[Ellipsis] Add MD_YAML Mode to Patch Function

[ellipsis-dev]

This change addresses review comments left by @jxnl on PR #674: Add markdown yaml input mode to optimize input tokens

⚠️ We couldn't build/test your project to verify our changes. Add a Dockerfile to significantly improve code quality.

Summary:

This PR adds a new MD_YAML mode to the patch function to support parsing YAML inputs within markdown and converting them to JSON.

Key points:

• Introduces new MD_YAML mode in patch function
• Allows parsing of YAML within markdown to JSON
• Example usage provided in the documentation

---

You can configure Ellipsis to address comments with a direct commit or a side PR, see docs.

---

Something look wrong?: If this Pull Request doesn't address the comments left on the above pull request, create a new PR review with more details. For more information, check the documentation.

Generated with ❤️ by ellipsis.dev

[cloudflare-pages]

Deploying instructor with    Cloudflare Pages

Latest commit:	f61fd78
Status:	✅  Deploy successful!
Preview URL:	https://bdb1aad4.instructor.pages.dev
Branch Preview URL:	https://ellipsis-wflow-akw3ynscskj2i.instructor.pages.dev

View logs

[sweep-ai]

Sweep: PR Review

instructor/client.py

The changes add support for the MD_YAML mode for the ANYSCALE, TOGETHER, and OPENAI providers.

---

instructor/function_calls.py

The changes introduce support for parsing and validating YAML formatted responses by adding a new import, modifying the from_response method to handle a new mode, and adding a new parse_yaml method to the OpenAISchema class.

---

instructor/mode.py

A new mode MD_YAML was added to the Mode enum class.

---

instructor/process_response.py

The changes introduce support for handling YAML formatted responses by adding the yaml module import and modifying the handle_response_model function to handle a new mode, Mode.MD_YAML.

Sweep Found These Issues

• The new conditional block for Mode.MD_YAML does not handle the case where new_kwargs["messages"] is empty, which could lead to an IndexError.

instructor/instructor/process_response.py

Lines 239 to 269 in f61fd78

	message = dedent(
	f"""
	As a genius expert, your task is to understand the content and provide
	the parsed objects in yaml that match the following yaml_schema:\n
	{yaml.dump(response_model.model_json_schema(), indent=2)}
	Make sure to return an instance of the YAML, not the schema itself
	"""
	)
	new_kwargs["messages"].append(
	{
	"role": "user",
	"content": "Return the correct YAML response within a ```yaml codeblock. not the YAML_SCHEMA",
	},
	)
	# check that the first message is a system message
	# if it is not, add a system message to the beginning
	if new_kwargs["messages"][0]["role"] != "system":
	new_kwargs["messages"].insert(
	0,
	{
	"role": "system",
	"content": message,
	},
	)
	# if it is, system append the schema to the end
	else:
	new_kwargs["messages"][0]["content"] += f"\n\n{message}"


View Diff

---

instructor/retry.py

A new elif condition was added to handle the Mode.MD_YAML mode in the reask_messages function, providing specific instructions for correcting YAML responses.

Sweep Found These Issues

• The new elif condition for Mode.MD_YAML does not include a return statement, which could lead to unintended fall-through behavior.

instructor/instructor/retry.py

Lines 118 to 122 in f61fd78

	elif mode == Mode.MD_YAML:
	yield {
	"role": "user",
	"content": f"Correct your YAML ONLY RESPONSE, based on the following errors:\n{exception}",
	}


View Diff

---

instructor/utils.py

The changes introduce the yaml module import and add a new function extract_json_from_yaml_codeblock to handle and convert YAML code blocks into JSON format.

Sweep Found These Issues

• The function extract_json_from_yaml_codeblock does not handle potential exceptions that could be raised by yaml.safe_load, such as yaml.YAMLError, which could lead to unhandled exceptions if the YAML content is malformed.

instructor/instructor/utils.py

Lines 74 to 81 in f61fd78

	def extract_json_from_yaml_codeblock(content: str) -> str:
	yaml_start = content.find("```yaml")
	if yaml_start != -1:
	yaml_end = content.find("```", yaml_start + 7)
	if yaml_end != -1:
	yaml_string = yaml.safe_load(content[yaml_start + 7 : yaml_end].strip())
	return json.dumps(yaml_string)
	return ""


View Diff

---

patch.md

The changes add documentation for a new mode MD_YAML that allows the system to handle YAML inputs within markdown and convert them into JSON for further processing.

Sweep Found These Issues

• The documentation does not specify how the system handles errors or invalid YAML inputs, which could lead to confusion or misuse.

instructor/patch.md

Lines 1 to 24 in f61fd78

	## YAML in Patching
	The system now supports a new mode, `MD_YAML`, to handle YAML inputs within markdown. This mode allows the system to parse YAML inputs and convert them into JSON for further processing.
	To use this mode, you need to specify it when calling the `patch` function, like so:
	```python
	client = instructor.patch(client, mode=instructor.Mode.MD_YAML)
	```
	You can then provide YAML inputs within markdown. Here's an example:
	```yaml
	Order Details:
	Customer: Jason
	Items:
	Name: Apple, Price: 0.50
	Name: Bread, Price: 2.00
	Name: Milk, Price: 1.50
	```
	The system will parse this input and convert it into a JSON object for further processing.


View Diff

---

tests/llm/test_openai/test_modes.py

A new test function test_yaml was added to validate the Order model using a YAML-formatted string as input.

---

[jxnl]

@ellipsis-dev delete everything below this

[ellipsis-dev]

@jxnl, I have addressed your comments in commit fb51ac6

---

You can configure Ellipsis to address comments with a direct commit or a side PR, see docs.