increase context len, decrease bsz for 8b (working)

casteryh · casteryh · commit 4318d83f269e · 2025-11-21T12:25:24.000-08:00
diff --git a/apps/grpo/qwen3_8b.yaml b/apps/grpo/qwen3_8b.yaml
@@ -3,9 +3,9 @@
 
 # Global configuration
 group_size: 8
-local_batch_size: 12 # per-device batch size
+local_batch_size: 8 # per-device batch size
 max_req_tokens: 1024
-max_res_tokens: 1024
+max_res_tokens: 2048
 model: "Qwen/Qwen3-8B"
 off_by_n: 1 # Off by one by default
 
