Add configs/kaggle_grpo.yaml

configs/kaggle_grpo.yaml

@@ -0,0 +1,45 @@
+# Kaggle GRPO defaults — T4 x2, 4-bit LoRA, eval-before-checkpoint
+accelerator: t4x2
+seed: 42
+
+model:
+  preferred:
+    - Qwen/Qwen2.5-Coder-1.5B-Instruct
+    - Qwen/Qwen2.5-Coder-0.5B-Instruct
+    - Qwen/Qwen2.5-1.5B-Instruct
+    - HuggingFaceTB/SmolLM2-1.7B-Instruct
+    - google/gemma-2-2b-it
+  max_seq_length: 2048
+  load_in_4bit: true
+  lora_r: 16
+  lora_alpha: 32
+  lora_dropout: 0.0
+
+data:
+  hf_dataset_id: vanishingradient/deepagents-rl-mbpp
+  train_split: train
+  eval_split: eval
+  prompt_column: prompt
+
+grpo:
+  learning_rate: 1.0e-6
+  num_generations: 4
+  max_completion_length: 512
+  per_device_train_batch_size: 1
+  gradient_accumulation_steps: 4
+  max_steps: 40
+  logging_steps: 1
+  # Save every N optimizer steps; ALWAYS run held-out eval + Langfuse BEFORE save
+  save_steps: 5
+  eval_before_checkpoint:
+    enabled: true
+    n_problems: 16
+    temperature: 0.0
+    max_new_tokens: 512
+  push_to_hub: true
+  hub_model_id: vanishingradient/deepagents-rl-coding
+
+observability:
+  wandb_project: deepagents-rl-coding
+  langfuse_ compulsory: true
+  langfuse_eval_trace_name: kaggle-ckpt-eval