From b9ceebfe7e815e2972756a5e20e7203d7495d48b Mon Sep 17 00:00:00 2001
From: tocmo0nlord <tocmo0nlord@192.168.1.63>
Date: Wed, 13 May 2026 12:52:25 +0000
Subject: [PATCH] fix deprecated type:sharegpt and flash_attention config keys

---
 human_chat_qlora.yml | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/human_chat_qlora.yml b/human_chat_qlora.yml
index dbe6e4db3..79ee0b40b 100644
--- a/human_chat_qlora.yml
+++ b/human_chat_qlora.yml
@@ -33,13 +33,11 @@ default_system_message: >-
 # OpenHermes-2.5: broad instruction coverage — sampled to 5% to keep balance
 datasets:
   - path: LDJnr/Capybara
-    type: sharegpt
-    conversation: llama3
+    type: chat_template
     split: train
 
   - path: teknium/OpenHermes-2.5
-    type: sharegpt
-    conversation: llama3
+    type: chat_template
     split: "train[:5%]"
 
 dataset_prepared_path: last_run_prepared
@@ -75,7 +73,7 @@ tf32: false
 
 # --- Memory & speed ---
 gradient_checkpointing: true
-flash_attention: true
+attn_implementation: flash_attention_2
 
 # --- Logging & checkpointing ---
 logging_steps: 10