feat(trl): improve eval dataset handling and documentation

- Add train/eval split to all training examples (SFT, DPO)
- Add eval_strategy configuration with proper dataset requirements
- Add critical troubleshooting section for training hangs
- Improve documentation with examples of correct vs incorrect patterns
- Make utility scripts executable (convert_to_gguf, estimate_cost, validate_dataset)
- Update references to point to example scripts for production training
- Remove outdated 'Quick Demo' and 'Production with Checkpoints' sections

FIXES: Training jobs hanging when eval_strategy is set without eval_dataset
IMPROVES: User guidance on best practices for monitoring and evaluation
ADDS: Comprehensive troubleshooting guide for common issues

trl/SKILL.md

@@ -156,15 +156,21 @@ trackio.init(project="my-training", space_id="username/my-dashboard")
 
 dataset = load_dataset("trl-lib/Capybara", split="train")
 
+# Create train/eval split for monitoring
+dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
+
 trainer = SFTTrainer(
     model="Qwen/Qwen2.5-0.5B",
-    train_dataset=dataset,
+    train_dataset=dataset_split["train"],
+    eval_dataset=dataset_split["test"],
     peft_config=LoraConfig(r=16, lora_alpha=32),
     args=SFTConfig(
         output_dir="my-model",
         push_to_hub=True,
         hub_model_id="username/my-model",
         num_train_epochs=3,
+        eval_strategy="steps",
+        eval_steps=50,
         report_to="trackio",
     )
 )

trl/references/training_methods.md

@@ -24,11 +24,14 @@ trainer = SFTTrainer(
         output_dir="my-model",
         push_to_hub=True,
         hub_model_id="username/my-model",
+        eval_strategy="no",  # Disable eval for simple example
     )
 )
 trainer.train()
 ```
 
+**Note:** For production training with evaluation monitoring, see `scripts/train_sft_example.py`
+
 **Documentation:** `hf_doc_fetch("https://huggingface.co/docs/trl/sft_trainer")`
 
 ## Direct Preference Optimization (DPO)
@@ -52,11 +55,14 @@ trainer = DPOTrainer(
     args=DPOConfig(
         output_dir="dpo-model",
         beta=0.1,  # KL penalty coefficient
+        eval_strategy="no",  # Disable eval for simple example
     )
 )
 trainer.train()
 ```
 
+**Note:** For production training with evaluation monitoring, see `scripts/train_dpo_example.py`
+
 **Documentation:** `hf_doc_fetch("https://huggingface.co/docs/trl/dpo_trainer")`
 
 ## Group Relative Policy Optimization (GRPO)

trl/references/training_patterns.md

@@ -2,79 +2,6 @@
 
 This guide provides common training patterns and use cases for TRL on Hugging Face Jobs.
 
-## Quick Demo (5-10 minutes)
-
-Test setup with minimal training:
-
-```python
-hf_jobs("uv", {
-    "script": "https://raw.githubusercontent.com/huggingface/trl/main/examples/scripts/sft.py",
-    "script_args": [
-        "--model_name_or_path", "Qwen/Qwen2.5-0.5B",
-        "--dataset_name", "trl-lib/Capybara",
-        "--dataset_train_split", "train[:50]",  # Only 50 examples
-        "--max_steps", "10",
-        "--output_dir", "demo",
-        "--push_to_hub",
-        "--hub_model_id", "username/demo"
-    ],
-    "flavor": "t4-small",
-    "timeout": "15m",
-    "secrets": {"HF_TOKEN": "$HF_TOKEN"}
-})
-```
-
-**Note:** The TRL maintained script above doesn't include Trackio. For production training with monitoring, see `scripts/train_sft_example.py` for a complete template with Trackio integration.
-
-## Production with Checkpoints
-
-Full training with intermediate saves. Use this pattern for long training runs where you want to save progress:
-
-```python
-hf_jobs("uv", {
-    "script": """
-# /// script
-# dependencies = ["trl>=0.12.0", "peft>=0.7.0", "trackio"]
-# ///
-
-from datasets import load_dataset
-from peft import LoraConfig
-from trl import SFTTrainer, SFTConfig
-import trackio
-
-trackio.init(project="production-training", space_id="username/my-dashboard")
-
-dataset = load_dataset("trl-lib/Capybara", split="train")
-
-config = SFTConfig(
-    output_dir="my-model",
-    push_to_hub=True,
-    hub_model_id="username/my-model",
-    hub_strategy="every_save",  # Push each checkpoint
-    save_strategy="steps",
-    save_steps=100,
-    save_total_limit=3,
-    num_train_epochs=3,
-    report_to="trackio",
-)
-
-trainer = SFTTrainer(
-    model="Qwen/Qwen2.5-0.5B",
-    train_dataset=dataset,
-    args=config,
-    peft_config=LoraConfig(r=16, lora_alpha=32),
-)
-
-trainer.train()
-trainer.push_to_hub()
-trackio.finish()
-""",
-    "flavor": "a10g-large",
-    "timeout": "6h",
-    "secrets": {"HF_TOKEN": "$HF_TOKEN"}
-})
-```
-
 ## Multi-GPU Training
 
 Automatic distributed training across multiple GPUs. TRL/Accelerate handles distribution automatically:
@@ -116,18 +43,24 @@ trackio.init(project="dpo-training", space_id="username/my-dashboard")
 
 dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")
 
+# Create train/eval split
+dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
+
 config = DPOConfig(
     output_dir="dpo-model",
     push_to_hub=True,
     hub_model_id="username/dpo-model",
     num_train_epochs=1,
     beta=0.1,  # KL penalty coefficient
+    eval_strategy="steps",
+    eval_steps=50,
     report_to="trackio",
 )
 
 trainer = DPOTrainer(
     model="Qwen/Qwen2.5-0.5B-Instruct",  # Use instruct model as base
-    train_dataset=dataset,
+    train_dataset=dataset_split["train"],
+    eval_dataset=dataset_split["test"],  # IMPORTANT: Provide eval_dataset when eval_strategy is enabled
     args=config,
 )
 
@@ -169,26 +102,66 @@ hf_jobs("uv", {
 
 | Use Case | Pattern | Hardware | Time |
 |----------|---------|----------|------|
-| Test setup | Quick Demo | t4-small | 5-10 min |
-| Small dataset (<1K) | Production w/ Checkpoints | t4-medium | 30-60 min |
-| Medium dataset (1-10K) | Production w/ Checkpoints | a10g-large | 2-6 hours |
+| SFT training | `scripts/train_sft_example.py` | a10g-large | 2-6 hours |
 | Large dataset (>10K) | Multi-GPU | a10g-largex2 | 4-12 hours |
 | Preference learning | DPO Training | a10g-large | 2-4 hours |
 | Online RL | GRPO Training | a10g-large | 3-6 hours |
 
+## Critical: Evaluation Dataset Requirements
+
+**⚠️ IMPORTANT**: If you set `eval_strategy="steps"` or `eval_strategy="epoch"`, you **MUST** provide an `eval_dataset` to the trainer, or the training will hang.
+
+### ✅ CORRECT - With eval dataset:
+```python
+dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
+
+trainer = SFTTrainer(
+    model="Qwen/Qwen2.5-0.5B",
+    train_dataset=dataset_split["train"],
+    eval_dataset=dataset_split["test"],  # ← MUST provide when eval_strategy is enabled
+    args=SFTConfig(eval_strategy="steps", ...),
+)
+```
+
+### ❌ WRONG - Will hang:
+```python
+trainer = SFTTrainer(
+    model="Qwen/Qwen2.5-0.5B",
+    train_dataset=dataset,
+    # NO eval_dataset but eval_strategy="steps" ← WILL HANG
+    args=SFTConfig(eval_strategy="steps", ...),
+)
+```
+
+### Option: Disable evaluation if no eval dataset
+```python
+config = SFTConfig(
+    eval_strategy="no",  # ← Explicitly disable evaluation
+    # ... other config
+)
+
+trainer = SFTTrainer(
+    model="Qwen/Qwen2.5-0.5B",
+    train_dataset=dataset,
+    # No eval_dataset needed
+    args=config,
+)
+```
+
 ## Best Practices
 
-1. **Always start with Quick Demo** - Verify setup before long runs
-2. **Use checkpoints for runs >1 hour** - Protect against failures
-3. **Enable Trackio** - Monitor progress in real-time
-4. **Add 20-30% buffer to timeout** - Account for loading/saving overhead
-5. **Test with small dataset slice first** - Use `"train[:100]"` to verify code
+1. **Use train/eval splits** - Create evaluation split for monitoring progress
+2. **Enable Trackio** - Monitor progress in real-time
+3. **Add 20-30% buffer to timeout** - Account for loading/saving overhead
+4. **Test with TRL official scripts first** - Use maintained examples before custom code
+5. **Always provide eval_dataset** - When using eval_strategy, or set to "no"
 6. **Use multi-GPU for large models** - 7B+ models benefit significantly
 
 ## See Also
 
-- `scripts/train_sft_example.py` - Complete SFT template with Trackio
+- `scripts/train_sft_example.py` - Complete SFT template with Trackio and eval split
 - `scripts/train_dpo_example.py` - Complete DPO template
 - `scripts/train_grpo_example.py` - Complete GRPO template
 - `references/hardware_guide.md` - Detailed hardware specifications
 - `references/training_methods.md` - Overview of all TRL training methods
+- `references/troubleshooting.md` - Common issues and solutions

trl/references/troubleshooting.md

@@ -2,6 +2,49 @@
 
 Common issues and solutions when training with TRL on Hugging Face Jobs.
 
+## Training Hangs at "Starting training..." Step
+
+**Problem:** Job starts but hangs at the training step - never progresses, never times out, just sits there.
+
+**Root Cause:** Using `eval_strategy="steps"` or `eval_strategy="epoch"` without providing an `eval_dataset` to the trainer.
+
+**Solution:**
+
+**Option A: Provide eval_dataset (recommended)**
+```python
+# Create train/eval split
+dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
+
+trainer = SFTTrainer(
+    model="Qwen/Qwen2.5-0.5B",
+    train_dataset=dataset_split["train"],
+    eval_dataset=dataset_split["test"],  # ← MUST provide when eval_strategy is enabled
+    args=SFTConfig(
+        eval_strategy="steps",
+        eval_steps=50,
+        ...
+    ),
+)
+```
+
+**Option B: Disable evaluation**
+```python
+trainer = SFTTrainer(
+    model="Qwen/Qwen2.5-0.5B",
+    train_dataset=dataset,
+    # No eval_dataset
+    args=SFTConfig(
+        eval_strategy="no",  # ← Explicitly disable
+        ...
+    ),
+)
+```
+
+**Prevention:**
+- Always create train/eval split for better monitoring
+- Use `dataset.train_test_split(test_size=0.1, seed=42)`
+- Check example scripts: `scripts/train_sft_example.py` includes proper eval setup
+
 ## Job Times Out
 
 **Problem:** Job terminates before training completes, all progress lost.
@@ -208,5 +251,6 @@ If issues persist:
    - `references/hub_saving.md` - Hub authentication issues
    - `references/hardware_guide.md` - Hardware selection and specs
    - `references/uv_scripts_guide.md` - UV script format issues
+   - `references/training_patterns.md` - Eval dataset requirements
 
 4. **Ask in HF forums:** https://discuss.huggingface.co/

trl/scripts/train_dpo_example.py

@@ -43,9 +43,18 @@ trackio.init(
 )
 
 # Load preference dataset
+print("📦 Loading dataset...")
 dataset = load_dataset("trl-lib/ultrafeedback_binarized", split="train")
 print(f"✅ Dataset loaded: {len(dataset)} preference pairs")
 
+# Create train/eval split
+print("🔀 Creating train/eval split...")
+dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
+train_dataset = dataset_split["train"]
+eval_dataset = dataset_split["test"]
+print(f"   Train: {len(train_dataset)} pairs")
+print(f"   Eval: {len(eval_dataset)} pairs")
+
 # Training configuration
 config = DPOConfig(
     # CRITICAL: Hub settings
@@ -69,6 +78,10 @@ config = DPOConfig(
     save_steps=100,
     save_total_limit=2,
 
+    # Evaluation - IMPORTANT: Only enable if eval_dataset provided
+    eval_strategy="steps",
+    eval_steps=100,
+
     # Optimization
     warmup_ratio=0.1,
     lr_scheduler_type="cosine",
@@ -79,9 +92,11 @@ config = DPOConfig(
 
 # Initialize and train
 # Note: DPO requires an instruct-tuned model as the base
+print("🎯 Initializing trainer...")
 trainer = DPOTrainer(
     model="Qwen/Qwen2.5-0.5B-Instruct",  # Use instruct model, not base model
-    train_dataset=dataset,
+    train_dataset=train_dataset,
+    eval_dataset=eval_dataset,  # CRITICAL: Must provide eval_dataset when eval_strategy is enabled
     args=config,
 )
 

trl/scripts/train_sft_example.py

@@ -16,6 +16,7 @@ This script demonstrates:
 - Trackio integration for real-time monitoring
 - LoRA/PEFT for efficient training
 - Proper Hub saving configuration
+- Train/eval split for monitoring
 - Checkpoint management
 - Optimized training parameters
 
@@ -48,10 +49,19 @@ trackio.init(
     }
 )
 
-# Load and validate
+# Load dataset
+print("📦 Loading dataset...")
 dataset = load_dataset("trl-lib/Capybara", split="train")
 print(f"✅ Dataset loaded: {len(dataset)} examples")
 
+# Create train/eval split
+print("🔀 Creating train/eval split...")
+dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
+train_dataset = dataset_split["train"]
+eval_dataset = dataset_split["test"]
+print(f"   Train: {len(train_dataset)} examples")
+print(f"   Eval: {len(eval_dataset)} examples")
+
 # Training configuration
 config = SFTConfig(
     # CRITICAL: Hub settings
@@ -72,6 +82,10 @@ config = SFTConfig(
     save_steps=100,
     save_total_limit=2,
 
+    # Evaluation - IMPORTANT: Only enable if eval_dataset provided
+    eval_strategy="steps",
+    eval_steps=100,
+
     # Optimization
     warmup_ratio=0.1,
     lr_scheduler_type="cosine",
@@ -91,9 +105,11 @@ peft_config = LoraConfig(
 )
 
 # Initialize and train
+print("🎯 Initializing trainer...")
 trainer = SFTTrainer(
     model="Qwen/Qwen2.5-0.5B",
-    train_dataset=dataset,
+    train_dataset=train_dataset,
+    eval_dataset=eval_dataset,  # CRITICAL: Must provide eval_dataset when eval_strategy is enabled
     args=config,
     peft_config=peft_config,
 )