Implementation:Run llama Llama index CohereRerankerFinetuneEngine

Overview

The CohereRerankerFinetuneEngine provides a fine-tuning engine for Cohere's reranking models. It handles data upload, custom model creation via the Cohere API, and retrieval of the fine-tuned model as a LlamaIndex CohereRerank postprocessor. This module resides in the llama-index-finetuning package under the rerankers submodule.

Source file: llama-index-finetuning/llama_index/finetuning/rerankers/cohere_reranker.py (78 lines)

Dependencies

Dependency	Purpose
importlib.util	Checking whether the cohere package is installed
os	Reading the COHERE_API_KEY environment variable
llama_index.finetuning.types.BaseCohereRerankerFinetuningEngine	Abstract base class defining the fine-tuning engine interface
llama_index.postprocessor.cohere_rerank.CohereRerank	LlamaIndex Cohere reranker postprocessor returned by get_finetuned_model
cohere	Cohere SDK client (imported conditionally at runtime)
cohere.custom_model_dataset.JsonlDataset	Dataset wrapper for JSONL training/validation files (imported at runtime in finetune)

Class: CohereRerankerFinetuneEngine

Inherits from: BaseCohereRerankerFinetuningEngine

Constructor

def __init__(
    self,
    train_file_name: str = "train.jsonl",
    val_file_name: Optional[str] = None,
    model_name: str = "exp_finetune",
    model_type: str = "RERANK",
    base_model: str = "english",
    api_key: Optional[str] = None,
) -> None

Parameter	Type	Default	Description
train_file_name	str	"train.jsonl"	Path to the JSONL training data file
val_file_name	Optional[str]	None	Optional path to a JSONL validation data file
model_name	str	"exp_finetune"	Name for the custom model on Cohere's platform
model_type	str	"RERANK"	Type of custom model to create (always "RERANK" for rerankers)
base_model	str	"english"	Base reranking model to fine-tune on top of
api_key	Optional[str]	None	Cohere API key; falls back to the COHERE_API_KEY environment variable

Initialization behavior:

1. Checks if the cohere module is available using importlib.util.find_spec("cohere"). If not installed, raises ImportError with installation instructions.
2. Resolves the API key from the parameter or the COHERE_API_KEY environment variable. Raises ValueError if neither is available.
3. Creates a cohere.Client with the API key and sets client_name="llama_index".
4. Stores all configuration parameters as private instance attributes.
5. Initializes self._finetune_model to None.

Method: finetune

def finetune(self) -> None

Launches the fine-tuning job on Cohere's platform.

Workflow:

1. Imports JsonlDataset from cohere.custom_model_dataset.
2. Creates a JsonlDataset instance:
   • If a validation file is provided, passes both train_file and eval_file.
   • If no validation file is provided, passes only train_file.
3. Calls self._model.create_custom_model() with:
   • name -- the configured model name
   • dataset -- the JSONL dataset object
   • model_type -- the model type string (e.g., "RERANK")
   • base_model -- the base model identifier
4. Stores the returned custom model object in self._finetune_model.

Method: get_finetuned_model

def get_finetuned_model(self, top_n: int = 5) -> CohereRerank

Returns a LlamaIndex CohereRerank postprocessor configured with the fine-tuned model.

Parameter	Type	Default	Description
top_n	int	5	Number of top-ranked results the reranker should return

Workflow:

1. Checks that self._finetune_model is not None; raises RuntimeError if finetune() has not been called yet.
2. Returns CohereRerank(model=self._finetune_model.id, top_n=top_n, api_key=self.api_key).

Environment Variables

Variable	Required	Default	Purpose
COHERE_API_KEY	Yes (unless api_key parameter is provided)	none	API key for Cohere authentication

Error Handling

Condition	Exception	Message
cohere package not installed	ImportError	"Cannot import cohere. Please install the package using pip install cohere."
No API key found	ValueError	"Must pass in cohere api key or specify via COHERE_API_KEY environment variable"
get_finetuned_model called before finetune	RuntimeError	"Finetuned model is not set yet. Please run the finetune method first."

Note: The API key fallback uses os.environ["COHERE_API_KEY"] which raises KeyError if the variable is not set, but the except clause catches IndexError instead. This means a missing environment variable would raise an unhandled KeyError rather than the intended ValueError.

See Also

• Run_llama_Llama_index_Reranker_Dataset_Gen -- Dataset generation for Cohere reranker fine-tuning
• Run_llama_Llama_index_CrossEncoderFinetuneEngine -- Cross-encoder fine-tuning engine (sentence-transformers based)