Add Gigachat model

[seldereyy]

Add a new model to the library using the API with chat templates. For authorization set environmental variables "GIGACHAT_CREDENTIALS" and "GIGACHAT_SCOPE" for your API auth_data and scope (GIGACHAT_API_CORP or GIGACHAT_API_PERS) respectively.

Here are some usage examples: Run gsm8k with logger:

lm_eval --model gigachat_llms \
    --model_args  model="GigaChat",profanity_check=False\
    --tasks gsm8k \
    --device cpu \
    --write_out \
    --log_samples \
    --output_path './gsm8k.json'

Before using the chat template, set the description in the YAML config. For example, for gsm8k, use the following description: "You have a simple math problem in front of you. Solve it by writing down all the arguments and the answer: final digits after symbols ####." We recommend to use chat template with --system_instruction in order to get better results.

Apply the chat template using the following command:

lm_eval --model gigachat_llms \
    --model_args  model="GigaChat",profanity_check=False\
    --apply_chat_template
    --tasks gsm8k \
    --device cpu \
    --write_out \
    --log_samples \
    --output_path './gsm8k.json'