diffusion alt: custom loss impl

examples/llama-3/diffusion-3.2-1b-pretrain.yaml

@@ -0,0 +1,57 @@
+base_model: meta-llama/Llama-3.2-1B
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+pretraining_dataset:
+  - path: wikitext
+    name: wikitext-103-raw-v1
+    type: completion
+    field: text
+
+plugins:
+  - diffusion.DiffusionPlugin
+noise_schedule: cosine
+min_mask_ratio: 0.15
+max_mask_ratio: 0.85
+eps: 5e-4
+importance_weighting: true
+mask_token_id: 128002
+generate_samples: true
+generation_interval: 10
+
+output_dir: ./outputs/model-out
+
+sequence_len: 512
+sample_packing: true
+
+gradient_accumulation_steps: 8
+micro_batch_size: 4
+max_steps: 10000
+
+optimizer: adamw_8bit
+lr_scheduler: cosine
+learning_rate: 3e-4
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+sdp_attention: true
+
+warmup_steps: 1000
+
+save_strategy: steps
+save_steps: 1000
+
+special_tokens:
+  pad_token: "<|end_of_text|>"
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/llama-3/diffusion-3.2-1b-sft.yaml

@@ -0,0 +1,58 @@
+base_model: meta-llama/Llama-3.2-1B
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+val_set_size: 0.05
+
+plugins:
+  - diffusion.DiffusionPlugin
+noise_schedule: cosine
+min_mask_ratio: 0.1
+max_mask_ratio: 0.9
+num_diffusion_steps: 128
+eps: 1e-3
+importance_weighting: true
+mask_token_id: 128002
+
+output_dir: ./outputs/model-out
+
+sequence_len: 512
+sample_packing: true
+eval_sample_packing: true
+
+gradient_accumulation_steps: 4
+micro_batch_size: 4
+num_epochs: 1
+
+optimizer: adamw_8bit
+lr_scheduler: cosine
+learning_rate: 1e-5
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+sdp_attention: true
+
+warmup_steps: 1000
+
+save_strategy: steps
+eval_strategy: steps
+save_steps: 500
+eval_steps: 500
+
+special_tokens:
+  pad_token: "<|end_of_text|>"
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

src/axolotl/core/builders/causal.py

@@ -10,6 +10,7 @@ import transformers
 from transformers import (
     DataCollatorWithFlattening,
     EarlyStoppingCallback,
+    Trainer,
 )
 from trl.trainer.utils import RewardDataCollatorWithPadding
 
@@ -385,10 +386,11 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                 **data_collator_kwargs,
             )
         sig = inspect.signature(trainer_cls)
-        if "processing_class" in sig.parameters:
+        if "processing_class" in sig.parameters or issubclass(trainer_cls, Trainer):
             trainer_kwargs["processing_class"] = self.tokenizer
         elif "tokenizer" in sig.parameters:
             trainer_kwargs["tokenizer"] = self.tokenizer
+
         if (
             trainer_cls not in [AxolotlRewardTrainer, AxolotlPRMTrainer]
             and self.cfg.datasets is not None

src/axolotl/core/trainers/base.py

@@ -82,7 +82,9 @@ class AxolotlTrainer(
         super().__init__(*_args, **kwargs)
 
         self.train_data_collator = self.data_collator
-        self._stored_metrics = defaultdict(lambda: defaultdict(list))
+        self._stored_metrics = defaultdict(
+            lambda: defaultdict(lambda: {"values": [], "reduction": "mean"})
+        )
         if self.args.orpo_alpha:
             self.loss_fct = torch.nn.CrossEntropyLoss(reduction="none")
 
@@ -573,9 +575,26 @@ class AxolotlTrainer(
         """
         # logs either has 'loss' or 'eval_loss'
         train_eval = "train" if "loss" in logs else "eval"
-        # Add averaged stored metrics to logs
-        for key, metrics in self._stored_metrics[train_eval].items():
-            logs[key] = torch.tensor(metrics).mean().item()
+
+        # Add reduced stored metrics to logs
+        for key, metric_data in self._stored_metrics[train_eval].items():
+            values = torch.tensor(metric_data["values"])
+            reduction_type = metric_data["reduction"]
+
+            if reduction_type == "mean":
+                logs[key] = values.mean().item()
+            elif reduction_type == "min":
+                logs[key] = values.min().item()
+            elif reduction_type == "max":
+                logs[key] = values.max().item()
+            elif reduction_type == "sum":
+                logs[key] = values.sum().item()
+            else:
+                raise NotImplementedError(
+                    "Metric reduction must be one of [mean, min, max, sum]"
+                )
+
+            logs[key] = round(logs[key], 4)
 
         if is_main_process():
             # Add memory usage
@@ -592,10 +611,27 @@ class AxolotlTrainer(
         return super().log(logs, start_time)
 
     def store_metrics(
-        self, metrics: dict[str, float], train_eval: Literal["train", "eval"] = "train"
+        self,
+        metrics: dict[str, float] | dict[str, tuple[int | float, str]],
+        train_eval: Literal["train", "eval"] = "train",
+        reduction: Literal["mean", "min", "max", "sum"] = "mean",
     ) -> None:
+        """
+        Store metrics with specified reduction type.
+
+        Args:
+            metrics: Dictionary of metric names to values, or metric names to (value,
+                reduction_type) tuples.
+            train_eval: Whether this is for training or evaluation.
+        """
         for key, value in metrics.items():
-            self._stored_metrics[train_eval][key].append(value)
+            if isinstance(value, tuple):
+                metric_value, metric_reduction = value
+            else:
+                metric_value, metric_reduction = value, reduction
+
+            self._stored_metrics[train_eval][key]["values"].append(metric_value)
+            self._stored_metrics[train_eval][key]["reduction"] = metric_reduction
 
     def _save_checkpoint(self, model, trial, **kwargs):
         # make sure the checkpoint dir exists, since trainer is flakey

src/axolotl/integrations/base.py

@@ -147,7 +147,7 @@ class BasePlugin:
         """
 
     # pylint: disable=unused-argument
-    def get_trainer_cls(self, cfg: DictDefault) -> Trainer | None:
+    def get_trainer_cls(self, cfg: DictDefault) -> type[Trainer] | None:
         """Returns a custom class for the trainer.
 
         Args:

src/axolotl/integrations/diffusion/README.md

@@ -0,0 +1,125 @@
+# Diffusion LM Training Plugin for Axolotl
+
+This plugin enables diffusion language model training using the LLaDA (Large Language
+And Diffusion Assistant) approach within the Axolotl framework.
+
+## Overview
+
+LLaDA is a diffusion-based approach to language model training that uses:
+- **Random token masking** during training instead of next-token prediction
+- **Bidirectional attention** to allow the model to see the full context
+- **Importance weighting** based on masking probabilities for stable training
+
+This approach can lead to more robust language models with better understanding of
+bidirectional context.
+
+## Installation
+
+The plugin is included with Axolotl. To use it, simply add the plugin configuration to
+your training config.
+
+## Quickstart
+
+### Basic Configuration
+
+Add the following to your Axolotl configuration YAML:
+
+```yaml
+# Enable diffusion LM training plugin
+plugins:
+  - axolotl.integrations.diffusion.DiffusionPlugin
+
+# Diffusion-specific configuration
+noise_schedule: linear  # or "cosine"
+min_mask_ratio: 0.1
+max_mask_ratio: 0.9
+num_diffusion_steps: 128
+eps: 1e-3
+importance_weighting: true
+mask_token_id: 128002
+
+# Sample generation (optional)
+generate_samples: true
+generation_interval: 100
+num_generation_samples: 3
+generation_steps: 128
+generation_temperature: 0.0
+generation_max_length: 100
+
+# Model configuration
+base_model: meta-llama/Llama-3.2-1B
+model_type: llama
+
+# Standard Axolotl configuration
+datasets:
+  - path: your_dataset
+    ...
+
+# Other config
+sequence_len: 1024
+micro_batch_size: 8
+gradient_accumulation_steps: 4
+learning_rate: 3e-4
+```
+
+## Supported Models
+
+Any models that support 4D attention masks should work out of the box. If not, please
+create an [issue](https://github.com/axolotl-ai-cloud/axolotl/issues)!
+
+## How It Works
+
+### Random Masking
+During training, tokens are randomly masked based on a sampled timestep:
+- Sample timestep `t` uniformly from [0, 1]
+- Calculate masking probability: `p = (1 - eps) * t + eps`
+- Randomly mask tokens with probability `p`
+
+### Bidirectional Attention
+The plugin uses native 4D attention masks to:
+- Enable bidirectional attention without patches
+- Allow all tokens to attend to all other tokens
+- Maintain proper padding masks
+- Work with modern `transformers` models out of the box
+
+### Diffusion Loss
+
+Loss is computed only on masked tokens with (optional) importance weighting:
+
+```python
+loss = sum(cross_entropy(pred, target) / p_mask) / total_tokens
+```
+
+## Sample Generation
+
+When `generate_samples: true`, the plugin generates samples during training:
+
+```
+Sample 1:
+   Original (45 tokens): The quick brown fox jumps over the lazy dog...
+   Masked (18/45 tokens, 40.0%): The [MASK] [MASK] fox [MASK] over [MASK] lazy [MASK]...
+   Generated: The quick brown fox jumps over the lazy dog...
+```
+
+Samples are logged to console and wandb (if enabled).
+
+## Metrics and Monitoring
+
+The plugin adds several metrics to track diffusion training:
+
+- `train/loss`: Weighted diffusion loss
+- `train/accuracy`: Accuracy on masked tokens
+- `train/mask_ratio`: Average fraction of tokens masked
+- `train/num_masked_tokens`: Number of tokens masked
+- `train/avg_p_mask`: Average masking probability
+- `train/ce_loss`: Unweighted cross-entropy loss
+- `train/importance_weight_avg`: Average importance weight
+
+## Limitations
+
+- No flash attention support
+
+## References
+
+- [LLaDA Paper](https://arxiv.org/abs/2404.10406)
+- [Axolotl Documentation](https://docs.axolotl.ai/)

src/axolotl/integrations/diffusion/__init__.py

@@ -0,0 +1,6 @@
+"""Diffusion LM training plugin init."""
+
+from .args import DiffusionArgs
+from .plugin import DiffusionPlugin
+
+__all__ = ["DiffusionArgs", "DiffusionPlugin"]

src/axolotl/integrations/diffusion/args.py

@@ -0,0 +1,70 @@
+"""Config args for diffusion LM training."""
+
+from typing import Literal
+
+from pydantic import BaseModel, Field
+
+
+class DiffusionArgs(BaseModel):
+    """Arguments for diffusion LM training plugin."""
+
+    # Noise schedule config
+    noise_schedule: Literal["linear", "cosine"] = Field(
+        default="linear", description="Type of noise schedule for diffusion training"
+    )
+    min_mask_ratio: float = Field(
+        default=0.1,
+        ge=0.0,
+        le=1.0,
+        description="Minimum masking ratio for diffusion noise schedule",
+    )
+    max_mask_ratio: float = Field(
+        default=0.9,
+        ge=0.0,
+        le=1.0,
+        description="Maximum masking ratio for diffusion noise schedule",
+    )
+    num_diffusion_steps: int = Field(
+        default=128, ge=1, description="Number of diffusion timesteps"
+    )
+    eps: float = Field(
+        default=1e-3,
+        ge=0.0,
+        le=1.0,
+        description="Epsilon value for minimum masking probability in forward process",
+    )
+
+    # Training config
+    importance_weighting: bool = Field(
+        default=True,
+        description="Apply importance weighting to loss based on masking probability",
+    )
+    mask_token_id: int = Field(
+        default=128002,
+        description=(
+            "Token ID to use for masking. Default is 128002 "
+            "(<|reserved_special_token_0|> for Llama 3.2)"
+        ),
+    )
+
+    # Sample generation config
+    generate_samples: bool = Field(
+        default=True, description="Enable sample generation during training"
+    )
+    generation_interval: int = Field(
+        default=100, ge=1, description="Generate samples every N steps"
+    )
+    num_generation_samples: int = Field(
+        default=3, ge=1, description="Number of samples to generate each time"
+    )
+    generation_steps: int = Field(
+        default=128, ge=1, description="Number of diffusion steps for generation"
+    )
+    generation_temperature: float = Field(
+        default=0.0,
+        ge=0.0,
+        description="Temperature for generation sampling (0.0 = deterministic)",
+    )
+    generation_max_length: int = Field(
+        default=100, ge=1, description="Maximum sequence length for generation"
+    )

src/axolotl/integrations/diffusion/callbacks.py

@@ -0,0 +1,113 @@
+"""Callbacks for diffusion training."""
+
+import wandb
+from transformers.trainer_callback import TrainerCallback, TrainerControl, TrainerState
+from transformers.training_args import TrainingArguments
+
+from axolotl.utils.logging import get_logger
+
+from .generation import generate_samples
+
+LOG = get_logger(__name__)
+
+
+class DiffusionGenerationCallback(TrainerCallback):
+    """Callback for generating samples during diffusion training."""
+
+    def __init__(self, trainer):
+        self.trainer = trainer
+
+    # pylint: disable=unused-argument
+    def on_step_end(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ):
+        """Generate samples at specified intervals."""
+        if (
+            state.global_step > 0
+            and state.global_step % self.trainer.config.generation_interval == 0
+        ):
+            # Use eval dataloader if available, otherwise use train dataloader
+            if (
+                hasattr(self.trainer, "eval_dataset")
+                and self.trainer.eval_dataset is not None
+            ):
+                dataloader = self.trainer.callback_handler.eval_dataloader
+            else:
+                dataloader = self.trainer.callback_handler.train_dataloader
+
+            # Generate samples
+            samples = generate_samples(
+                model=self.trainer.model,
+                tokenizer=self.trainer.tokenizer,
+                dataloader=dataloader,
+                num_generation_samples=self.trainer.config.num_generation_samples,
+                max_length=self.trainer.config.generation_max_length,
+                num_diffusion_steps=self.trainer.config.generation_steps,
+                temperature=self.trainer.config.generation_temperature,
+                mask_token_id=self.trainer.config.mask_token_id,
+            )
+
+            # Log samples
+            self._log_samples(samples, state.global_step)
+
+    def _log_samples(self, samples: list, step: int):
+        """Log generated samples."""
+        if not samples:
+            return
+
+        LOG.info("=" * 60)
+        LOG.info("GENERATED SAMPLES")
+        LOG.info("=" * 60)
+
+        for i, sample_data in enumerate(samples, 1):
+            original = sample_data["original"]
+            masked = sample_data["masked"]
+            generated = sample_data["generated"]
+            mask_ratio = sample_data["mask_ratio"]
+            masked_tokens = sample_data["masked_tokens"]
+            total_tokens = sample_data["total_tokens"]
+
+            LOG.info(f"\nSample {i}:")
+            LOG.info(f"\tOriginal ({total_tokens} tokens): {original}")
+            LOG.info(
+                f"\tMasked ({masked_tokens}/{total_tokens} tokens, "
+                f"{mask_ratio:.1%}): {masked}"
+            )
+            LOG.info(f"\tGenerated: {generated}")
+
+        LOG.info("=" * 60)
+
+        if self.trainer.config.use_wandb and self.trainer.state.is_world_process_zero:
+            if wandb.run is not None:
+                wandb.log(
+                    {
+                        "generated_samples": wandb.Table(
+                            columns=[
+                                "step",
+                                "original",
+                                "masked",
+                                "generated",
+                                "mask_ratio",
+                                "masked_tokens",
+                                "total_tokens",
+                            ],
+                            data=[
+                                [
+                                    step,
+                                    sample["original"],
+                                    sample["masked"],
+                                    sample["generated"],
+                                    f"{sample['mask_ratio']:.1%}",
+                                    sample["masked_tokens"],
+                                    sample["total_tokens"],
+                                ]
+                                for sample in samples
+                            ],
+                        )
+                    },
+                    step=step,
+                )

src/axolotl/integrations/diffusion/generation.py

@@ -0,0 +1,269 @@
+"""Sample generation utilities for diffusion training."""
+
+import logging
+from typing import Any, List, Optional
+
+import torch
+
+logger = logging.getLogger(__name__)
+
+
+def generate_samples(
+    model: torch.nn.Module,
+    tokenizer: Any,
+    dataloader: Optional[Any] = None,
+    num_generation_samples: int = 3,
+    max_length: int = 100,
+    num_diffusion_steps: int = 128,
+    temperature: float = 0.0,
+    mask_token_id: int = 32000,
+) -> List[dict]:
+    """
+    Generate text samples using the diffusion model by randomly masking sequences from
+    the given dataset and running the reverse diffusion process.
+
+    Args:
+        model: The wrapped or unwrapped model
+        tokenizer: Tokenizer for encoding/decoding
+        dataloader: Validation dataloader (for sampling sequences)
+        num_generation_samples: Number of samples to generate
+        max_length: Maximum length of sequences to use
+        num_diffusion_steps: Number of diffusion steps for generation
+        temperature: Temperature for sampling (0.0 = deterministic)
+        mask_token_id: Token ID used for masking
+
+    Returns:
+        List of dictionaries with original text, masked text, and generated text
+    """
+    if dataloader is None:
+        logger.warning("No validation dataloader provided, cannot generate samples")
+        return []
+
+    # Get the actual model (unwrap if needed)
+    unwrapped_model = model.module if hasattr(model, "module") else model
+    unwrapped_model.eval()
+    generations = []
+
+    # Sample sequences from validation dataset
+    sampled_sequences = _sample_sequences_from_dataloader(
+        dataloader, num_generation_samples, max_length, unwrapped_model.device
+    )
+    logger.info(f"Sampled {len(sampled_sequences)} sequences from validation dataset")
+
+    # Generate samples using reverse diffusion process
+    with torch.no_grad():
+        for original_sequence in sampled_sequences:
+            generation_result = _generate(
+                unwrapped_model,
+                tokenizer,
+                original_sequence,
+                num_diffusion_steps,
+                temperature,
+                mask_token_id,
+            )
+            generations.append(generation_result)
+
+    unwrapped_model.train()
+    return generations
+
+
+def _sample_sequences_from_dataloader(
+    dataloader: Any, num_samples: int, max_length: int, device: torch.device
+) -> List[torch.Tensor]:
+    """Sample sequences from validation dataloader."""
+    sampled_sequences = []
+    sample_count = 0
+
+    # Add randomness by skipping a random number of batches
+    skip_batches = torch.randint(0, 6, (1,)).item()
+    batch_count = 0
+
+    for batch in dataloader:
+        # Skip some batches for variety
+        if batch_count < skip_batches:
+            batch_count += 1
+            continue
+
+        if sample_count >= num_samples:
+            break
+
+        batch_count += 1
+        input_ids = batch["input_ids"]
+        attention_mask = batch.get("attention_mask")
+
+        # Randomly sample from sequences in this batch
+        batch_indices = torch.randperm(input_ids.size(0)).tolist()
+
+        for i in batch_indices:
+            if sample_count >= num_samples:
+                break
+
+            # Get actual sequence length (non-padded)
+            if attention_mask is not None:
+                seq_len = attention_mask[i].sum().item()
+            else:
+                seq_len = input_ids.size(1)
+
+            # Limit sequence length to max_length
+            actual_length = min(seq_len, max_length)
+            if actual_length < 10:  # Skip very short sequences
+                continue
+
+            # Extract the sequence
+            sequence = input_ids[i][:actual_length].unsqueeze(0).to(device)
+            sampled_sequences.append(sequence)
+            sample_count += 1
+
+    return sampled_sequences
+
+
+def _generate(
+    model: torch.nn.Module,
+    tokenizer: Any,
+    original_sequence: torch.Tensor,
+    num_diffusion_steps: int,
+    temperature: float,
+    mask_token_id: int,
+) -> dict:
+    """Generate a single sample using reverse diffusion."""
+    # Get original text for comparison
+    original_text = tokenizer.decode(
+        original_sequence[0].cpu(), skip_special_tokens=True
+    )
+
+    # Apply custom masking with random ratio (10% to 70%)
+    total_tokens = original_sequence.size(1)
+    min_ratio, max_ratio = 0.1, 0.7
+    target_mask_ratio = torch.rand(1).item() * (max_ratio - min_ratio) + min_ratio
+    target_masked_tokens = int(total_tokens * target_mask_ratio)
+
+    # Create random mask indices
+    mask_positions = torch.randperm(total_tokens)[:target_masked_tokens]
+    masked_indices = torch.zeros(
+        1, total_tokens, dtype=torch.bool, device=original_sequence.device
+    )
+    masked_indices[0, mask_positions] = True
+
+    # Create masked sequence
+    masked_sequence = original_sequence.clone()
+    masked_sequence[masked_indices] = mask_token_id
+
+    # Calculate actual mask ratio
+    masked_tokens = masked_indices.sum().item()
+    mask_ratio = masked_tokens / total_tokens
+
+    # Get masked text for comparison
+    masked_text = tokenizer.decode(masked_sequence[0].cpu(), skip_special_tokens=False)
+    # Clean up mask token representation
+    masked_text = _clean_masked_text(masked_text, tokenizer, mask_token_id)
+
+    # Run reverse diffusion process
+    sequence = masked_sequence.clone()
+    for step in range(num_diffusion_steps):
+        sequence = _diffusion_step(
+            model, sequence, step, num_diffusion_steps, temperature, mask_token_id
+        )
+
+    # Get final generated text
+    generated_text = tokenizer.decode(sequence[0].cpu(), skip_special_tokens=True)
+
+    return {
+        "original": original_text,
+        "masked": masked_text,
+        "generated": generated_text,
+        "mask_ratio": mask_ratio,
+        "masked_tokens": masked_tokens,
+        "total_tokens": total_tokens,
+        "formatted": (
+            f"Original: '{original_text}' → Masked: '{masked_text}' "
+            f"({mask_ratio:.1%}) → Generated: '{generated_text}'"
+        ),
+    }
+
+
+def _clean_masked_text(masked_text: str, tokenizer: Any, mask_token_id: int) -> str:
+    """Clean up masked text for display."""
+    mask_token_repr = tokenizer.decode([mask_token_id], skip_special_tokens=False)
+    cleaned = masked_text.replace(mask_token_repr, "[MASK]")
+
+    if hasattr(tokenizer, "special_tokens_map"):
+        for token_value in tokenizer.special_tokens_map.values():
+            if token_value and isinstance(token_value, str):
+                cleaned = cleaned.replace(token_value, "")
+
+    cleaned = " ".join(cleaned.split()).strip()
+
+    return cleaned
+
+
+def _diffusion_step(
+    model: torch.nn.Module,
+    sequence: torch.Tensor,
+    step: int,
+    num_diffusion_steps: int,
+    temperature: float,
+    mask_token_id: int,
+) -> torch.Tensor:
+    """Perform a single diffusion step with remasking."""
+    # Only process if there are masked tokens remaining
+    current_mask = sequence == mask_token_id
+    if not current_mask.any():
+        return sequence
+
+    # Create bidirectional attention mask for diffusion
+    batch_size, seq_len = sequence.shape
+    attention_mask = torch.ones(
+        batch_size, 1, seq_len, seq_len, dtype=torch.bool, device=sequence.device
+    )
+
+    # Forward pass
+    outputs = model(input_ids=sequence, attention_mask=attention_mask)
+    logits = outputs.logits
+
+    # Only sample at currently masked positions
+    if current_mask.any():
+        masked_logits = logits[current_mask]
+
+        # Apply temperature scaling
+        if temperature > 0:
+            scaled_logits = masked_logits / temperature
+        else:
+            scaled_logits = masked_logits
+
+        # Suppress mask token in outputs
+        scaled_logits[:, mask_token_id] = -float("inf")
+
+        # Sample predictions
+        if temperature > 0:
+            # Add Gumbel noise for sampling
+            gumbel_noise = -torch.log(
+                -torch.log(torch.rand_like(scaled_logits, dtype=torch.float32))
+            )
+            gumbel_logits = scaled_logits + gumbel_noise
+            predicted_tokens = torch.argmax(gumbel_logits, dim=-1)
+        else:
+            # Deterministic sampling when temperature is 0
+            predicted_tokens = torch.argmax(scaled_logits, dim=-1)
+
+        # Calculate probabilities for confidence scoring
+        probs = torch.softmax(scaled_logits, dim=-1)
+        predicted_token_probs = probs[range(len(predicted_tokens)), predicted_tokens]
+
+        # Determine how many tokens to unmask this step
+        remaining_masked = current_mask.sum().item()
+        if step == num_diffusion_steps - 1:
+            num_to_unmask = remaining_masked
+        else:
+            unmask_ratio = 1.0 / (num_diffusion_steps - step)
+            num_to_unmask = max(1, int(remaining_masked * unmask_ratio))
+
+        # Select highest confidence predictions to unmask
+        if num_to_unmask >= remaining_masked:
+            sequence[current_mask] = predicted_tokens
+        else:
+            _, top_indices = predicted_token_probs.topk(num_to_unmask)
+            mask_positions = torch.where(current_mask)[1]
+            positions_to_unmask = mask_positions[top_indices]
+            sequence[0, positions_to_unmask] = predicted_tokens[top_indices]
+
+    return sequence

src/axolotl/integrations/diffusion/loss.py

@@ -0,0 +1,115 @@
+"""Diffusion LM loss function for integration with transformers LOSS_MAPPING."""
+
+from typing import Optional
+
+import torch
+import torch.nn.functional as F
+
+
+def ForDiffusionLMLoss(
+    logits: torch.Tensor,
+    labels: torch.Tensor,
+    vocab_size: int,
+    config: Optional[dict] = None,
+    inputs: Optional[dict] = None,
+    model: Optional[torch.nn.Module] = None,
+    **kwargs,
+) -> torch.Tensor:
+    """
+    Diffusion Language Modeling loss function.
+
+    This function computes cross-entropy loss only on masked tokens using
+    diffusion info stored by the model patch during forward pass.
+
+    Args:
+        logits: Model predictions [batch_size, seq_len, vocab_size]
+        labels: Ground truth tokens [batch_size, seq_len]
+        vocab_size: Size of vocabulary
+        config: Model configuration (contains diffusion parameters)
+        inputs: Input batch dictionary (contains input_ids, attention_mask)
+        model: The model instance (to access stored diffusion info)
+        **kwargs: Additional arguments
+
+    Returns:
+        loss: Computed diffusion loss
+    """
+    # Get diffusion info stored by model patch
+    if model is None or not hasattr(model, "_diffusion_info"):
+        # Fallback to regular causal LM loss if no diffusion info
+        shift_logits = logits[..., :-1, :].contiguous()
+        shift_labels = labels[..., 1:].contiguous()
+        loss_fct = torch.nn.CrossEntropyLoss()
+        return loss_fct(
+            shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1)
+        )
+
+    diffusion_info = model._diffusion_info
+    original_input_ids = diffusion_info["original_input_ids"]
+    masked_indices = diffusion_info["masked_indices"]
+    p_mask = diffusion_info["p_mask"]
+
+    # Get diffusion config parameters
+    diffusion_config = getattr(config, "diffusion_config", {})
+    importance_weighting = diffusion_config.get("importance_weighting", True)
+
+    # Check if we have any masked tokens
+    if not masked_indices.any():
+        return torch.tensor(0.0, device=logits.device, requires_grad=True)
+
+    # Get predictions and targets for masked positions only
+    masked_logits = logits[masked_indices]
+    masked_targets = original_input_ids[masked_indices]  # Original unmasked tokens
+
+    # Compute cross-entropy loss without reduction
+    token_loss = F.cross_entropy(
+        masked_logits.float(), masked_targets, reduction="none"
+    )
+
+    if importance_weighting:
+        # Apply importance weighting: 1 / p_mask
+        masked_p_mask = p_mask.expand_as(masked_indices)[masked_indices]
+        weighted_loss = token_loss / masked_p_mask
+
+        if labels is not None:
+            # For SFT data: normalize by answer length per sample
+            answer_mask = labels != -100
+            answer_lengths = answer_mask.sum(dim=1).float()
+
+            # Group losses by batch sample
+            batch_indices = torch.arange(
+                original_input_ids.shape[0], device=original_input_ids.device
+            )
+            batch_indices = batch_indices.unsqueeze(1).expand_as(masked_indices)
+            masked_batch_indices = batch_indices[masked_indices]
+
+            # Sum losses per sample and normalize by answer length
+            loss_per_sample = torch.zeros(
+                original_input_ids.shape[0], device=original_input_ids.device
+            )
+            for i in range(original_input_ids.shape[0]):
+                sample_mask = masked_batch_indices == i
+                if sample_mask.any():
+                    sample_loss = weighted_loss[sample_mask].sum()
+                    loss_per_sample[i] = sample_loss / max(answer_lengths[i], 1)
+
+            loss = loss_per_sample.mean()
+        else:
+            # For completion data: simple average
+            loss = weighted_loss.mean()
+    else:
+        # No importance weighting
+        loss = token_loss.mean()
+
+    return loss
+
+
+def register_diffusion_loss():
+    """Register the diffusion loss function in transformers LOSS_MAPPING."""
+    try:
+        from transformers.loss.loss_utils import LOSS_MAPPING
+
+        LOSS_MAPPING["ForDiffusionLM"] = ForDiffusionLMLoss
+        return True
+    except ImportError:
+        # Fallback for older transformers versions
+        return False

src/axolotl/integrations/diffusion/model_patch.py

@@ -0,0 +1,149 @@
+"""Model patches for diffusion training."""
+
+import torch
+
+
+def patch_model_for_bidirectional_attention(model):
+    """
+    Patch model to handle diffusion training with forward process and bidirectional
+    attention.
+
+    This monkey-patches the model's forward method to:
+    - Apply forward diffusion process (masking) during training
+    - Use bidirectional attention masks
+    - Store info for loss computation
+    """
+    original_forward = model.forward
+
+    def diffusion_forward(
+        self,
+        input_ids: torch.Tensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        labels: torch.Tensor | None = None,
+        **kwargs,
+    ):
+        # Check if this is diffusion training
+        if (
+            hasattr(self.config, "loss_type")
+            and self.config.loss_type == "ForDiffusionLM"
+            and self.training
+        ):
+
+            # Store original input_ids for loss computation
+            original_input_ids = input_ids.clone()
+
+            # Apply forward diffusion process (masking)
+            diffusion_config = getattr(self.config, "diffusion_config", {})
+            noisy_input_ids, masked_indices, p_mask = _forward_process(
+                input_ids, attention_mask, labels, diffusion_config
+            )
+
+            # Use noisy input for model forward
+            input_ids = noisy_input_ids
+
+            # Convert attention mask to bidirectional
+            if attention_mask is not None:
+                attention_mask = _create_bidirectional_attention_mask(
+                    input_ids, attention_mask
+                )
+
+            # Store diffusion info in the model for loss computation
+            self._diffusion_info = {
+                "original_input_ids": original_input_ids,
+                "masked_indices": masked_indices,
+                "p_mask": p_mask,
+            }
+
+        return original_forward(
+            input_ids=input_ids, attention_mask=attention_mask, labels=labels, **kwargs
+        )
+
+    # Replace the forward method
+    model.forward = diffusion_forward.__get__(model, model.__class__)
+
+
+def _create_bidirectional_attention_mask(
+    input_ids: torch.Tensor, attention_mask: torch.Tensor
+) -> torch.Tensor:
+    """
+    Create bidirectional attention mask from 2D attention mask.
+
+    Args:
+        input_ids: Input token IDs [batch_size, seq_len]
+        attention_mask: 2D attention mask [batch_size, seq_len]
+
+    Returns:
+        bidirectional_mask: 4D attention mask [batch_size, 1, seq_len, seq_len]
+    """
+    batch_size, seq_len = input_ids.shape
+
+    # Simple bidirectional mask - all tokens can attend to all valid tokens
+    # Expand 2D mask to 4D: [batch_size, seq_len] -> [batch_size, 1, seq_len, seq_len]
+    bidirectional_mask = attention_mask.unsqueeze(1).unsqueeze(2)  # [B, 1, 1, S]
+    bidirectional_mask = bidirectional_mask.expand(batch_size, 1, seq_len, seq_len)
+
+    # Apply row-wise masking (padded tokens can't attend to anything)
+    row_mask = attention_mask.unsqueeze(1).unsqueeze(3)  # [B, 1, S, 1]
+    bidirectional_mask = bidirectional_mask & row_mask
+
+    return bidirectional_mask
+
+
+def _forward_process(
+    input_ids: torch.Tensor,
+    attention_mask: torch.Tensor | None = None,
+    labels: torch.Tensor | None = None,
+    diffusion_config: dict | None = None,
+) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """
+    Apply forward diffusion process (random masking).
+
+    Args:
+        input_ids: Input token IDs [batch_size, seq_len]
+        attention_mask: Attention mask [batch_size, seq_len]
+        labels: Labels for SFT training [batch_size, seq_len]
+        diffusion_config: Diffusion configuration dict
+
+    Returns:
+        noisy_input_ids: Input with masked tokens
+        masked_indices: Boolean mask of which tokens were masked
+        p_mask: Masking probabilities used
+    """
+    if diffusion_config is None:
+        diffusion_config = {}
+
+    batch_size, seq_len = input_ids.shape
+    device = input_ids.device
+
+    eps = diffusion_config.get("eps", 1e-3)
+    mask_token_id = diffusion_config.get("mask_token_id", 128002)
+
+    # Sample random timesteps for each sample
+    t = torch.rand(batch_size, device=device)
+
+    # Calculate masking probability with epsilon
+    p_mask = (1 - eps) * t + eps  # [batch_size]
+    p_mask = p_mask.unsqueeze(1).expand(-1, seq_len)  # [batch_size, seq_len]
+
+    # Don't mask padding tokens
+    if attention_mask is not None:
+        p_mask = p_mask * attention_mask.float()
+
+    # Create random mask based on p_mask
+    random_values = torch.rand_like(p_mask)
+    masked_indices = random_values < p_mask
+
+    # Apply attention mask constraints
+    if attention_mask is not None:
+        masked_indices = masked_indices & attention_mask.bool()
+
+    # For SFT data, only mask answer tokens (where labels != -100)
+    if labels is not None:
+        answer_mask = labels != -100
+        masked_indices = masked_indices & answer_mask
+
+    # Create noisy input by replacing masked tokens
+    noisy_input_ids = input_ids.clone()
+    noisy_input_ids[masked_indices] = mask_token_id
+
+    return noisy_input_ids, masked_indices, p_mask

src/axolotl/integrations/diffusion/plugin.py

@@ -0,0 +1,96 @@
+"""Diffusion LM training plugin for Axolotl."""
+
+from peft import PeftModel
+from transformers import PreTrainedModel
+
+from axolotl.integrations.base import BasePlugin
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.logging import get_logger
+
+from .args import DiffusionArgs
+from .callbacks import DiffusionGenerationCallback
+from .loss import register_diffusion_loss
+from .model_patch import patch_model_for_bidirectional_attention
+
+LOG = get_logger(__name__)
+
+
+class DiffusionPlugin(BasePlugin):
+    """
+    Plugin for diffusion language model training.
+
+    This plugin enables diffusion-based training using the LLaDA approach, which uses
+    random masking and bidirectional attention to train language models.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.cfg = None
+
+        if register_diffusion_loss():
+            LOG.info("Registered ForDiffusionLM loss function")
+        else:
+            LOG.warning(
+                "Failed to register diffusion loss - older transformers version"
+            )
+
+    def get_input_args(self) -> str:
+        """Returns the pydantic model for LLaDA plugin arguments."""
+        return "axolotl.integrations.diffusion.DiffusionArgs"
+
+    def post_model_load(self, cfg: DictDefault, model: PreTrainedModel | PeftModel):
+        """Configure model for diffusion training after loading."""
+        self.cfg = cfg
+
+        # Set loss type for diffusion training
+        if hasattr(model, "config"):
+            model.config.loss_type = "ForDiffusionLM"
+
+            # Store diffusion config in model config
+            model.config.diffusion_config = {
+                "eps": getattr(cfg, "eps", 1e-3),
+                "importance_weighting": getattr(cfg, "importance_weighting", True),
+                "mask_token_id": getattr(cfg, "mask_token_id", 128002),
+            }
+
+            LOG.info("Configured model for diffusion training with ForDiffusionLM loss")
+
+        # Patch model for bidirectional attention during training
+        patch_model_for_bidirectional_attention(model)
+        LOG.info("Applied bidirectional attention patch to model")
+
+        return model
+
+    def post_trainer_create(self, cfg: DictDefault, trainer):
+        """Configure trainer after creation."""
+        # Create diffusion config from cfg
+        diffusion_config = DiffusionArgs(
+            noise_schedule=getattr(cfg, "noise_schedule", "linear"),
+            min_mask_ratio=getattr(cfg, "min_mask_ratio", 0.1),
+            max_mask_ratio=getattr(cfg, "max_mask_ratio", 0.9),
+            num_diffusion_steps=getattr(cfg, "num_diffusion_steps", 128),
+            eps=getattr(cfg, "eps", 1e-3),
+            importance_weighting=getattr(cfg, "importance_weighting", True),
+            mask_token_id=getattr(cfg, "mask_token_id", 128002),
+            generate_samples=getattr(cfg, "generate_samples", True),
+            generation_interval=getattr(cfg, "generation_interval", 100),
+            num_generation_samples=getattr(cfg, "num_generation_samples", 3),
+            generation_steps=getattr(cfg, "generation_steps", 128),
+            generation_temperature=getattr(cfg, "generation_temperature", 0.0),
+            generation_max_length=getattr(cfg, "generation_max_length", 100),
+        )
+
+        # Store diffusion config on trainer for callbacks to access
+        trainer.diffusion_config = diffusion_config
+        LOG.info("Stored diffusion config on trainer")
+
+    def add_callbacks_post_trainer(self, cfg: DictDefault, trainer):
+        """Add diffusion generation callback if enabled."""
+        if (
+            hasattr(trainer, "diffusion_config")
+            and trainer.diffusion_config.generate_samples
+        ):
+            generation_callback = DiffusionGenerationCallback(trainer)
+            LOG.info("Added diffusion generation callback")
+            return [generation_callback]
+        return []

src/axolotl/loaders/model.py

@@ -681,6 +681,23 @@ class ModelLoader:
 
         return hf_ds_cfg
 
+    def _load_model_from_config(self) -> PreTrainedModel:
+        """Load model with random initialization using from_config."""
+        if self.auto_model_loader in [AutoModelForCausalLM, AutoModelForVision2Seq]:
+            return self.auto_model_loader.from_config(config=self.model_config)
+        return self.auto_model_loader(config=self.model_config)
+
+    def _load_model_from_pretrained(self, model_loader_class=None) -> PreTrainedModel:
+        """Load model from pretrained weights."""
+        loader = model_loader_class or self.auto_model_loader
+        kwargs = {
+            **self.model_kwargs,
+            "config": self.model_config,
+            "trust_remote_code": self.cfg.trust_remote_code or False,
+            **self.model_kwargs,
+        }
+        return loader.from_pretrained(self.base_model, **kwargs)
+
     def _build_model(self) -> bool:
         """Load model, with load strategy depending on config."""
         skip_move_to_device = False
@@ -695,7 +712,8 @@ class ModelLoader:
         if self.is_fsdp_enabled:
             if self.cfg.fsdp_config.cpu_ram_efficient_loading:
                 skip_move_to_device = True
-                # Don't delete device_map for QLoRA + FSDP - it was set correctly in _set_device_map
+                # Don't delete device_map for QLoRA + FSDP - it was set correctly in
+                # _set_device_map
                 if (
                     "device_map" in self.model_kwargs
                     and not self.is_qlora_and_fsdp_enabled
@@ -724,6 +742,11 @@ class ModelLoader:
                 or self.cfg.qlora_sharded_model_loading
             )
         ):
+            if self.cfg.reinit_weights:
+                LOG.warning(
+                    "reinit_weights is not supported with sharded quantized loading. "
+                    "Loading from pretrained weights instead."
+                )
             quant_storage = self.cfg.torch_dtype
             quantization_config = getattr(
                 self.model_config, "quantization_config", None
@@ -739,33 +762,12 @@ class ModelLoader:
                 quantization_config=quantization_config,
             )
             skip_move_to_device = True
-        elif (
-            self.model_config.model_type in ["llama", "llama4"]
-            and not self.cfg.trust_remote_code
-            and not self.cfg.gptq
-        ):
-            # Please don't remove underscore binding without reading the fn docstring.
-            _ = self._configure_zero3_memory_efficient_loading()
-
-            # Load model with random initialization if specified
-            if self.cfg.random_init_weights:
-                # AutoModel classes support the from_config method
-                if self.auto_model_loader in [
-                    AutoModelForCausalLM,
-                    AutoModelForVision2Seq,
-                ]:
-                    self.model = self.auto_model_loader.from_config(
-                        config=self.model_config,
-                    )
-                else:
-                    self.model = self.auto_model_loader(config=self.model_config)
-            else:
-                self.model = self.auto_model_loader.from_pretrained(
-                    self.base_model,
-                    config=self.model_config,
-                    **self.model_kwargs,
-                )
         elif self.model_type == "MambaLMHeadModel":
+            if self.cfg.reinit_weights:
+                LOG.warning(
+                    "reinit_weights is not supported with MambaLMHeadModel. "
+                    "Loading from pretrained weights instead."
+                )
             # FIXME this is janky at best and hacked together to make it work
             MambaLMHeadModel = fix_mamba_attn_for_loss()  # pylint: disable=invalid-name
 
@@ -778,41 +780,27 @@ class ModelLoader:
                 self.base_model,
                 **self.model_kwargs,
             )
-        elif (
-            self.model_type
-            and self.model_type != "AutoModelForCausalLM"
-            and not self.cfg.trust_remote_code
-        ):
-            if self.cfg.gptq:
-                self.model = self.auto_model_loader.from_pretrained(
-                    self.base_model,
-                    config=self.model_config,
-                    trust_remote_code=self.cfg.trust_remote_code or False,
-                    **self.model_kwargs,
-                )
-            else:
-                self.model = getattr(transformers, self.model_type).from_pretrained(
-                    self.base_model,
-                    config=self.model_config,
-                    trust_remote_code=self.cfg.trust_remote_code or False,
-                    **self.model_kwargs,
-                )
-        elif self.cfg.gptq:
-            self.model = self.auto_model_loader.from_pretrained(
-                self.base_model,
-                config=self.model_config,
-                trust_remote_code=self.cfg.trust_remote_code or False,
-                **self.model_kwargs,
-            )
         else:
-            # Please don't remove underscore binding without reading the fn docstring.
+            # Please don't remove underscore binding without reading the fn docstring
             _ = self._configure_zero3_memory_efficient_loading()
-            self.model = self.auto_model_loader.from_pretrained(
-                self.base_model,
-                config=self.model_config,
-                trust_remote_code=self.cfg.trust_remote_code or False,
-                **self.model_kwargs,
-            )
+
+            if (
+                self.model_type
+                and self.model_type != "AutoModelForCausalLM"
+                and not self.cfg.trust_remote_code
+                and not self.cfg.gptq
+            ):
+                # Use model type from transformers
+                model_loader_class = getattr(transformers, self.model_type)
+            else:
+                # Use auto model loader (handles gptq and default cases)
+                model_loader_class = self.auto_model_loader
+
+            if self.cfg.reinit_weights:
+                self.model = self._load_model_from_config()
+            else:
+                self.model = self._load_model_from_pretrained(model_loader_class)
+
         if is_deepspeed_zero3_enabled():
             skip_move_to_device = True
 

src/axolotl/prompt_tokenizers.py

@@ -75,7 +75,7 @@ class PromptTokenizingStrategy(abc.ABC):
     ) -> BatchEncoding:
         empty = BatchEncoding(data={"input_ids": [], "attention_mask": []})
         if not prompt:
-            LOG.warning("Empty text requested for tokenization.")
+            LOG.warning_once("Empty text requested for tokenization.")
             return empty
 
         result = self.tokenizer(

src/axolotl/utils/schemas/config.py

@@ -109,6 +109,12 @@ class AxolotlInputConfig(
             "description": "Don't upcast the embeddings to float32 when using PEFT. Useful for low-VRAM GPUs"
         },
     )
+    reinit_weights: bool | None = Field(
+        default=None,
+        json_schema_extra={
+            "description": "Reinitialize model weights randomly instead of loading pretrained weights"
+        },
+    )
 
     trainer_cls: str | None = Field(
         default=None,

tests/e2e/test_diffusion.py

@@ -0,0 +1,119 @@
+"""E2E smoke test for diffusion training plugin."""
+
+from axolotl.common.datasets import load_datasets
+from axolotl.train import train
+from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.dict import DictDefault
+
+from tests.e2e.utils import check_model_output_exists
+
+
+class TestDiffusion:
+    """Test case for diffusion training plugin."""
+
+    def test_diffusion_smoke_test(self, temp_dir):
+        """
+        Smoke test for diffusion training to ensure the plugin loads and trains without
+        error.
+        """
+        cfg = DictDefault(
+            {
+                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "tokenizer_type": "AutoTokenizer",
+                "trust_remote_code": True,
+                "sequence_len": 256,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "pad_token": "<|endoftext|>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 1,
+                "max_steps": 3,
+                "micro_batch_size": 1,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.0001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "bf16": True,
+                "save_safetensors": True,
+                "save_first_step": False,
+                "logging_steps": 1,
+                "eval_steps": 3,
+                # Diffusion-specific config
+                "plugins": ["axolotl.integrations.diffusion.DiffusionPlugin"],
+                "diffusion_mask_token_id": 16,
+                "diffusion_eps": 1e-3,
+                "diffusion_importance_weighting": False,
+            }
+        )
+
+        cfg = validate_config(cfg)
+        normalize_config(cfg)
+        dataset_meta = load_datasets(cfg=cfg)
+
+        train(cfg=cfg, dataset_meta=dataset_meta)
+        check_model_output_exists(temp_dir, cfg)
+
+    def test_diffusion_sft_labels(self, temp_dir):
+        """Test that diffusion training properly handles SFT data with labels."""
+        cfg = DictDefault(
+            {
+                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "tokenizer_type": "AutoTokenizer",
+                "trust_remote_code": True,
+                "sequence_len": 256,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "pad_token": "<|endoftext|>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 1,
+                "max_steps": 3,
+                "micro_batch_size": 1,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.0001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "bf16": True,
+                "save_safetensors": True,
+                "save_first_step": False,
+                "logging_steps": 1,
+                "eval_steps": 2,
+                # Diffusion-specific config
+                "plugins": ["axolotl.integrations.diffusion.DiffusionPlugin"],
+                "diffusion_mask_token_id": 16,
+                "diffusion_eps": 1e-3,
+                "diffusion_importance_weighting": True,
+                # Ensure we have proper SFT labels
+                "train_on_inputs": False,
+            }
+        )
+
+        cfg = validate_config(cfg)
+        normalize_config(cfg)
+        dataset_meta = load_datasets(cfg=cfg)
+
+        # Verify that the dataset has labels
+        sample = dataset_meta.train_dataset[0]
+        assert "labels" in sample, "SFT dataset should have labels"
+
+        # Check that some labels are -100 (prompt tokens)
+        labels = sample["labels"]
+        if hasattr(labels, "tolist"):
+            labels = labels.tolist()
+        assert -100 in labels, "SFT dataset should have -100 labels for prompt tokens"
+
+        train(cfg=cfg, dataset_meta=dataset_meta)
+        check_model_output_exists(temp_dir, cfg)

tests/integrations/test_diffusion.py

@@ -0,0 +1,335 @@
+"""Tests for diffusion trainer integration."""
+
+# pylint: disable=redefined-outer-name,protected-access
+
+from unittest.mock import Mock, patch
+
+import pytest
+import torch
+
+from axolotl.integrations.diffusion.args import DiffusionArgs
+from axolotl.integrations.diffusion.loss import (
+    ForDiffusionLMLoss,
+    register_diffusion_loss,
+)
+from axolotl.integrations.diffusion.model_patch import (
+    _create_bidirectional_attention_mask,
+    _forward_process,
+    patch_model_for_bidirectional_attention,
+)
+from axolotl.integrations.diffusion.plugin import DiffusionPlugin
+
+
+@pytest.fixture
+def diffusion_config():
+    """Create a diffusion config."""
+    return DiffusionArgs(
+        eps=1e-3,
+        importance_weighting=False,
+        mask_token_id=32000,
+        generate_samples=False,
+    )
+
+
+@pytest.fixture
+def mock_model():
+    """Create a mock model."""
+    model = Mock()
+    model.config = Mock()
+    model.config.loss_type = "ForDiffusionLM"
+    model.config.diffusion_config = {
+        "eps": 1e-3,
+        "importance_weighting": False,
+        "mask_token_id": 32000,
+    }
+    model.training = True
+    return model
+
+
+class TestDiffusionLoss:
+    """Test the ForDiffusionLMLoss function."""
+
+    def test_loss_with_diffusion_info(self, mock_model):
+        """Test loss computation with stored diffusion info."""
+        batch_size, seq_len, vocab_size = 1, 5, 1000
+
+        # Mock stored diffusion info
+        original_input_ids = torch.tensor([[1, 10, 20, 30, 2]], dtype=torch.long)
+        masked_indices = torch.tensor(
+            [[False, True, True, False, False]], dtype=torch.bool
+        )
+        p_mask = torch.tensor([[0.5, 0.5, 0.5, 0.5, 0.5]], dtype=torch.float)
+
+        mock_model._diffusion_info = {
+            "original_input_ids": original_input_ids,
+            "masked_indices": masked_indices,
+            "p_mask": p_mask,
+        }
+
+        # Mock logits
+        logits = torch.randn(batch_size, seq_len, vocab_size, requires_grad=True)
+        labels = torch.tensor([[-100, -100, 20, 30, 2]], dtype=torch.long)
+
+        loss = ForDiffusionLMLoss(
+            logits=logits,
+            labels=labels,
+            vocab_size=vocab_size,
+            config=mock_model.config,
+            model=mock_model,
+        )
+
+        assert isinstance(loss, torch.Tensor)
+        assert loss.requires_grad
+        assert loss.item() >= 0
+
+    def test_loss_fallback_without_diffusion_info(self, mock_model):
+        """Test fallback to causal LM loss when no diffusion info."""
+        batch_size, seq_len, vocab_size = 1, 5, 1000
+
+        # Remove diffusion info to trigger fallback
+        if hasattr(mock_model, "_diffusion_info"):
+            delattr(mock_model, "_diffusion_info")
+
+        logits = torch.randn(batch_size, seq_len, vocab_size, requires_grad=True)
+        labels = torch.tensor([[1, 10, 20, 30, 2]], dtype=torch.long)
+
+        loss = ForDiffusionLMLoss(
+            logits=logits,
+            labels=labels,
+            vocab_size=vocab_size,
+            config=mock_model.config,
+            model=mock_model,
+        )
+
+        assert isinstance(loss, torch.Tensor)
+        assert loss.requires_grad
+
+    def test_loss_no_masked_tokens(self, mock_model):
+        """Test loss when no tokens are masked."""
+        batch_size, seq_len, vocab_size = 1, 3, 1000
+
+        # No masked tokens
+        original_input_ids = torch.tensor([[1, 10, 2]], dtype=torch.long)
+        masked_indices = torch.tensor([[False, False, False]], dtype=torch.bool)
+        p_mask = torch.tensor([[0.1, 0.1, 0.1]], dtype=torch.float)
+
+        mock_model._diffusion_info = {
+            "original_input_ids": original_input_ids,
+            "masked_indices": masked_indices,
+            "p_mask": p_mask,
+        }
+
+        logits = torch.randn(batch_size, seq_len, vocab_size)
+        labels = torch.tensor([[1, 10, 2]], dtype=torch.long)
+
+        loss = ForDiffusionLMLoss(
+            logits=logits,
+            labels=labels,
+            vocab_size=vocab_size,
+            config=mock_model.config,
+            model=mock_model,
+        )
+
+        assert loss.item() == 0.0
+
+
+class TestModelPatch:
+    """Test the model patching functionality."""
+
+    def test_forward_process_basic(self):
+        """Test basic forward process."""
+        input_ids = torch.tensor([[1, 10, 20, 30, 2]], dtype=torch.long)
+        diffusion_config = {"eps": 0.1, "mask_token_id": 32000}
+
+        noisy_input_ids, masked_indices, p_mask = _forward_process(
+            input_ids, diffusion_config=diffusion_config
+        )
+
+        # Check shapes
+        assert noisy_input_ids.shape == input_ids.shape
+        assert masked_indices.shape == input_ids.shape
+        assert p_mask.shape == input_ids.shape
+
+        # Check that mask token is applied where masked
+        if masked_indices.any():
+            assert (noisy_input_ids[masked_indices] == 32000).all()
+
+    def test_forward_process_with_labels(self):
+        """Test forward process with SFT labels."""
+        input_ids = torch.tensor([[1, 10, 20, 30, 2]], dtype=torch.long)
+        labels = torch.tensor([[-100, -100, 20, 30, 2]], dtype=torch.long)
+        diffusion_config = {"eps": 0.1, "mask_token_id": 32000}
+
+        _, masked_indices, _ = _forward_process(
+            input_ids, labels=labels, diffusion_config=diffusion_config
+        )
+
+        # Check that only answer tokens can be masked (where labels != -100)
+        non_answer_mask = labels == -100
+        assert not masked_indices[non_answer_mask].any()
+
+    def test_forward_process_with_attention_mask(self):
+        """Test forward process with attention mask."""
+        input_ids = torch.tensor([[1, 10, 20, 0]], dtype=torch.long)
+        attention_mask = torch.tensor([[1, 1, 1, 0]], dtype=torch.long)
+        diffusion_config = {"eps": 0.1, "mask_token_id": 32000}
+
+        _, masked_indices, p_mask = _forward_process(
+            input_ids, attention_mask=attention_mask, diffusion_config=diffusion_config
+        )
+
+        # Check that padding tokens are not masked
+        padding_positions = attention_mask == 0
+        assert not masked_indices[padding_positions].any()
+        assert (p_mask[padding_positions] == 0).all()
+
+    def test_bidirectional_attention_mask(self):
+        """Test bidirectional attention mask creation."""
+        input_ids = torch.tensor([[1, 10, 20, 2]], dtype=torch.long)
+        attention_mask = torch.tensor([[1, 1, 1, 1]], dtype=torch.long)
+
+        mask = _create_bidirectional_attention_mask(input_ids, attention_mask)
+
+        # Should be all-to-all attention
+        expected_shape = (1, 1, 4, 4)
+        assert mask.shape == expected_shape
+        assert mask.all()
+
+    def test_bidirectional_attention_mask_with_padding(self):
+        """Test bidirectional attention mask with padding."""
+        input_ids = torch.tensor([[1, 10, 20, 0]], dtype=torch.long)
+        attention_mask = torch.tensor([[1, 1, 1, 0]], dtype=torch.long)
+
+        mask = _create_bidirectional_attention_mask(input_ids, attention_mask)
+
+        # Padding positions should not attend or be attended to
+        assert not mask[0, 0, 3, :].any()  # Padding can't attend to anything
+        assert not mask[0, 0, :, 3].any()  # Nothing can attend to padding
+
+    def test_patch_model_for_bidirectional_attention(self):
+        """Test that model patching works."""
+        mock_model = Mock()
+        mock_model.config = Mock()
+        mock_model.config.loss_type = "ForDiffusionLM"
+        mock_model.config.diffusion_config = {"eps": 1e-3, "mask_token_id": 32000}
+        mock_model.training = True
+
+        original_forward = Mock()
+        mock_model.forward = original_forward
+
+        # Patch the model
+        patch_model_for_bidirectional_attention(mock_model)
+
+        # Check that forward method was replaced
+        assert mock_model.forward != original_forward
+
+
+class TestDiffusionPlugin:
+    """Test the DiffusionPlugin."""
+
+    def test_plugin_registers_loss_function(self):
+        """Test that plugin registers diffusion loss function."""
+        with patch(
+            "axolotl.integrations.diffusion.plugin.register_diffusion_loss",
+            return_value=True,
+        ) as mock_register:
+            plugin = DiffusionPlugin()
+            mock_register.assert_called_once()
+
+    def test_post_model_load_configuration(self):
+        """Test that post_model_load configures model correctly."""
+        plugin = DiffusionPlugin()
+
+        # Mock model and config
+        mock_model = Mock()
+        mock_model.config = Mock()
+        mock_cfg = Mock()
+        mock_cfg.eps = 1e-3
+        mock_cfg.importance_weighting = True
+        mock_cfg.mask_token_id = 32000
+
+        with patch(
+            "axolotl.integrations.diffusion.plugin.patch_model_for_bidirectional_attention"
+        ) as mock_patch:
+            result = plugin.post_model_load(mock_cfg, mock_model)
+
+            # Check model configuration
+            assert mock_model.config.loss_type == "ForDiffusionLM"
+            assert mock_model.config.diffusion_config is not None
+            assert mock_model.config.diffusion_config["eps"] == 1e-3
+
+            # Check model was patched
+            mock_patch.assert_called_once_with(mock_model)
+
+            # Should return the model
+            assert result == mock_model
+
+    def test_post_trainer_create_stores_config(self, diffusion_config):
+        """Test that post_trainer_create stores config on trainer."""
+        plugin = DiffusionPlugin()
+        mock_trainer = Mock()
+        mock_cfg = Mock()
+
+        # Set config attributes
+        for attr, value in diffusion_config.model_dump().items():
+            setattr(mock_cfg, attr, value)
+
+        plugin.post_trainer_create(mock_cfg, mock_trainer)
+
+        # Check that diffusion config was stored on trainer
+        assert hasattr(mock_trainer, "diffusion_config")
+        assert mock_trainer.diffusion_config.eps == diffusion_config.eps
+
+    def test_add_callbacks_post_trainer_with_generation_enabled(self):
+        """Test callback addition when generation is enabled."""
+        plugin = DiffusionPlugin()
+        mock_trainer = Mock()
+        mock_cfg = Mock()
+
+        # Mock trainer with diffusion config that has generation enabled
+        mock_trainer.diffusion_config = DiffusionArgs(generate_samples=True)
+
+        with patch(
+            "axolotl.integrations.diffusion.plugin.DiffusionGenerationCallback"
+        ) as mock_callback_class:
+            callbacks = plugin.add_callbacks_post_trainer(mock_cfg, mock_trainer)
+
+            # Should return one callback
+            assert len(callbacks) == 1
+            mock_callback_class.assert_called_once_with(mock_trainer)
+
+    def test_add_callbacks_post_trainer_with_generation_disabled(self):
+        """Test callback addition when generation is disabled."""
+        plugin = DiffusionPlugin()
+        mock_trainer = Mock()
+        mock_cfg = Mock()
+
+        # Mock trainer with diffusion config that has generation disabled
+        mock_trainer.diffusion_config = DiffusionArgs(generate_samples=False)
+
+        callbacks = plugin.add_callbacks_post_trainer(mock_cfg, mock_trainer)
+
+        # Should return no callbacks
+        assert len(callbacks) == 0
+
+
+class TestLossRegistration:
+    """Test loss function registration."""
+
+    def test_register_diffusion_loss(self):
+        """Test that loss function can be registered."""
+        with patch("transformers.loss.loss_utils.LOSS_MAPPING", {}) as mock_mapping:
+            result = register_diffusion_loss()
+            assert result is True
+            assert "ForDiffusionLM" in mock_mapping
+            assert mock_mapping["ForDiffusionLM"] == ForDiffusionLMLoss
+
+    def test_register_diffusion_loss_import_error(self):
+        """Test fallback when LOSS_MAPPING import fails."""
+        # Patch the import to raise ImportError
+        with patch(
+            "builtins.__import__",
+            side_effect=ImportError("transformers.loss.loss_utils not found"),
+        ):
+            result = register_diffusion_loss()
+            assert result is False