This PR introduces a new improved diff mode for the code modification prompt.
Instead of using the Standard Unified Diff Format, which works fine with Claude 3.5 Sonnet, this PR introduces a diff format more akin to Conflict Markers in Version Control which should be easier for LLMs to output.
This PR also:
Adds Deepseek API / deepseek-coder support
Configures the --diff, --no-diff mode automatically based on the selected model. This behaviour can be overriden using the --diff and --no-diff flags.
the --diff flag is no longer set to be true or false (see above)
adds the new 'gpt-4o-2024-08-06' version of gpt-4o which is cheaper and has a 4x longer output token limit
adds a more detailed model recommendation to the README
This PR introduces a new improved diff mode for the code modification prompt.
Instead of using the Standard Unified Diff Format, which works fine with Claude 3.5 Sonnet, this PR introduces a diff format more akin to Conflict Markers in Version Control which should be easier for LLMs to output.
This PR also:
Adds Deepseek API / deepseek-coder support
Configures the --diff, --no-diff mode automatically based on the selected model. This behaviour can be overriden using the --diff and --no-diff flags.
the --diff flag is no longer set to be true or false (see above)
adds the new 'gpt-4o-2024-08-06' version of gpt-4o which is cheaper and has a 4x longer output token limit
adds a more detailed model recommendation to the README
fixes #84 #85 #87