replace Capybara with SlimOrca (compatible ShareGPT format)

This commit is contained in:
2026-05-13 12:58:29 +00:00
parent c02a76f132
commit 3f29fa017b

View File

@@ -29,15 +29,16 @@ default_system_message: >-
a clear, useful, conversational answer as if talking to a knowledgeable friend. a clear, useful, conversational answer as if talking to a knowledgeable friend.
# --- Datasets --- # --- Datasets ---
# Both Capybara and OpenHermes use ShareGPT format: # Both use ShareGPT format: conversations field, from/value keys
# field: conversations, role key: from, content key: value # SlimOrca: ~15k sample of high-quality multi-turn conversations
# OpenHermes-2.5: broad instruction coverage, 5% sample (~50k)
datasets: datasets:
- path: LDJnr/Capybara - path: Open-Orca/SlimOrca
type: chat_template type: chat_template
field_messages: conversations field_messages: conversations
message_field_role: from message_field_role: from
message_field_content: value message_field_content: value
split: train split: "train[:3%]"
- path: teknium/OpenHermes-2.5 - path: teknium/OpenHermes-2.5
type: chat_template type: chat_template