Add codestral prompt for code completions
What does this MR do and why?
This MR adds support for using a codestral
model for Code Completions via Code Suggestions.
It does so by:
- Create a new abstract class
AiGatewayCodeCompletionMessage
requiring only theprompt
method to be implemented by subclasses. - Create the respective
CodestralMessage
subclass that implements the FIM prompt required by Codestral. - Refactor
CodeGemmaMessage
to also subclassAiGatewayCodeCompletionMessage
, removing some duplicated code.
Part of #467433 (closed)
Evaluation results
I ran the code-suggestions eval
pipeline against a local gdk instance that uses codestral for code completions.
config
{
"beam_config": {
"pipeline_options": {
"runner": "DirectRunner",
"project": "dev-ai-research-0e2f8974",
"region": "us-central1",
"temp_location": "gs://prompt-library/tmp/",
"save_main_session": false
}
},
"input_bq_table": "dev-ai-research-0e2f8974.code_suggestion.input_testcases_v1",
"output_bq_table": "dev-ai-research-0e2f8974.code_suggestion_experiments.bc_codestral_completions_py_only_temp0",
"throttle_sec": 1,
"batch_size": 10,
"min_length": 0,
"eval_setups": [
{
"llm_config": {
"name": "gitlab-code-completion",
"parameters": {
"base_url": "http://gdk.test:3000",
"intent": "completion"
}
},
"languages": [
"python"
],
"post_transformations": [],
"include_suffix": true
}
]
}
model | avg_similarity_score | sample_size | table |
---|---|---|---|
codestral:latest | 0.77 | 510 | dev-ai-research-0e2f8974.code_suggestion_experiments.bc_codestral_completions_py_only_temp0 |
prod (code-gecko) | 0.79 | 510 | dev-ai-research-0e2f8974.code_suggestion_experiments.bc_becko_completions_py_only |
Example query:
SELECT model, round(avg(similarity_score), 2) as avg_score, count(*) FROM `dev-ai-research-0e2f8974.code_suggestion_experiments.bc_codestral_completions_py_only_temp0` WHERE NOT STARTS_WITH(model_completion, '{') GROUP BY model
MR acceptance checklist
Please evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.
Screenshots or screen recordings
Screenshots are required for UI changes, and strongly recommended for all other merge requests.
Before | After |
---|---|
How to set up and validate locally
Numbered steps to set up and validate the change are strongly suggested.
Edited by Bruno Cardoso