [//]: # (<br>)
<br>
We experimented with three model as judges: `gpt-4-turbo-2024-04-09`, `gpt-4o`, and `claude-3-opus-20240229`. All three produced similar results, with the same ranking order. This strengthens the validity of our testing protocol. [//]: # (In practice, open-source models are unsuitable for most real-world code tasks, and require further fine-tuning to produce acceptable results.)
The evaluation prompt can be found [here](https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_evaluate_prompt_response.toml)
Here is an example of a judge model feedback: [//]: # ()
[//]: # (_Qodo Merge pull request benchmark_ aims to benchmark models on their ability to be fine-tuned for a coding task.)
``` [//]: # (Specifically, we chose to fine-tune open-source models on the task of analyzing a pull request, and providing useful feedback and code suggestions.)
command: improve
model1_score: 9,
model2_score: 6,
why: |
Response 1 is better because it provides more actionable and specific suggestions that directly
enhance the code's maintainability, performance, and best practices. For example, it suggests
using a variable for reusable widget instances and using named routes for navigation, which
are practical improvements. In contrast, Response 2 focuses more on general advice and less
actionable suggestions, such as changing variable names and adding comments, which are less
critical for immediate code improvement."
```
## Comparing Top Closed-Source Models [//]: # ()
[//]: # (Here are the results:)
Another application of the Pull Request Benchmark is comparing leading closed-source models to determine which performs better at analyzing pull request code. [//]: # (<br>)
The evaluation methodology resembles the approach used for evaluating fine-tuned models: [//]: # (<br>)
- We ran each model across 200 diverse pull requests, asking them to generate code suggestions using Qodo Merge's `improve` tool [//]: # ()
- A third top model served as judge to determine which response better fulfilled the prompt and would likely be perceived as superior by human users [//]: # (**Model performance:**)
[//]: # ()
[//]: # (| Model name | Model size [B] | Better than gpt-4 rate, after fine-tuning [%] |)
[//]: # (|-----------------------------|----------------|----------------------------------------------|)
[//]: # (| **DeepSeek 34B-instruct** | **34** | **40.7** |)
[//]: # (| DeepSeek 34B-base | 34 | 38.2 |)
[//]: # (| Phind-34b | 34 | 38 |)
[//]: # (| Granite-34B | 34 | 37.6 |)
[//]: # (| Codestral-22B-v0.1 | 22 | 32.7 |)
[//]: # (| QWEN-1.5-32B | 32 | 29 |)
[//]: # (| | | |)
[//]: # (| **CodeQwen1.5-7B** | **7** | **35.4** |)
[//]: # (| Llama-3.1-8B-Instruct | 8 | 35.2 |)
[//]: # (| Granite-8b-code-instruct | 8 | 34.2 |)
[//]: # (| CodeLlama-7b-hf | 7 | 31.8 |)
[//]: # (| Gemma-7B | 7 | 27.2 |)
[//]: # (| DeepSeek coder-7b-instruct | 7 | 26.8 |)
[//]: # (| Llama-3-8B-Instruct | 8 | 26.8 |)
[//]: # (| Mistral-7B-v0.1 | 7 | 16.1 |)
[//]: # ()
[//]: # (<br>)
[//]: # ()
[//]: # (**Fine-tuning impact:**)
[//]: # ()
[//]: # (| Model name | Model size [B] | Fine-tuned | Better than gpt-4 rate [%] |)
[//]: # (|---------------------------|----------------|------------|----------------------------|)
[//]: # (| DeepSeek 34B-instruct | 34 | yes | 40.7 |)
[//]: # (| DeepSeek 34B-instruct | 34 | no | 3.6 |)
[//]: # ()
[//]: # (## Results analysis)
[//]: # ()
[//]: # (- **Fine-tuning is a must** - without fine-tuning, open-source models provide poor results on most real-world code tasks, which include complicated prompt and lengthy context. We clearly see that without fine-tuning, deepseek model was 96.4% of the time inferior to GPT-4, while after fine-tuning, it is better 40.7% of the time.)
[//]: # (- **Always start from a code-dedicated model** — When fine-tuning, always start from a code-dedicated model, and not from a general-usage model. The gaps in downstream results are very big.)
[//]: # (- **Don't believe the hype** —newer models, or models from big-tech companies &#40;Llama3, Gemma, Mistral&#41;, are not always better for fine-tuning.)
[//]: # (- **The best large model** - For large 34B code-dedicated models, the gaps when doing proper fine-tuning are small. The current top model is **DeepSeek 34B-instruct**)
[//]: # (- **The best small model** - For small 7B code-dedicated models, the gaps when fine-tuning are much larger. **CodeQWEN 1.5-7B** is by far the best model for fine-tuning.)
[//]: # (- **Base vs. instruct** - For the top model &#40;deepseek&#41;, we saw small advantage when starting from the instruct version. However, we recommend testing both versions on each specific task, as the base model is generally considered more suitable for fine-tuning.)
[//]: # ()
[//]: # (## Dataset)
[//]: # ()
[//]: # (### Training dataset)
[//]: # ()
[//]: # (Our training dataset comprises 25,000 pull requests, aggregated from permissive license repos. For each pull request, we generated responses for the three main tools of Qodo Merge:)
[//]: # ([Describe]&#40;https://qodo-merge-docs.qodo.ai/tools/describe/&#41;, [Review]&#40;https://qodo-merge-docs.qodo.ai/tools/improve/&#41; and [Improve]&#40;https://qodo-merge-docs.qodo.ai/tools/improve/&#41;.)
[//]: # ()
[//]: # (On the raw data collected, we employed various automatic and manual cleaning techniques to ensure the outputs were of the highest quality, and suitable for instruct-tuning.)
[//]: # ()
[//]: # (Here are the prompts, and example outputs, used as input-output pairs to fine-tune the models:)
[//]: # ()
[//]: # (| Tool | Prompt | Example output |)
[//]: # (|----------|------------------------------------------------------------------------------------------------------------|----------------|)
[//]: # (| Describe | [link]&#40;https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_description_prompts.toml&#41; | [link]&#40;https://github.com/Codium-ai/pr-agent/pull/910#issue-2303989601&#41; |)
[//]: # (| Review | [link]&#40;https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_reviewer_prompts.toml&#41; | [link]&#40;https://github.com/Codium-ai/pr-agent/pull/910#issuecomment-2118761219&#41; |)
[//]: # (| Improve | [link]&#40;https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_code_suggestions_prompts.toml&#41; | [link]&#40;https://github.com/Codium-ai/pr-agent/pull/910#issuecomment-2118761309&#41; |)
[//]: # ()
[//]: # (### Evaluation dataset)
[//]: # ()
[//]: # (- For each tool, we aggregated 200 additional examples to be used for evaluation. These examples were not used in the training dataset, and were manually selected to represent diverse real-world use-cases.)
[//]: # (- For each test example, we generated two responses: one from the fine-tuned model, and one from the best code model in the world, `gpt-4-turbo-2024-04-09`.)
[//]: # ()
[//]: # (- We used a third LLM to judge which response better answers the prompt, and will likely be perceived by a human as better response.)
[//]: # (<br>)
[//]: # ()
[//]: # (We experimented with three model as judges: `gpt-4-turbo-2024-04-09`, `gpt-4o`, and `claude-3-opus-20240229`. All three produced similar results, with the same ranking order. This strengthens the validity of our testing protocol.)
[//]: # (The evaluation prompt can be found [here]&#40;https://github.com/Codium-ai/pr-agent/blob/main/pr_agent/settings/pr_evaluate_prompt_response.toml&#41;)
[//]: # ()
[//]: # (Here is an example of a judge model feedback:)
[//]: # ()
[//]: # (```)
[//]: # (command: improve)
[//]: # (model1_score: 9,)
[//]: # (model2_score: 6,)
[//]: # (why: |)
[//]: # ( Response 1 is better because it provides more actionable and specific suggestions that directly)
[//]: # ( enhance the code's maintainability, performance, and best practices. For example, it suggests)
[//]: # ( using a variable for reusable widget instances and using named routes for navigation, which)
[//]: # ( are practical improvements. In contrast, Response 2 focuses more on general advice and less)
[//]: # ( actionable suggestions, such as changing variable names and adding comments, which are less)
[//]: # ( critical for immediate code improvement.")
[//]: # (```)
[//]: # ()
[//]: # (## Comparing Top Closed-Source Models)
[//]: # ()
[//]: # (Another application of the Pull Request Benchmark is comparing leading closed-source models to determine which performs better at analyzing pull request code.)
[//]: # ()
[//]: # (The evaluation methodology resembles the approach used for evaluating fine-tuned models:)
[//]: # ()
[//]: # (- We ran each model across 200 diverse pull requests, asking them to generate code suggestions using Qodo Merge's `improve` tool)
[//]: # (- A third top model served as judge to determine which response better fulfilled the prompt and would likely be perceived as superior by human users)