https://github.com/cedarscript/cedarscript-integration-aider
Allows Aider to use CEDARScript as an edit format
https://github.com/cedarscript/cedarscript-integration-aider
Last synced: 8 months ago
JSON representation
Allows Aider to use CEDARScript as an edit format
- Host: GitHub
- URL: https://github.com/cedarscript/cedarscript-integration-aider
- Owner: CEDARScript
- License: agpl-3.0
- Created: 2024-10-20T23:42:19.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-11-18T03:50:16.000Z (over 1 year ago)
- Last Synced: 2024-11-18T04:38:15.151Z (over 1 year ago)
- Language: Python
- Size: 267 KB
- Stars: 7
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# CEDARScript Integration: Aider
[](https://pypi.org/project/cedarscript-integration-aider/)
[](https://pypi.org/project/cedarscript-integration-aider/)
[](https://github.com/psf/black)
[](https://www.gnu.org/licenses/agpl-3.0)
`CEDARScript Integration: Aider` enables [`Aider`](https://aider.chat/) to use
[**CEDARScript**](https://github.com/CEDARScript/cedarscript-grammar#readme)
as an [_edit format_](https://aider.chat/docs/benchmarks.html#edit-formats).
## Table of Contents
- [What is CEDARScript?](#what-is-cedarscript)
- [Installation](#installation)
- [Running a Benchmark](#running-a-benchmark)
- [Why Use CEDARScript?](#why-use-cedarscript)
- [Performance Comparison](#performance-comparison)
- [Notable Achievements](#notable-achievements)
- [The Gemini 1.5 Flash benchmark highlights](#the-gemini-15-flash-benchmark-highlights)
- [Individual Test Analysis](#individual-test-analysis)
- [Detailed Analysis](#detailed-analysis)
- [Contributing](#contributing)
- [License](#license)
## What is CEDARScript?
[CEDARScript](https://bit.ly/cedarscript) (_Concise Examination, Development, And Refactoring Script_)
is a domain-specific language designed to improve how AI coding assistants interact with codebases and communicate their code modification intentions.
It provides a standardized way to express complex code modification and analysis operations, making it easier for
AI-assisted development tools to understand and execute these tasks.
## Installation
1. Install **Aider with _CEDARScript and CedarTL support_** via this command below:
```shell
python -m ensurepip --upgrade
pip install --upgrade --force-reinstall \
git+https://github.com/elifarley/aider@cedarscript \
aider-chat
```
2. Now, simply use the [`--edit-format` switch](https://aider.chat/docs/more/edit-formats.html) and select `cedarscript`:
```shell
aider --edit-format cedarscript
```
## Running a Benchmark
### One-Time Actions
First, [install Aider with CEDARScript](#installation);
Then, follow the [**benchmark setup instructions**](https://github.com/Aider-AI/aider/blob/main/benchmark/README.md#setup-for-benchmarking) once.
Then, install the [`refactor-benchmark`](https://github.com/Aider-AI/refactor-benchmark/tree/main#benchmark-details),
which will perform refactorings on a _non-trivial_ amount of code found in fairly **large** files:
```shell
( cd tmp.benchmarks && git clone https://github.com/Aider-AI/refactor-benchmark.git )
```
## For Every Benchmark Run
Finally, for every new benchmark you want to run:
```shell
# Launch the docker container
benchmark/docker.sh
# Inside the container, install aider as a development build.
# This way you're running the code that you cloned above, including any local changes.
pip install -e .
###
benchmark/benchmark.py gemini-flash-cedarscript-version-refactor \
--model gemini/gemini-1.5-flash-latest \
--edit-format cedarscript \
--exercises-dir refactor-benchmark \
--threads 1 #### Must be only 1 ####
```
## Why use CEDARScript?
`TL;DR`: You can get higher success rates when refactoring large files, comparing to other edit formats.
1. **Higher Success Rates**: Significantly improves the performance of AI models in code refactoring tasks.
2. **Cost-Effective Performance**: Enables more affordable models to compete with top-tier options.
3. **Standardized Communication**: Provides a consistent format for AI-code interaction in coding tasks.
4. **Enhanced Accuracy**: Reduces errors and improves the quality of AI-generated code modifications.
## Performance Comparison
CEDARScript has shown remarkable improvements in AI model performance for code refactoring:
| Model | Format | Pass Rate | Well-Formed Cases | Syntax Errors | Indentation Errors | Cost | Avg. Time per case |
|-------------------|-------------|-----------|-------------------|---------------|--------------------|------|--------------------|
| Gemini 1.5 PRO | CEDARScript | 77.5% | 86.5% | 4 | 3 | 26.2 | 29 |
| Gemini 1.5 Flash | CEDARScript | 76.4% | 94.4% | 3 | 5 | 0.68 | 14.7 |
| Claude 3.5 Sonnet | diff | 64.0% | 76.4% | n/a | n/a | n/a | n/a |
| Gemini 1.5 PRO | diff-fenced | 49.4% | 7.9% | 21 | 93 | 28.3 | 110.1 |
### Notable Achievements:
- **Gemini 1.5 _PRO_** with **CEDARScript** outperformed both its diff-fenced format and **Claude 3.5 Sonnet**.
- Most remarkably, the more cost-effective **Gemini 1.5 _Flash_** model, using **CEDARScript**, outperformed **Claude 3.5 Sonnet**.
- It goes to show that even a more affordable model can surpass top-tier competitors when equipped with the _right_ tools.
This suggests that **CEDARScript** can level the playing field, enabling more accessible AI models
to compete with and even _exceed_ the capabilities of more expensive options in certain coding tasks.
### The Gemini 1.5 Flash benchmark highlights
- 48% of tests (43 total) showed improvements
- 103% increase in Pass 1 success rate (75 tests)
- Test duration reduced by 93% (from 5:17:26 to 0:25:17)
- Token efficiency greatly improved:
- Sent tokens: -37% (7.59M)
- Received tokens: -96% (180K)
- Error reduction:
- Error outputs: -94% (35 total)
- Malformed outputs: -94% (6 cases)
- Syntax errors: -85% (3 cases)
- Indent errors eliminated (100% reduction)
Delta...
### Individual Test Analysis
Individual Test Diff
```diff
--- 2024-10-22-05-21-41--gemini-1.5-flash-refactoring-whole
+++ 2024-10-22-05-13-37--gemini-1.5-flash-refactoring-cedarscript-i0.0.9-e0.3.3
# ============= Failed Attempts per Test =============
# N >= 0: it eventually passed after N failed attempts
# N < 0 : All attempts failed and limit was reached
@@ Improved, now PASSED (36) @@
++analyzer_cli_DebugAnalyzer__make_source_table: -4 -> 0
++autodetector_MigrationAutodetector__trim_to_apps: -4 -> 0
++backends_ModelBackend_with_perm: -4 -> 0
++builtin_BuiltinVariable_call_setattr: -4 -> 0
++checks_BaseModelAdminChecks__check_ordering_item: -4 -> 0
++checks_BaseModelAdminChecks__check_raw_id_fields_item: -4 -> 0
++checks_ModelAdminChecks__check_action_permission_methods: -4 -> 0
++checks_ModelAdminChecks__check_inlines_item: -4 -> 0
++checks_ModelAdminChecks__check_list_display_item: -4 -> 0
++clustering_ops_KMeans__mini_batch_training_op: -4 -> 0
++codeeditor_CodeEditor___get_brackets: -4 -> 0
++config_AppConfig__path_from_module: -4 -> 0
++config_ConfigCLI__get_settings_vars: -4 -> 0
++coordinator_HERETransitDataUpdateCoordinator__parse_transit_response: -4 -> 0
++cuda_cpp_scheduling_CUDACPPScheduling__can_fuse_epilogue_impl: -4 -> 0
++dataframeeditor_DataFrameView_next_index_name: -4 -> 0
++diffsettings_Command_output_hash: -4 -> 0
++dim2_Dim2CompatTests_test_reductions_2d_axis0: -4 -> 3
++distribution_DistributionFiles_parse_distribution_file_SUSE: -4 -> 0
++doc_DocCLI_get_role_man_text: -4 -> 0
++figure_FigureBase_colorbar: -4 -> 0
++functional_Functional__conform_to_reference_input: -4 -> 0
++galaxy_GalaxyCLI_execute_list_collection: -4 -> 0
++kernel_SpyderKernel_get_fault_text: -4 -> 1
++main_widget_PylintWidget_parse_output: -4 -> 0
++methods_BaseMethodsTests_test_where_series: -4 -> 0
++ogrinspect_Command_add_arguments: -4 -> 0
++onnxfunction_dispatcher_OnnxFunctionDispatcher__get_aten_name: -4 -> 1
++operations_DatabaseOperations_last_executed_query: -4 -> 0
++polar_RadialTick__determine_anchor: -4 -> 0
++profile_analyzer_cli_ProfileAnalyzer__get_list_profile_lines: -4 -> 0
++split_cat_SplitCatSimplifier_replace_cat: -4 -> 0
++split_cat_SplitCatSimplifier_replace_split: -4 -> 0
++text_CountVectorizer__limit_features: -4 -> 0
++triton_TritonScheduling_define_kernel: -4 -> 0
++triton_TritonScheduling_generate_node_schedule: -4 -> 0
@@ Improved, minor (3) @@
+ checks_BaseModelAdminChecks__check_autocomplete_fields_item: 3 -> 0
+ dataloader_DataLoader__is_role: 2 -> 0
+ operations_OracleOperations_convert_extent: 1 -> 0
@@ Worsened, now FAILED (7) @@
--base_BaseHandler_adapt_method_mode: 0 -> -4
--feedgenerator_Atom1Feed_add_item_elements: 0 -> -4
--generic_bsd_GenericBsdIfconfigNetwork_parse_inet_line: 0 -> -4
--graph_drawer_FxGraphDrawer__stringify_tensor_meta: 0 -> -4
--group_batch_fusion_GroupLinearFusion_fuse: 0 -> -4
--inspectdb_Command_normalize_col_name: 0 -> -4
--introspection_DatabaseIntrospection__parse_column_or_constraint_definition: 0 -> -4
@@ Stable: PASSED (30) @@
=+autosave_AutosaveForPlugin_get_files_to_recover: 0
=+base_BaseHandler_check_response: 0
=+baseconv_BaseConverter_convert: 0
=+compile_utils_MetricsContainer__get_metric_object: 0
=+concat__Concatenator__clean_keys_and_objs: 0
=+config_NetworkConfig_parse: 0
=+csrf_CsrfViewMiddleware__set_csrf_cookie: 0
=+dumpdata_Command_add_arguments: 0
=+finders_FileSystemFinder_check: 0
=+gateway_Gateway_get_and_delete_all_sms: 0
=+getitem_BaseGetitemTests_test_get: 0
=+grad_scaler_GradScaler__unscale_grads_: 0
=+gradient_checker_GradientChecker__assertInferTensorChecks: 0
=+graph_MigrationGraph_iterative_dfs: 0
=+grpc_debug_server_EventListenerBaseServicer__process_tensor_event_in_chunks: 0
=+i18n_JavaScriptCatalog_get_paths: 0
=+inspectdb_Command_get_field_type: 0
=+inspectdb_Command_get_meta: 0
=+introspection_DatabaseIntrospection__get_column_collations: 0
=+load_v1_in_v2__EagerSavedModelLoader__extract_signatures: 0
=+makemessages_Command_add_arguments: 0
=+makemigrations_Command_add_arguments: 0
=+migrate_Command_add_arguments: 0
=+operations_DatabaseOperations_bulk_insert_sql: 0
=+operations_DatabaseOperations_check_expression_support: 0
=+reshaping_BaseReshapingTests_test_concat_mixed_dtypes: 0
=+schema_DatabaseSchemaEditor_quote_value: 0
=+shell_Command_python: 0
=+special_RunSQL__run_sql: 0
=+weather_NWSWeather__forecast: 0
@@ Stable: FAILED (13) @@
=-autodetector_MigrationAutodetector_check_dependency: -4 -> -4
=-checks_ModelAdminChecks__check_list_editable_item: -4 -> -4
=-common_methods_invocations_foreach_inputs_sample_func__sample_rightmost_arg: -4 -> -4
=-common_utils_TestCase_genSparseTensor: -4 -> -4
=-doc_DocCLI_display_plugin_list: -4 -> -4
=-generator_GenOpTestCase_out_variant_op_test_case_generator: -4 -> -4
=-options_ModelAdmin_message_user: -4 -> -4
=-patches__Curve__get_arrow_wedge: -4 -> -4
=-quiver_Barbs__make_barbs: -4 -> -4
=-reshaping_BaseReshapingTests_test_unstack: -4 -> -4
=-sharding_policies_MaxShardSizePolicy__add_partition: -4 -> -4
=-split_cat_SplitCatSimplifier_get_transform_params: -4 -> -4
=-symbolic_shapes_ShapeEnv_bind_symbols: -4 -> -4
--- 2024-10-22-05-21-41--gemini-1.5-flash-refactoring-whole
+++ 2024-10-26-23-32-08--gemini-1.5-pro-refactoring-cedarscript-i0.0.18-e0.3.13
@@ ============= TEST STATUS CHANGES ============ @@
< REMOVED : 71 ( 80% of total)
<+ PASSED : 35 ( 39% of total)
<- FAILED : 36 ( 40% of total)
+ IMPROVED : 11 ( 12% of total)
++ Now PASSES : 10 ( 11% of total)
+ Minor : 1 ( 1% of total)
- WORSENED : 3 ( 3% of total)
-- Now FAILED : 3 ( 3% of total)
- Minor : 0 ( 0% of total)
= STABLE : 4 ( 4% of total)
=+ PASSED : 1 ( 1% of total)
#- FAILED : 3 ( 3% of total)
@@ ============= PERFORMANCE METRICS ============ @@
# TOTAL TEST COUNT : 18 (-71)
# DURATION hh:mm:ss: 0:10:29 (- 5:32:14, -97%) ---------
# COST ($) : 9.40 ( +7.02, +294%) ++++++++++++++++++++
# TOKENS SENT : 2,683,000 (-9,312,700, -78%) -------
# TOKENS RECEIVED : 12,382 (-4,970,531, -100%) ---------
# TIMEOUTS : 0 N/A
# ERROR OUTPUTS : 62 N/A
# USER ASKS : 37 ( -521, -93%) ---------
# CONTEXT EXHAUSTS : 0 N/A
# MALFORMED : 4 N/A
# SYNTAX ERRORS : 1 ( -105, -99%) ---------
# INDENT ERRORS : 9 ( -11, -55%) -----
# LAZY COMMENTS : 0 ( -28, -100%) ----------
```
#### Detailed Analysis
This overview suggests that the CEDARScript edit format has had a significant positive impact on the task of method extraction,
with improvements in nearly half of the tests and only a small percentage of tests worsening.
**Improvements:**
36 tests that previously failed now pass. This is a substantial improvement, indicating that CEDARScript is more
effective in correctly extracting methods from classes across a wide range of codebases.
**Notable improvements include:**
- `analyzer_cli_DebugAnalyzer__make_source_table`: Suggests better handling of debug-related code refactoring.
- `autodetector_MigrationAutodetector__trim_to_apps`: Indicates improved capability in refactoring Django migration-related code.
- `cuda_cpp_scheduling_CUDACPPScheduling__can_fuse_epilogue_impl`: Shows better performance in handling complex CUDA-related code.
- `triton_TritonScheduling_define_kernel` and triton_TritonScheduling_generate_node_schedule: Demonstrates improved capability in refactoring GPU computing-related code.
**Minor Improvements:**
3 tests showed minor improvements, such as `checks_BaseModelAdminChecks__check_autocomplete_fields_item`, suggesting
slight enhancements in handling Django admin-related code.
**Regressions:**
7 tests that previously passed now fail. While concerning, it's a relatively small number compared to the improvements.
**Notable regressions include:**
- `feedgenerator_Atom1Feed_add_item_elements`: Suggests potential issues with refactoring feed generation code.
- `generic_bsd_GenericBsdIfconfigNetwork_parse_inet_line`: Indicates challenges in refactoring network-related parsing code.
- `introspection_DatabaseIntrospection__parse_column_or_constraint_definition`: Shows difficulties in handling database schema introspection code.
**Stability:**
30 tests remained stable and passing, indicating that CEDARScript maintained performance in many areas, including various
Django commands, database operations, and utility functions.
13 tests remained stable but failing, suggesting that some challenging areas were not addressed by either format.
These include complex operations like `symbolic_shapes_ShapeEnv_bind_symbols` and `reshaping_BaseReshapingTests_test_unstack`.
**Analysis by Domain:**
- **Web Frameworks (e.g., Django)**: Generally improved, with better handling of model admin checks, configuration, and
database operations.
- **Data Science and ML**: Mixed results. Improvements in areas like clustering operations (`clustering_ops_KMeans__mini_batch_training_op`)
and data frame handling, but persistent issues in some reshaping operations.
- **System-level Operations**: Some improvements (e.g., `distribution_DistributionFiles_parse_distribution_file_SUSE`)
but also regressions (e.g., `generic_bsd_GenericBsdIfconfigNetwork_parse_inet_line`).
- **GPU and High-Performance Computing**: Significant improvements, especially in `CUDA` and `Triton`-related code.
**Interpretation:**
CEDARScript appears more effective in handling complex code structures, especially in areas related to web frameworks,
data processing, and high-performance computing.
It shows improved capability in understanding class contexts and correctly extracting methods across various domains.
However, it may introduce new challenges in certain specific areas, possibly due to its different approach to code manipulation.
**Areas for Further Investigation:**
Understanding why certain tests regressed (e.g., `feedgenerator_Atom1Feed_add_item_elements`) could provide insights for improvement.
Analyzing the stable failing tests (e.g., `symbolic_shapes_ShapeEnv_bind_symbols`) to see if CEDARScript can be enhanced
to address these persistent issues.
**Conclusion:**
- The introduction of CEDARScript appears to be a significant improvement for the task of extracting methods from classes.
- It shows particular strength in handling complex codebases, especially those related to web frameworks, data processing, and high-performance computing.
- However, care should be taken to address the areas where regressions occurred, particularly in system-level operations and certain parsing tasks.
- The consistent performance across various domains suggests that CEDARScript offers a more robust and versatile approach to code refactoring.
This analysis indicates that CEDARScript is a promising enhancement to Aider, offering more accurate and comprehensive
method extraction capabilities across a wide range of codebases.
However, it also highlights the need for continued refinement, especially in areas where regressions were observed.
### Benchmark Metrics
Sonnet 3.5 + diff
```yaml
- dirname: refac-claude-3.5-sonnet-diff-not-lazy
model: claude-3.5-sonnet (diff)
edit_format: diff
pass_rate_1: 64.0
percent_cases_well_formed: 76.4
```
Gemini 1.5 PRO + diff-fenced (leaderboard site)
```yaml
- dirname: refac-gemini
model: gemini/gemini-1.5-pro-latest
edit_format: diff-fenced
pass_rate_1: 49.4
percent_cases_well_formed: 7.9
```
Gemini 1.5 PRO + diff-fenced (own tests)
```yaml
- dirname: 2024-10-05-00-43-21--diff-fenced-Gemini-Refactoring
test_cases: 89
model: gemini/gemini-1.5-pro-latest
edit_format: diff-fenced
commit_hash: 772710b-dirty
pass_rate_1: 18.0
pass_rate_2: 21.3
pass_rate_3: 24.7
percent_cases_well_formed: 34.8
error_outputs: 180
num_malformed_responses: 180
num_with_malformed_responses: 58
user_asks: 128
lazy_comments: 2
syntax_errors: 21
indentation_errors: 93
exhausted_context_windows: 0
test_timeouts: 0
command: aider --model gemini/gemini-1.5-pro-latest
date: 2024-10-05
versions: 0.57.2.dev
seconds_per_case: 110.1
total_cost: 28.2515
```
Gemini 1.5 PRO + CEDARScript
```yaml
- dirname: 2024-10-19-22-48-07--cedarscript-0.3.1-refactoring-gemini1.5pro
test_cases: 89
model: gemini/gemini-1.5-pro-latest
edit_format: cedarscript-g
commit_hash: 4da1e9b-dirty
pass_rate_1: 77.5
percent_cases_well_formed: 86.5
error_outputs: 337
num_malformed_responses: 19
num_with_malformed_responses: 12
user_asks: 12
lazy_comments: 0
syntax_errors: 4
indentation_errors: 3
exhausted_context_windows: 0
test_timeouts: 0
command: aider --model gemini/gemini-1.5-pro-latest
date: 2024-10-19
versions: 0.59.2.dev
seconds_per_case: 29.0
total_cost: 26.2374
```
Gemini 1.5 Flash + whole
```yaml
- dirname: 2024-10-22-05-21-41--gemini-1.5-flash-refactoring-whole
test_cases: 89
model: gemini/gemini-1.5-flash-002
edit_format: whole
commit_hash: feb1c38
pass_rate_1: 41.6
pass_rate_2: 42.7
pass_rate_3: 43.8
pass_rate_4: 44.9
percent_cases_well_formed: 100.0
error_outputs: 0
num_malformed_responses: 0
num_with_malformed_responses: 0
user_asks: 558
lazy_comments: 28
syntax_errors: 106
indentation_errors: 20
exhausted_context_windows: 0
test_timeouts: 0
command: aider --model gemini/gemini-1.5-flash-002
date: 2024-10-22
versions: 0.59.2.dev
seconds_per_case: 231.1
total_cost: 2.3894
```
Gemini 1.5 Flash + CEDARScript
```yaml
- dirname: 2024-10-20-00-33-27--cedarscript-0.3.1-refactoring-gemini-1.5-flash
test_cases: 89
model: gemini/gemini-1.5-flash-latest
edit_format: cedarscript-g
commit_hash: 4da1e9b-dirty
pass_rate_1: 76.4
percent_cases_well_formed: 94.4
error_outputs: 403
num_malformed_responses: 13
num_with_malformed_responses: 5
user_asks: 21
lazy_comments: 0
syntax_errors: 3
indentation_errors: 5
exhausted_context_windows: 0
test_timeouts: 0
command: aider --model gemini/gemini-1.5-flash-latest
date: 2024-10-20
versions: 0.59.2.dev
seconds_per_case: 14.7
total_cost: 0.6757
```
#### functional_Functional__conform_to_reference_input
diff-fenced
```yaml
"cost": 0.33188854999999995,
"duration": 27.793912172317505,
"test_timeouts": 0,
"commit_hash": "772710b-dirty",
"num_error_outputs": 2,
"num_user_asks": 3,
"num_exhausted_context_windows": 0,
"num_malformed_responses": 2,
"syntax_errors": 0,
"indentation_errors": 3,
"lazy_comments": 0,
```
cedarscript
```yaml
"cost": 0.18178265,
"duration": 11.176445960998535,
"test_timeouts": 0,
"commit_hash": "772710b-dirty",
"num_error_outputs": 0,
"num_user_asks": 1,
"num_exhausted_context_windows": 0,
"num_malformed_responses": 0,
"syntax_errors": 0,
"indentation_errors": 0,
"lazy_comments": 0,
```
## Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
## License
This project is licensed under the MIT License.