From c02a76f13249282a941760a87b03d20275369ada Mon Sep 17 00:00:00 2001 From: tocmo0nlord Date: Wed, 13 May 2026 12:56:03 +0000 Subject: [PATCH] fix field_messages mapping for Capybara/OpenHermes ShareGPT format --- human_chat_qlora.yml | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/human_chat_qlora.yml b/human_chat_qlora.yml index 79ee0b40b..cb17d5b6e 100644 --- a/human_chat_qlora.yml +++ b/human_chat_qlora.yml @@ -29,15 +29,21 @@ default_system_message: >- a clear, useful, conversational answer as if talking to a knowledgeable friend. # --- Datasets --- -# Capybara: ~16k carefully curated multi-turn conversations — best for natural tone -# OpenHermes-2.5: broad instruction coverage — sampled to 5% to keep balance +# Both Capybara and OpenHermes use ShareGPT format: +# field: conversations, role key: from, content key: value datasets: - path: LDJnr/Capybara type: chat_template + field_messages: conversations + message_field_role: from + message_field_content: value split: train - path: teknium/OpenHermes-2.5 type: chat_template + field_messages: conversations + message_field_role: from + message_field_content: value split: "train[:5%]" dataset_prepared_path: last_run_prepared