Fix: Trying to fix dependency issues

training/train_grpo.py

@@ -46,11 +46,11 @@ if not args.test_local:
     _TRAIN_DEPS = [
         "wandb==0.18.7",
         "datasets==3.0.2",
-        "transformers==4.46.3",
+        "transformers==4.44.2",
         "accelerate==1.0.1",
-        "trl==0.14.0",
+        "trl==0.12.2",
         "peft==0.13.2",
-        "bitsandbytes>=0.49.0",
+        "bitsandbytes==0.43.3",
     ]
     print("Installing training dependencies...", flush=True)
     ret = os.system(
@@ -466,7 +466,6 @@ config = GRPOConfig(
     warmup_steps=10 if args.test else 30,
     num_generations=_num_gen,
     max_completion_length=_max_comp,
-    temperature=0.9,
     logging_steps=5,
     save_steps=50,
     report_to="wandb" if WANDB_API_KEY else "none",
@@ -477,7 +476,7 @@ trainer = GRPOTrainer(
     args=config,
     train_dataset=make_dataset(0),
     reward_funcs=reward_fn,
-    processing_class=tokenizer,
+    tokenizer=tokenizer,
 )
 
 # ── Curriculum callback ───────────────────────────────────────────────────────